AB-100 Agentic AI Business Solutions Architect

https://www.marks4sure.com/ab-100-agentic-ai-business-solutions-architect-beta-questions.html

1 Plan AI-powered business solutions

Case study

Overview

Contoso, Ltd. is a high-tech manufacturing company that uses Microsoft Dynamics 365 Finance. Dynamics 365 Supply Chain Management, and Dynamics 365 Commerce for its North American operations. The company designs and develops innovative products that have many patents and proprietary technologies. The patents and engineering designs are closely guarded secrets.

Contoso executives want to integrate and adopt AI solutions to help scale the company in preparation for an anticipated period of rapid growth.

The company has multiple legal entities and Azure subscriptions that will be used in the adopted AI solutions.

Requirements

AI Adoption

The following executives will have specific responsibilities in the overall AI adoption:

Chief Technology Officer (CTO): Select one Dynamics 365 Finance, Dynamics 365 Supply Chain Management or Dynamics 365 Commerce prebuilt AI agent and one custom Microsoft Copilot Studio AI agent to prioritize and deploy during the initial AI adoption phase.
Chief Information Officer (CIO): Ensure that appropriate security labels are assigned to the data used by the AI agents.
Chief Financial Officer (CFO): Analyze the return on investment (ROI) for the AI agents being deployed. Chief Information Security Officer (CISO): Discover and inventory AI resources for auditing.
Chief Executive Officer (CEO): Ensure that all solutions adhere to industry-standard responsible AI practices.

All AI initiatives and agents will have a detailed business use case, a defined audience profile, and an estimated ROI that will compare the cost savings of the current process against the estimated costs of using the new AI solutions.

The company's research and development (R&D) department already has a custom Model Context Protocol (MCP) server that contains comprehensive product specifications and compliance data.

Prebuilt AI Agent

The CTO has NOT yet selected which prebuilt AI agent to use in Dynamics 365 Supply Chain Management. The CTO wants to view available agent templates to identify which agent will add the most business value.

Depending on which high-priority AI agents are identified, its agent capabilities must be previewed in a discovery meeting with the relevant business operation stakeholders.

Custom AI Agent

Contoso has identified the following custom AI agent requirements:

The custom AI agent will use data from Dynamics 365 Supply Chain Management to answer questions for the manufacturing team as a low-code solution.
The custom AI agent will be accessible from within Microsoft Teams.
The custom AI agent must be designed to eventually connect to other agents that can be selected based on their description.
The topics used in the custom AI agent will be selected based NOT on a trigger phrase, but on a description of the purpose of the query, to make the interactions more conversational.
The custom AI agent must be able to answer questions about product specifications by using existing technologies. The product specifications are maintained by the R&D department.
The custom AI agent must be integrated with and accessible from Dynamics 365 Supply Chain Management.
The custom AI agent must be able to use Dynamics 365 Supply Chain Management business logic that is stored outside of the application.

Analysis, Reporting, and Troubleshooting

Contoso has identified the following analysis, reporting, and troubleshooting requirements:

The CISO will audit all the AI solutions monthly for compliance and security.

The CFO will analyze all the AI solutions quarterly to compare the estimated ROI against actual measured efficiencies and adoption. The CFO will use the Copilot Studio agent usage estimator to perform this analysis.

The CISO wants to identify how much sensitive data was accessed for a given AI agent run and who accessed the data. Too much sensitive data accessed by a single user might indicate a high security risk.

The CTO wants to track user feedback on the quality of the AI agent responses during user interactions with the agents. Consistently poor feedback will trigger an escalated reengineering discussion.

The CEO wants a quarterly assessment of all the required metrics for their specific responsibilities. The tools used for the assessments must be Microsoft-recommended and must verify reliability, interpretability, fairness, and compliance.

The CFO wants to identify how many interactions with the AI agents are abandoned on a given day as compared to resolved conversations. Too many abandoned sessions might indicate that Copilot Studio credits are being used inefficiently by end users.

QUESTION 1 💩

Which two components in the custom AI agent design should the CFO evaluate in the quarterly agent analysis? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. the GPT models used for the agent

B. the average characters in a chat message

C. the agent orchestration method

D. the average session time per agent

你这个质疑是对的。如果这是当前版本 Copilot Studio Agent Usage Estimator 的考试题，我会更倾向于 C + D，而不是之前给你的 A + C。

正确答案：✅ C、D

选项	判断	原因
A. GPT models used	❌	属于底层模型实现细节，不是 CFO 进行季度 usage/ROI 分析的主要指标
B. Average characters in a chat message	❌	不是主要的 Agent usage 分析指标
C. Agent orchestration method	✅	Classic vs. Generative orchestration 会影响 Agent 如何进行推理、调用工具和知识源，从而影响使用量和成本
D. Average session time per agent	✅	反映用户使用情况和会话效率，可用于分析 adoption、session usage 以及成本效率

为什么这题容易误导？

题目不是问：

“哪些因素会影响模型 token 消耗？”

如果这样问，GPT model 很容易成为答案。

它实际问的是：

CFO should evaluate in the quarterly agent analysis

而且明确给了：

compare estimated ROI against actual measured efficiencies and adoption

所以更偏向 Agent usage / adoption / efficiency analytics。

因此：

C = orchestration / consumption behavior

D = session usage / adoption efficiency

📌 考试记忆

看到：

CFO + quarterly + ROI + usage estimator + adoption

优先想到：

Orchestration + Session usage

所以这道题我建议你记：

✅ C + D

Correct Answer: CD

Explanation/Reference:

Explanation:

Scenario:

The CFO will analyze all the AI solutions quarterly to compare the estimated ROI against actual measured efficiencies and adoption. The CFO will use the Copilot Studio agent usage estimator to perform this analysis.

-Quarterly Estimated ROI (Forecasting) Use the Microsoft Agent Usage Estimator to model quarterly expectations before each period.

Orchestration Method Input: Select between Classic (logic-driven) or Generative (AI-driven) orchestration. Generative orchestration typically consumes more credits but reduces manual development time.

Session Time Variables: Model the average session time per agent to estimate total message volume. The estimator uses this to project credit consumption based on interaction depth.

Target ROI Formula: Define the benchmark as:

Estimated Savings = (Projected Deflection × Human Agent Cost) - Estimated Credit Cost.

QUESTION 2 💩💩💩💩💩

What should you configure for the custom AI agent?

A. AI-assisted evaluators

B. classic orchestration

C. generative orchestration

D. Azure OpenAI reasoning models

正确答案： C. generative orchestration（生成式编排）

解析与考点分析

为什么选 C？

直接看案例原文中关于 Custom AI Agent（自定义 AI Agent） 的明确要求：

"The topics used in the custom AI agent will be selected based NOT on a trigger phrase, but on a description of the purpose of the query, to make the interactions more conversational."

（自定义 AI Agent 中使用的主题将基于对查询目的的描述来选择，而不是基于触发词，以使交互更具对话性。）

在 Microsoft Copilot Studio 中：

Classic Orchestration（传统编排，选项 B）： 依赖用户输入的触发词（Trigger Phrases）来精准匹配并触发特定的 Topic（主题）。
Generative Orchestration（生成式编排，选项 C）： 允许 Copilot Studio 使用大语言模型（LLM）理解用户的意图，并根据每个 Topic、Plugin（插件）或 Agent 的描述（Description）动态选择并调用最合适的内容，完全不需要预定义触发词。

另外，案例中还提到：

"The custom AI agent must be designed to eventually connect to other agents that can be selected based on their description."

（自定义 AI Agent 必须设计为最终能够连接到其他可以根据其描述进行选择的 Agent。）

这同样是 Copilot Studio 中 Generative Orchestration（生成式编排）/ Dynamic Chaining（动态链接） 的标准功能（即 Agent 间通过 Description 互相路由与协作）。

为什么不选其他选项？

A. AI-assisted evaluators（AI 辅助评估器）： 这是 Azure AI Foundry / Responsible AI 中用于测试和评估模型输出质量（如可靠性、公平性）的工具，用于 CEO/CTO 的季度质量评估，而不是 Agent 运行时的话题匹配机制。
B. classic orchestration（传统编排）： 与案例要求的“不使用触发词（NOT on a trigger phrase）”直接冲突。
D. Azure OpenAI reasoning models（Azure OpenAI 推理模型）： 指的是 o1/o3 等具备高级推理能力的模型，而题目问的是针对 Agent 话题路由与调用机制的配置（Orchestration）。

Correct Answer: C

Explanation/Reference:

Explanation:

Generative orchestration is the most appropriate choice for this Microsoft Dynamics 365 AI agent solution. This selection directly addresses your requirements for a low-code, conversational, and interconnected agent ecosystem within the Microsoft Power Platform and Dynamics 365 environment.

Why Generative Orchestration?

Generative orchestration (available in Microsoft Copilot Studio) is specifically designed to move away from rigid, trigger-phrase-based logic toward a flexible, intent-based model.

Mapping to Your Requirements

Intent-Based Selection: Unlike "Classic" orchestration which relies on exact trigger phrases, generative orchestration uses Natural Language Understanding (NLU). It selects the correct topic or "sub-agent" based on a description of the purpose, allowing for the conversational flow you requested.

Low-Code Integration: Copilot Studio is the primary low-code tool for Dynamics 365. It provides native connectors to Supply Chain Management (SCM) data and can be embedded directly into the SCM interface or deployed to Microsoft Teams.

External Business Logic: It can trigger Power Automate flows or API calls to execute business logic stored in external databases or legacy systems, bringing that data back into the conversation.

Product Specifications: By using Generative Answers, the agent can crawl "existing technologies" like SharePoint libraries, internal wikis, or SCM data tables to answer complex spec questions without manual topic authoring.

Incorrect:

[Not A]

AI-Assisted Evaluators are testing and diagnostic tools, not runtime execution engines.

You would use these to measure how well your agent is performing, but they cannot be the agent or manage the logic flow.

[Not B]

Classic Orchestration is entirely dependent on trigger phrases.

It creates a "command-and-control" feel rather than the fluid, conversational interaction you are looking for.

It also scales poorly when trying to connect multiple agents.

[Not D]

Azure OpenAI Models (Reasoning Models) while powerful, this is a pro-code path (API-heavy).

Using raw Azure OpenAI models would require significant custom development, missing the "low-code" requirement. While Generative Orchestration uses these models under the hood, the orchestration layer itself is what manages the "which agent to call" logic.

Scenario: Custom AI Agent

Contoso has identified the following custom AI agent requirements:

*-> The custom AI agent will use data from Dynamics 365 Supply Chain Management to answer questions for the manufacturing team as a low-code solution.

The custom AI agent will be accessible from within Microsoft Teams.

The custom AI agent must be designed to eventually connect to other agents that can be selected based on their description

*-> The topics used in the custom AI agent will be selected based NOT on a trigger phrase, but on a description of the purpose of the query, to make the interactions more conversational.

The custom AI agent must be able to answer questions about product specifications by using existing technologies. The product specifications are maintained by the R&D department.

*-> The custom AI agent must be integrated with and accessible from Dynamics 365 Supply Chain Management.

*-> The custom AI agent must be able to use Dynamics 365 Supply Chain Management business logic that is stored outside of the application.

2 Plan AI-powered business solutions

Background

Fabrikam, Inc., is a global consumer goods company that is undergoing a digital transformation initiative to migrate its entire infrastructure to the Microsoft cloud. As a key element of this cloud migration, the company will implement Microsoft Dynamics 365 Sales, moving away from the current on-premises proprietary technologies used by its business-to-business (B2B) sales team.

As part of the cloud migration, Fabrikam will adopt an AI-first approach to its business solutions and implement AI solutions, wherever possible, to streamline operations.

Problem Statements

Fabrikam's infrastructure currently relies on various on-premises systems that require sales executives to use corporate computers with physical keyboards to access business information during customer interactions. Mobile phones cannot be used for these purposes, as the systems depend on keyboard input. As a result, the sales executives spend a lot of time using keyboards to search for data on several disparate systems and file servers, rather than focusing on the customers. This affects the customer experience.

Fabrikam stakeholders are concerned that users will be hesitant to adopt AI. If the AI initiatives are NOT adopted, cost savings will never be realized. Additionally, funding for future AI initiatives will depend on demonstrating an increase in AI adoption month over month. As the AI agent initiative for the sales team will be the first for Fabrikam, the rapid adoption of the agent is a high priority

Planned Initiatives

General

Fabrikam management has prioritized AI-driven projects to improve efficiency, customer engagement, and responsible AI adoption. The current application infrastructure is on-premises and must be migrated to the cloud to support the adoption of these technologies.

Infrastructure Migration

Fabrikam plans to migrate from its current on-premises infrastructure to a completely cloud-based topology; this will include user authentication, the security framework, and, primarily, the adoption of the services by end users.

All the data from the different systems will be consolidated into a single data source - a common data model that will use a Microsoft Dataverse environment as a single source of truth (SSOT) for the sales team.

Sales Cycle Enablement

To achieve the company's objectives, Fabrikam intends to implement the following strategies to enhance the sales cycle:

Use low-code development to create a single AI agent that has Dataverse as its core component. Ensure that sales managers can access unanswered correspondence from prospects and intervene as appropriate.

Replace the previous proprietary software with Dynamics 365 Sales to track sales cycles and customer interactions.

Have the sales executives use Dynamics 365 Sales to track interactions for open opportunities and send follow-up communications to prospects.

Have the sales executives use handsfree headsets to interact with an AI agent when they have questions about internal policies or customer data.

Requirements

Infrastructure Migration

Fabrikam has identified the following infrastructure migration requirements:

Azure must be used for all future infrastructure workloads.

The company must follow Microsoft-recommended methodologies for infrastructure migration to the cloud.

Any created AI agents must have their return on investment (ROI) calculated to ensure that the solution will save the company money.

Sales Cycle Enablement

Fabrikam has identified the following requirements for sales cycle enablement:

The final AI agent must follow Microsoft recommendations for a conversational user experience.

A designated checklist must be reviewed to ensure that the AI agent follows Microsoft deployment recommendations for a compliant solution.

Detailed telemetry must be logged for the first created AI agent to help troubleshoot and optimize the agent during the initial AI agent adoption process.

Unexpected AI agent actions must end in an escalation to a live representative. For example, a sales executive must be rerouted to a representative if the agent cannot answer a question after two failed attempts.

The return on investment (ROI) of switching from the current process to the future process is required for stakeholder sign off.

The sales team must use Dynamics 365 Sales to correspond with prospects more quickly and efficiently than currently.

Sales managers must report on the adoption of the AI agent to key Fabrikam stakeholders on a monthly basis.

Any sensitive information, such as user IDs and names, shared via the AI agent must be tracked for future auditing.

QUESTION 1 💩💩💩

Which framework should you use to meet the AI agent requirements for the sales cycle enablement? To answer, select the appropriate options in the answer area.

Hot Area:

Answer Area

For Microsoft Copilot Studio best practices:

the ALM Accelerator for Microsoft Power Platform
Microsoft Cloud Adoption Framework for Azure
Microsoft Power Platform Well-Architected framework
Success by Design

For conversational user experiences:

the ALM Accelerator for Microsoft Power Platform
Microsoft Cloud Adoption Framework for Azure
Microsoft Power Platform Well-Architected framework
Success by Design

For Microsoft Copilot Studio best practices: → Microsoft Power Platform Well-Architected framework
For conversational user experiences: → Success by Design

为什么？

1. Copilot Studio best practices → Power Platform Well-Architected

这个框架关注 Power Platform 解决方案的：

Security
Reliability
Operational excellence
Performance
Cost optimization

因此更适合题目中 AI Agent 的部署、治理、Telemetry、合规和成本/ROI等要求。

2. Conversational user experiences → Success by Design

Success by Design 更强调 Dynamics 365 / Power Platform 项目的实施方法和用户体验设计，包括 conversational experience、设计检查、实施风险和 adoption 等。

题目明确说：

The final AI agent must follow Microsoft recommendations for a conversational user experience.

因此这里选择 Success by Design 更符合题意。

🧠 考试记忆

Copilot Studio / Power Platform architecture & best practices → Well-Architected、

Dynamics 365 implementation + conversational experience → Success by Design

QUESTION 2 💩💩💩

Which framework should you use for the infrastructure migration?

A. Microsoft Cloud Adoption Framework for Azure (CAF)

B. Success by Design

C. Microsoft Power Platform Center of Excellence (CoE)

D. Microsoft Power Platform Project Setup Wizard

正确答案：✅ A. Microsoft Cloud Adoption Framework for Azure

原因

题目问的是：

Which framework should you use for the infrastructure migration?

而案例明确要求：

follow Microsoft-recommended methodologies for infrastructure migration to the cloud

Microsoft Cloud Adoption Framework for Azure (CAF) 就是 Microsoft 用于规划和执行 Azure Cloud Adoption / Migration 的推荐框架。

其他选项

❌ B. Success by Design → 主要针对 Dynamics 365 项目的实施方法论。
❌ C. Power Platform CoE → 用于治理和管理 Power Platform adoption。
❌ D. Power Platform Project Setup Wizard → 用于 Power Platform 项目设置，不是 Azure infrastructure migration framework。

🎯 最终答案

A. Microsoft Cloud Adoption Framework for Azure

记忆：

Cloud migration → CAF

Dynamics 365 implementation → Success by Design

Power Platform governance → CoE

Correct Answer: A

Explanation/Reference:

For migrating a legacy on-premises infrastructure to Microsoft Dynamics 365 Sales with Dataverse as the Single Source of Truth (SSOT), the recommended framework is the Microsoft Cloud Adoption Framework for Azure (CAF), specifically utilized in conjunction with the Data Management Framework (DMF) for Dynamics 365.

This combined approach ensures a structured transition by focusing on both the strategic adoption of cloud technology and the technical, granular migration of data.

Recommended Framework: Microsoft Cloud Adoption Framework (CAF) The CAF provides a holistic structure to ensure the migration is secure, compliant, and aligned with business goals.

Plan: Assess legacy data, prioritize workloads, and define the SSOT requirements.

Ready: Set up the Dataverse environment (landing zone) and configure security (Azure Active Directory/ Microsoft Entra ID).

Adopt (Migrate): Perform the technical migration of data using ETL (Extract, Transform, Load) processes.

3 Scenario: Infrastructure Migration

Fabrikam plans to migrate from its current on-premises infrastructure to a completely cloud-based topology; this will include user authentication, the security framework, and, primarily, the adoption of the services by end users.

All the data from the different systems will be consolidated into a single data source - a common data model that will use a Microsoft Dataverse environment as a single source of truth (SSOT) for the sales team.

Background

Fabrikam, Inc., is a global consumer goods company that is undergoing a digital transformation initiative to migrate its entire infrastructure to the Microsoft cloud. As a key element of this cloud migration, the company will implement Microsoft Dynamics 365 Sales, moving away from the current on-premises proprietary technologies used by its business-to-business (B2B) sales team.

Reference:

Plan AI-powered business solutions

QUESTION 1. 💩💩💩

A company uses Microsoft Dynamics 365 Sales to manage leads that are stored in a Microsoft Dataverse table named Lead and use non-standard terminology and custom columns.

You need to configure business terms in the Lead table so that Microsoft Copilot controls can summarize the leads efficiently. The solution must minimize administrative effort.

How should you configure the business terms?

A. Combine all the fields into one custom field.

B. Map the field display names as business terms.

C. Add the schema names as business terms.

D. Create new business terms for each field.

正确答案：✅ B. Map the field display names as business terms.

原因

题目关键：

Lead 使用 non-standard terminology
有 custom columns
需要让 Microsoft Copilot 正确理解字段
minimize administrative effort

Business Terms 的目的就是把 Copilot 使用的业务术语映射到 Dataverse 字段。

因此直接：

Field display name → Business term

不需要为每个字段重新创建一套独立的业务术语。

其他选项

❌ A：把字段合并会破坏数据结构。
❌ C：Schema name 是技术名称，不是用户/业务使用的术语。
❌ D：逐个创建新的 Business Terms，管理成本更高。

最终答案：B. Map the field display names as business terms.

Correct Answer: B

Explanation/Reference:

To configure Microsoft Copilot to efficiently summarize leads with non-standard terminology and custom columns in Microsoft Dynamics 365 Sales, you must map these unique fields to business terms within the Sales AI Glossary in Microsoft Copilot Studio.

Note:

To map your field display names as business terms:

Access Copilot Studio: Open Microsoft Copilot Studio and select the environment containing your Dynamics 365 Sales instance.
Select the Sales Agent: Navigate to Agents and select the agent named Copilot in Dynamics 365 Sales (formerly Sales Copilot Power Virtual Agents Bot).
Navigate to Knowledge: Under the Knowledge section, select the SalesSpecificQnA knowledge source.
Add Glossary Entries:

Go to the Glossary tab.

Term: Enter the non-standard or custom field display name (e.g., your custom business term).

Description: Define how this term relates to the Dataverse schema. This helps Copilot understand the logic behind the custom column.

Configure Synonyms: In the Synonyms section, map your custom field to alternative names that sellers might use in natural language queries (e.g., mapping "Custom Revenue" to "Opportunity Revenue").
Publish Changes: Select Publish to apply these mappings, allowing Copilot to use the newly defined terms when generating lead summaries.

QUESTION 2 💩💩

You are designing two Microsoft Copilot Studio agents named Agent1 and Agent2. Each agent must meet the following requirements:

Each agent must use a standard model.
Each agent must NOT use generative orchestration.
Agent1 must support simple and short phrases for a given topic.
Agent2 must integrate with Microsoft Dynamics 365 Contact Center voice channel.

You need to recommend language models for the agents.

What should you recommend for each agent? To answer, drag the appropriate language models to the correct agents. Each language model may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

Select and Place:

Language models

Azure Language in Foundry Tools
Azure OpenAI
Conversational language understanding (CLU)
Natural language understanding (NLU)
Natural language understanding + (NLU+)

Answer Area

Agent1:
Agent2:

答案：

Agent1: ✅ Natural language understanding (NLU)
Agent2: ✅ Natural language understanding + (NLU+)

共同要求：

✅ Each agent must use a standard model
✅ NOT use generative orchestration

所以可以先排除：

Azure OpenAI ❌（属于生成式 AI，不符合要求）
Azure Language in Foundry Tools ❌（属于 Azure AI 服务，不是 Copilot Studio 的标准语言模型）

真正需要在下面三者中选择：

CLU
NLU
NLU+

Agent1

要求：

support simple and short phrases for a given topic

关键词：

simple
short phrases
given topic

这是 Natural Language Understanding (NLU) 的典型应用。

NLU：

Intent Recognition
Trigger phrases
简单短语
规则式 Topic 匹配

例如：

"Reset password"

"My order"

"Track shipment"

都属于简单 Trigger Phrase。

因此：

✅ Agent1 → NLU

Agent2

要求：

integrate with Microsoft Dynamics 365 Contact Center voice channel

关键词：

Voice
Contact Center

微软官方建议：

Dynamics 365 Contact Center Voice Agent

使用： NLU+

原因： NLU+：

更适合 Voice
Better intent detection
Better speech recognition integration
更自然的对话

微软文档中：

Voice Channel 推荐：

Natural Language Understanding+ (NLU+)

因此：

✅ Agent2 → NLU+

为什么不是 CLU？

很多人容易混淆。

CLU（Conversational Language Understanding）：属于 Azure AI Language。

主要用于：

Azure Bot
Azure AI Language

而不是 Copilot Studio Standard Model。

题目已经说明：

must use a standard model

Copilot Studio Standard Models：

NLU
NLU+

不是 CLU。

为什么其他选项错？

Azure OpenAI ❌ Generative AI。

题目明确：

must NOT use generative orchestration

排除。

Azure Language in Foundry Tools ❌

Azure AI 服务。不是 Copilot Studio Standard Language Model。

CLU ❌

Azure AI Language 的模型。不是 Copilot Studio Standard Model。

最终答案

Agent	Language Model
Agent1	✅ Natural language understanding (NLU)
Agent2	✅ Natural language understanding + (NLU+)

考试速记

模型	适用场景
NLU	简单 Trigger Phrase、短语匹配、经典 Topic
NLU+	Voice、Dynamics 365 Contact Center、更自然的意图识别
CLU	Azure AI Language（不是 Copilot Studio Standard Model）
Azure OpenAI	Generative Orchestration / LLM
Azure Language in Foundry Tools	Azure AI Language 服务

Box 1: Natural Language Understanding (NLU)

Box 2: Natural Language Understanding + (NLU +)

Explanation/Reference:

Box 1: Natural Language Understanding (NLU)

Agent1 must support simple and short phrases for a given topic.

For a Microsoft Copilot Studio agent that must not use generative orchestration and requires support for simple, short trigger phrases, the best choice is the Classic NLU (Natural Language Understanding) model.

When you disable generative orchestration (also known as "Generative mode" or "Generative AI" orchestration), the agent reverts to Classic orchestration. In this mode, the agent relies on predefined trigger phrases to map user input directly to specific topics.

Box 2: Natural Language Understanding + (NLU +)

Agent2 must integrate with Microsoft Dynamics 365 Contact Center voice channel.

For a Microsoft Copilot Studio agent using classic orchestration (no generative orchestration) and integrating with the Dynamics 365 Contact Center voice channel, the best language model is NLU+.

Why NLU+ is the Best Choice While standard agents offer three "classic" Natural Language Understanding (NLU) options, NLU+ is specifically designed for high-performance, enterprise-grade scenarios like voice channels.

Note:

Comparison of Classic Models

Model	Best For	Key Characteristic
NLU	Simpler bots	Uses 5-20 phrases per topic; basic RegEx/List entities.
NLU+	Voice / Enterprise	Optimizes speech recognition and handles large training sets with high accuracy.
CLU	Advanced devs	Integrates with Azure Conversational Language Understanding for external model management.

QUESTION 3 💩💩💩

A company uses Microsoft Dynamics 365 finance and operations apps.

The company plans to use Microsoft Copilot in-app help and guidance to generate responses for internal business processes.

You need to add an additional knowledge source for the business processes.

The solution must NOT add new topics to the Copilot agent for the finance and operations apps.

Which knowledge source should you add?

A. Microsoft Dataverse

B. a public website

C. Azure AI Search

D. a file upload

正确答案：✅ D. a file upload

题目关键是：

add an additional knowledge source

must NOT add new topics

对于 Dynamics 365 Finance and Operations 的 Copilot in-app help and guidance，可以通过添加文件作为知识源来提供额外的业务流程信息，而不需要创建新的 Topic。

✅ D. File upload → 添加额外知识内容，不需要新增 Topic。
❌ A. Microsoft Dataverse → 不是该场景下 in-app help and guidance 的合适知识源。
❌ B. Public website → 不适合内部业务流程知识。
❌ C. Azure AI Search → 更适合自定义 AI/Agent 的外部知识 grounding，不是这里的最佳选择。

🎯 记忆

Finance & Operations Copilot + additional knowledge + no new topics → File upload

答案：D

Correct Answer: D Section: (none)

To add an additional knowledge source for internal business processes to the Microsoft Copilot in-app experience for Dynamics 365 finance and operations apps—without creating new topics—you should add File Uploads (such as PDF, Word, or text documents) to the "Copilot for finance and operations apps" agent in Copilot Studio.

QUESTION 4

A company has an AI business solution.

You need to extend the solution so that Microsoft 365 Copilot can invoke external logic hosted in Azure services.

What should you include in the solution?

A. Microsoft Copilot Studio skills

B. Microsoft Power Platform connectors

C. custom engine agents

B. Microsoft Power Platform connectors（Microsoft Power Platform 连接器）

核心考点速记（备考复习）

Microsoft 365 Copilot 插件扩展机制：
- 当需要扩展 Microsoft 365 Copilot，使其能够调用托管在 Azure 服务（如 Azure Functions、Azure App Services、API Management 等）中的外部业务逻辑时，最标准、低代码的做法是将 Azure REST API 封装为 Microsoft Power Platform connectors（自定义连接器 Custom Connectors）。
- 将其配置为 Microsoft 365 Copilot 的插件（Plugin / Action）后，Copilot 即可在对话过程中自动识别并触发底层的 Azure 业务逻辑。

Correct Answer: B Section: (none)

To enhance an AI business solution with Microsoft 365 Copilot and integrate external logic hosted in Azure, you should use Copilot Studio to create Actions. These actions act as plugins that allow Copilot to invoke external services through Power Platform components.

Implementation Strategy

Key Components *->

Connector: Wraps the Azure API using an OpenAPI definition or Postman collection.

Plugin/Action: Defines how Copilot identifies when to use the connector based on user prompts.

Authentication: Ensure the connector is configured with appropriate security (e.g., OAuth 2.0) to safely access your Azure resources.

QUESTION 5 💩

You need to design a shared prompt library that will be used across multiple business units. The solution must meet the following requirements:

Ensure consistent AI responses with reusable formats.
Support governance and version control.
Minimize administrative effort.
Minimize ongoing costs.

What should you recommend for each requirement? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Hot Area:

Answer Area

Ensure consistent AI responses:

Delegate department-specific prompt templates.
Define standardized prompt templates.
Maintain a prompt history.

Support governance and version control:

Define standardized prompt templates.
Store prompts in a Git repository.
Categorize prompts by business function.

✅ 答案

Ensure consistent AI responses: → Define standardized prompt templates
Support governance and version control: → Store prompts in a Git repository

🧠 记忆

Consistency → Standardized templates
Governance + Version control → Git

Box 1: Define standardized prompt templates

Box 2: Store prompts in a Git repository

Explanation/Reference:

Explanation:

Box 1: Define standardized prompt templates

Ensure consistent AI responses with reusable formats.

To ensure consistent AI responses across multiple business units, your shared prompt library should be built on a foundation of standardized, modular templates that balance centralized governance with unitspecific flexibility.

Box 2: Store prompts in a Git repository

Support governance and version control.

Storing AI prompts in a Git repository allows you to treat prompts as "first-class artifacts" with the same accountability and lifecycle management as source code. For an enterprise solution serving multiple business units, this approach provides the necessary structure for governance, collaboration, and scalability.

Repository Organization for Business Units
Governance and Version Control Workflow Branching Strategy: Use a dedicated branch for each experiment or new use case (e.g., feature/ marketing-seo-v2) to ensure the main branch remains stable.

Pull Requests (PRs): Mandate PRs for all changes to enable peer reviews. PRs should include descriptions of changes, linked issues, and test results.

Semantic Versioning: Apply tags (e.g., v1.0.1) to mark significant updates, allowing business units to pin their applications to specific, stable prompt versions.

Auditability: Git maintains a full historical record of who changed a prompt, what was modified, and when it occurred.

QUESTION 6 💩

A company has a Microsoft Foundry project that uses a single agent and a single prompt to complete a series of tasks.

The agent encounters the following issues:

It frequently produces incomplete results.
It struggles with domain-specific reasoning.
Agent response times are remarkably slow.

You need to recommend a solution to improve the overall performance and accuracy of the agent.

What should you include in the recommendation?

To answer, drag the appropriate actions to the correct requirements. Each action may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

NOTE: Each correct selection is worth one point.

Select and Place:

Actions

Add a grounding data source.
Add a prebuilt connector.
Move to a multi-agent architecture.
Upgrade to a larger generative AI model.

Answer Area

To improve performance: [______]
To improve accuracy: [______]

Action

[______]
[______]

✅ 答案

To improve performance: → Move to a multi-agent architecture
To improve accuracy: → Add a grounding data source

🧠 记忆：

Performance / slow response → Multi-agent architecture
Accuracy / domain-specific reasoning → Grounding data source
To improve performance: 👉 Move to a multi-agent architecture.（改为多 Agent 架构。）
To improve accuracy: 👉 Add a grounding data source.（添加接地数据源。）

解析与考点分析

提升性能与响应速度 (To improve performance)
对应问题： Agent 响应时间极慢（Response times are remarkably slow），且频繁产生不完整的结果（frequently produces incomplete results）。
诊断分析：
- 题目说明当前系统使用单个 Agent 和单个 Prompt去完成一连串复杂的任务（"a single agent and a single prompt to complete a series of tasks"）。
- 将所有复杂的串行任务塞给单一 Agent 和长 Prompt，会导致模型推理耗时极长、上下文超载（Context Overload）以及 Token 限制导致的回答截断（不完整）。
解决方案： Move to a multi-agent architecture
- 改为多 Agent 架构（Multi-Agent System）可以将庞大的任务拆解给多个专门的子 Agent 协作处理（如主管 Agent 分发任务给专门的子 Agent 模块）。这能显著降低单个 Prompt 的复杂度，减少推断延时，大幅提升整体系统性能和响应速度。
提高回答与推理准确性 (To improve accuracy)
对应问题： Agent 在领域特定推理方面存在困难（struggles with domain-specific reasoning）。
诊断分析：
- 通用大语言模型缺乏企业内部特定领域的最新知识、业务逻辑和私有数据。
解决方案： Add a grounding data source
- 引入接地数据源（Grounding Data Source）（如 RAG 检索增强生成，引入企业文档、知识库或领域数据库），能够向 Agent 提供精准的领域上下文（Domain-specific context），从而消除模型幻觉，显著提升领域推理的准确性（Accuracy）。

Box 1: Move to a multi-agent architecture

Box 2: Add a grounding data source

Explanation/Reference:

Box 1: Move to a multi-agent architecture To improve performance

To improve performance

Moving to a multi-agent architecture in Azure AI Foundry is a highly effective strategy to overcome performance bottlenecks, as single-agent systems often struggle with long-running tasks, leading to high latency and timeout issues. By decomposing complex tasks into smaller, specialized subtasks, you can improve response times through parallel processing and targeted tool usage.

Incorrect:

Upgrade to a larger generative AI model

To address slow response times in your Microsoft Foundry agent, upgrading to a larger generative AI model is one option, but it may increase latency in some scenarios due to higher processing demands. Instead, a combination of prompt optimization, model selection, and architectural changes in Microsoft Foundry is recommended to improve performance.

Box 2: Add a grounding data source

To improve accuracy

To improve the performance of an agent in a Microsoft Foundry project experiencing incomplete results, weak domain reasoning, and high latency, adding a grounding data source is a highly effective strategy. Grounding connects the Large Language Model (LLM) to verified external data, ensuring responses are accurate, contextual, and less likely to hallucinate.

QUESTION 7 💩

A financial services company uses Microsoft Dynamics 365 Finance.

Currently, the company's support staff manually reviews customer transaction histories to detect potential fraud cases before escalating the cases.

You need to recommend an automation solution for the review process.

The solution must ensure that escalations reach a human analyst for final decision making. What should you recommend?

A. Deploy an autonomous agent that closes non-fraud cases automatically.

B. Use Microsoft 365 Copilot in Word to automatically finalize fraud detection policies.

C. Configure a task agent to generate fraud risk scores for the human analyst to review.

D. Export the data to a data lake for analysis in Microsoft Power BI.

C. Configure a task agent to generate fraud risk scores for the human analyst to review.（配置一个任务 Agent 生成欺诈风险评分，供人工分析师进行审核。）

解析与考点分析

为什么选 C？

业务场景：金融服务公司希望将原本人工审查交易历史以识别潜在欺诈的过程自动化。
核心约束条件（Human-in-the-Loop）：
- 题目明确要求：“必须确保升级后的案件到达人工分析师手中以做出最终决策（The solution must ensure that escalations reach a human analyst for final decision making）”。
解决方案分析：
- 配置任务 Agent 生成风险评分： AI Agent 负责自动化繁琐的数据分析和模式识别工作，自动生成欺诈风险评分（Fraud risk scores），并将其呈现给人工分析师（Human Analyst）。
- 人工分析师作为“人机协同（Human-in-the-Loop）”中的最终决策者，基于 AI 提供的评分和分析做出是否升级或封禁的最终决定。这既实现了审核流程的自动化和提效，又完全符合负责任 AI（Responsible AI）的高风险场景合规要求。

为什么不选其他选项？

A. Deploy an autonomous agent that closes non-fraud cases automatically（部署一个自主 Agent 自动关闭非欺诈案件）：完全自主关闭案件缺乏人工监督，在金融欺诈识别这种高风险场景中容易因误判（假阴性）造成重大损失，且违背了题目要求的“由人工分析师做出最终决策”。
B. Use Microsoft 365 Copilot in Word to automatically finalize fraud detection policies（使用 Word 中的 Copilot 自动敲定欺诈检测策略）：在 Word 中撰写策略文档无法自动化交易历史的审查与欺诈检测流程。
D. Export the data to a data lake for analysis in Microsoft Power BI（导出数据到数据湖以便在 Power BI 中分析）：数据报表展示只能提供离线数据可视化，无法提供实时的自动化欺诈风险评分与事件升级工作流。

Correct Answer: C

Explanation/Reference:

Explanation:

To automate the fraud review process in Microsoft Dynamics 365 Finance while ensuring a human analyst makes the final decision, you should configure the Dynamics 365 Fraud Protection (or integrated Copilot AI agents) to generate risk scores and route high-risk transactions to a manual review queue.

Here is the configuration approach to achieve this:

Implement AI-Driven Risk Scoring: Utilize Dynamics 365 Fraud Protection, which uses AI to analyze customer transaction history and assign a risk score (0-999) to each transaction.
Define Rules for Escalation: Set up fraud rules in the system to determine which transactions require human intervention. For instance, define a threshold (e.g., a "Minimum score value") where transactions with high fraud probability are automatically flagged.
Establish Manual Review Queues: Configure the Manual Review tool to create queues for suspected fraudulent transactions, allowing human analysts to review the AI-generated risk score and transaction history, such as customer behavior, for final, informed decision-making.
Use Copilot/AI Agents for Monitoring: Enable AI agents to continuously monitor financial data, such as invoice, payment, and vendor data, and generate alerts for unusual patterns before escalating.

This setup, particularly through the Manual Review workspace, allows for an automated, intelligent, and scalable approach to fraud management.

QUESTION 8 💩

A company plans to deploy a Microsoft Copilot Studio agent that will analyze historical business data to predict customer behavior.

The data is currently stored in an Azure SQL database, flat files, APIs, and logs.

You need to organize the data into a format that can be used as a knowledge source in Copilot Studio.

What should you include in the solution?

A. Azure AI Search

B. Azure Data Lake Storage

C. Azure Cosmos DB

D. Azure Translator in Foundry Tools

正确答案：✅ A. Azure AI Search

原因

题目中的关键是：

数据来自 Azure SQL、Flat files、APIs、Logs
数据来源很多、格式不同
需要整理后作为 Copilot Studio Knowledge Source
用于历史数据的检索和分析

Azure AI Search 可以把来自不同来源的数据建立索引，并提供搜索/RAG 能力，适合作为 Copilot Studio 的知识来源。

其他选项

❌ Azure Data Lake Storage → 适合集中存储大量数据，但本身不是知识检索层。
❌ Azure Cosmos DB → NoSQL 数据库，不是主要的知识检索方案。
❌ Azure Translator → 翻译服务。

最终答案：A. Azure AI Search

Correct Answer: A

Explanation/Reference:

Microsoft Copilot Studio agents can analyze customer behavior by leveraging business data from Azure SQL, files, and APIs by using Azure AI Search as a knowledge source. By importing and vectorizing this structured and unstructured data into an Azure AI Search index, the agent can perform semantic, meaning-based searches to retrieve context-relevant information.

QUESTION 9 💩

A retail company plans to deploy Microsoft Copilot Studio agents to support:

Microsoft Dynamics 365 Commerce scenarios.
A Microsoft Power Apps inventory management solution.

You need to recommend a solution to organize product catalog data as a consistent source for multiple AI systems.

What should you recommend?

A. Let each agent scrape product details from Microsoft SharePoint Online libraries.

B. Store the product catalog data in a separate custom table for each agent.

C. Configure prompts to pull product details from the PDFs of external vendors.

D. Centralize the product catalog data in Microsoft Dataverse and expose the data to both agents.

正确答案与详细解析

正确答案：

D. Centralize the product catalog data in Microsoft Dataverse and expose the data to both agents.（将产品目录数据集中在 Microsoft Dataverse 中，并将数据公开给两个 Agent。）
业务场景与数据一致性需求：
- 公司需要同时为 Dynamics 365 Commerce（电商/零售场景）和 Power Apps（库存管理方案）部署 Copilot Studio Agent。
- 核心要求是：建立一个一致的数据源（Consistent source / Single Source of Truth）供多个 AI 系统共享使用。
微软架构最佳实践：
- Microsoft Dataverse 是 Power Platform 和 Dynamics 365 体系的核心数据底座（Common Data Model）。
- 将产品目录集中存储在 Dataverse 中，可以实现：
  - 单一事实来源（SSOT）：避免不同系统和 Agent 之间的数据冗余与不一致。
  - 原生无缝集成： Dynamics 365 Commerce、Power Apps 和 Copilot Studio 均能够以极低的开发成本原生的读写 Dataverse 中的表与数据，确保数据更新实时同步给所有 AI Agent。

为什么不选其他选项？

A. Let each agent scrape product details from Microsoft SharePoint Online libraries（让每个 Agent 从 SharePoint 抓取产品信息）：从非结构化文档库抓取产品数据不够精准且响应较慢，无法保证电商和库存系统所需的高一致性结构化数据。
B. Store the product catalog data in a separate custom table for each agent（为每个 Agent 分别存储在独立的自定义表中）：这会导致数据孤岛和数据冗余，一旦产品信息更新就需要多处修改，直接违背了“一致的数据源”的要求。
C. Configure prompts to pull product details from the PDFs of external vendors（配置 Prompt 从外部供应商的 PDF 中拉取）：通过 Prompt 实时解析 PDF 属于非结构化文档提取，准确率受限且无法满足零售与库存场景下的结构化业务查询需求。

Correct Answer: D

Explanation:

In the scenario described, centralizing product catalog data in Microsoft Dataverse is the recommended architectural approach to ensure consistency across multiple AI systems. Dataverse acts as a unified "knowledge network" that allows different agents to share a single source of truth for both structured and unstructured data.

Key Benefits of Centralizing in Dataverse

Cross-App Consistency: By storing the catalog in Dataverse, both the Dynamics 365 Commerce agent and the Power Apps inventory management agent access the exact same records. This prevents data silos where inventory levels or product descriptions might drift apart between systems.

Native Copilot Studio Integration: You can directly add Dataverse tables as knowledge sources in Microsoft Copilot Studio. This allows agents to use Retrieval-Augmented Generation (RAG) to answer queries grounded in your live product data.

Security & Governance: Dataverse enforces Role-Based Access Control (RBAC), ensuring that agents only interact with data they are authorized to see, based on the user's existing permissions.

Automated Updates: You can use Power Platform Dataflows to ingest and synchronize catalog data from external sources into Dataverse, keeping the information fresh for all connected AI agents.

QUESTION 10 💩💩💩

A company has a portfolio of AI initiatives at different stages of development.

You need to recommend a structured approach to evaluating the return on AI investment (ROAI) across all the initiatives.

The solution must balance immediate results with long-term values and strategic innovations.

What should you include in the recommendation?

A. a simple cost and benefit analysis

B. a horizon-based framework

C. the internal rate of return (IRR) function

D. a prioritization grid

✅ 答案：B. a horizon-based framework

原因：

多个不同阶段的 AI initiatives
Immediate results
Long-term value
Strategic innovations

这正是 Horizon-based framework 的使用场景：

Horizon 1 → 当前业务、短期收益
Horizon 2 → 新业务、成长机会
Horizon 3 → 长期战略创新

🧠 记忆：

Short-term + Long-term + Innovation → Horizon-based framework

答案：B ✅

Correct Answer: B

Explanation/Reference:

In this scenario, a horizon-based framework is best for evaluating the return on AI investment (ROAI). While a prioritization grid is excellent for immediate tactical choices (e.g., effort vs. impact), a horizonbased framework is specifically designed to balance diverse timeframes—mapping immediate efficiency gains alongside the long-term strategic and transformational value required for a comprehensive Microsoft AI portfolio.

Why the Horizon-Based Framework Wins

This approach categorizes AI initiatives into three distinct "horizons," allowing you to measure different types of value across the development lifecycle:

Horizon 1: Core Operations (Immediate ROI)

Focuses on extending current capabilities for rapid results, such as using Microsoft Copilot to automate routine coding tasks or IT support.

*Horizon 2: Adjacent Opportunities (Mid-Term Value)

Targets growth by expanding into new areas related to your core, such as developing agentic platforms or AI-driven specialized tools that build on existing infrastructure.

Comparison for This Scenario

Horizon 3: Transformational Innovation (Long-Term Strategy)

Invests in "future-forward" innovations that may have uncertain immediate returns but offer massive strategic upside or business model reimagination.

Feature	Priority Grid (e.g., 2x2 Matrix)	Horizon-Based Framework
Primary Goal	Selecting "what to do first" based on constraints.	Managing a balanced portfolio of value over time.
Best For	Tactical, sprint-level feature ranking.	Strategic roadmap and investment allocation.
ROAI Focus	Often emphasizes "quick wins" and high-certainty projects.	Validates high-risk, high-reward innovations as essential for long-term health.

Recommendation: Use the Microsoft AI Maturity Model to baseline your current state, then apply the Horizon-Based Framework to structure your ROAI evaluation. This ensures you don't starve long-term innovation in favor of only "low-hanging fruit" efficiency projects.

QUESTION 11 💩

You need to recommend a Microsoft Power Platform business solution that consolidates data from multiple internal and external data sources.

The solution must meet the following requirements:

Provide the data as a centralized source for multiple AI systems, including Microsoft Copilot Studio agents, Dynamics 365 applications, and external AI models.
Support built-in data classification and protection policies.
Provide data for grounding and analytics.

What should you include in the recommendation?

A. Microsoft Dataverse

B. Azure Data Lake Storage

C. a Microsoft Power BI semantic model

D. Azure Cosmos DB

正确答案：✅ A. Microsoft Dataverse

题目几个关键词基本直接指向 Dataverse：

Centralized source → 可作为 Power Platform / Dynamics 365 的统一数据平台
Copilot Studio agents → 原生支持 Dataverse 作为知识/数据来源
Dynamics 365 → Dynamics 365 应用原生建立在 Dataverse 上
External AI models → 可通过 API/connectors 等方式访问
Data classification & protection policies → 支持 Power Platform / Purview 等治理和数据保护能力
Grounding + analytics → 可作为 AI grounding 和业务分析的数据基础

其他选项

❌ B. Azure Data Lake Storage → 强于大规模数据湖和分析，不是 Power Platform/Dynamics 统一业务数据平台。
❌ C. Power BI semantic model → 主要用于分析语义层，不适合作为多个 AI 系统的中央业务数据源。
❌ D. Azure Cosmos DB → NoSQL/全球分布式数据库，虽然可以作为 AI 数据源，但不是本题要求的 Power Platform 中央业务数据平台。

🎯 最终答案

A. Microsoft Dataverse ✅

Correct Answer: A

Explanation/Reference:

Microsoft Dataverse is the ideal foundational component for this scenario, serving as the secure, centralized data platform for the Microsoft Power Platform, Dynamics 365, and AI integrations. By using Dataverse, the business can consolidate, protect, and utilize data across Copilot Studio agents, Dynamics 365 applications, and analytics tools, addressing all stated requirements.

Here is how Dataverse fulfills the requirements in the described scenario:

Centralized Source: It serves as the primary data storage and management engine for Power Platform, Dynamics 365, and Copilot Studio. It consolidates data into a unified data model (Common Data Model), ensuring consistency across internal and external sources.

Internal & External Integration: Dataverse supports data integration from diverse sources using tools like Power Query, Azure Data Factory, and virtual tables that map external data in real-time.

Data Classification & Protection: It features built-in, enterprise-grade security.

AI Grounding & Analytics:

Grounding: It acts as the knowledge source for Copilot Studio agents, providing the necessary business context for accurate generative AI responses.
Analytics: It integrates natively with Power BI for reporting and can synchronize with Microsoft Fabric or Azure Synapse for large-scale data modeling and advanced analytics.

QUESTION 12 💩💩

A company plans to deploy an AI-based customer service app that will autonomously manage interactions, escalate complex cases, and learn from historical ticket data.

You need to perform a return on AI investment (ROAI) analysis of the app deployment. The solution must ensure that the analysis is accurate.

What should you do first?

A. Establish the AI performance metrics.

B. Conduct an AI market benchmarking study.

C. Model the customer experience.

D. Identify and quantify all the development, deployment, and operating costs.

正确答案：✅ D. Identify and quantify all the development, deployment, and operating costs.

原因

ROAI（Return on AI Investment）的基础公式：

[ ROAI = \frac{\text{Benefits} - \text{Costs}}{\text{Costs}} ]

因此要进行准确的 ROAI 分析，第一步必须先建立成本基线（Cost Baseline），包括：

Development costs（开发成本）
Deployment costs（部署成本）
Operating costs（运营成本）
Maintenance costs（维护成本）

如果连成本都没有量化，就无法计算投资回报率。

其他选项

❌ A. Establish the AI performance metrics 有助于评估效果，但不能单独计算 ROAI。
❌ B. Conduct an AI market benchmarking study 属于市场分析，不是 ROAI 的第一步。
❌ C. Model the customer experience 用于评估业务影响，不是计算投资回报的起点。

🎯 AB100 / AI-900 类考试记忆

ROAI Analysis First Step → Establish Cost Baseline

→ Identify and quantify development, deployment, and operating costs

答案：D ✅

Correct Answer: D

Explanation/Reference:

Explanation:

To conduct a robust Return on AI Investment (ROAI) analysis for your Microsoft-based AI customer service application, you must first categorize and quantify three distinct cost phases: Development, Deployment, and Operations. For a system capable of managing complex escalations and learning from historical data, your project aligns with "Advanced" or "Agentic" AI profiles.

1.Development Costs (Upfront Investment)

This phase covers the creation of the core AI logic, custom integrations, and data preparation.

2.Deployment Costs (One-Time Setup)

These are the costs to move the application from a development environment to a live production state.

3.Operating & Maintenance Costs (Recurring)

Ongoing expenses are critical for ROAI as they impact the net gain over time.

QUESTION 13

You are designing end-to-end test scenarios for a business solution that uses Microsoft Dynamics 365 Sales and Dynamics 365 Finance.

You need to ensure that the business solution meets the following test requirements:

Properly exchanges data between the Dynamics 365 apps
Aligns with defined user workflows and business processes

Which type of testing should you use for each requirement?

To answer, drag the appropriate testing types to the correct requirements. Each testing type may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

NOTE: Each correct selection is worth one point.

Select and Place

Testing types

Drift
Exploratory
Integration
Performance
User acceptance

Answer Area

Properly exchanges data between the Dynamics 365 apps. []
Aligns with defined user workflows and business processes. []

这是该道拖拽匹配题（Select and Place）的正确答案与详细解析：

QUESTION 答案 (Answer Area)

Properly exchanges data between the Dynamics 365 apps: 👉 Integration（集成测试）
Aligns with defined user workflows and business processes: 👉 User acceptance（用户验收测试 / UAT）

解析与考点分析

1. 在 Dynamics 365 各应用之间正确交换数据 (Properly exchanges data between the Dynamics 365 apps)

正确选择： Integration
解析：
验证不同系统或应用之间（如 Dynamics 365 Sales 与 Dynamics 365 Finance，通过 Dual-write 或 Dataverse）的数据流动、接口对接以及数据同步正确性，属于标准的集成测试（Integration Testing）范畴。

2. 符合已定义的用户工作流和业务流程 (Aligns with defined user workflows and business processes)

正确选择： User acceptance
解析：
用户验收测试（UAT, User Acceptance Testing）： 由业务专家（SMEs）或最终用户按照真实的业务流程和工作流进行测试，旨在验证整个解决方案是否满足业务需求，以及用户能否顺畅完成既定的日常工作。

其他干扰项排除说明

Drift（漂移测试）： 主要指 AI/ML 中的模型漂移（Model Drift）或数据漂移（Data Drift）监测，用于评估模型准确率随时间衰减的情况，与系统流程测试无关。
Exploratory（探索性测试）： 无脚本的随机测试，用于发现意外的缺陷，不适用于验证既定的业务流程规范。
Performance（性能测试）： 用于验证高并发或大数据量下的响应时间和吞吐量，与数据交换逻辑和业务流程对齐无关。

Box 1: Integration

Box 2: User acceptance

Box 1: Integration

Properly exchanges data between the Dynamics 365 apps

For a project involving Dynamics 365 Sales and Dynamics 365 Finance, the best testing type to verify proper data exchange is Integration Testing, specifically within a System Integration Testing (SIT) phase.

Integration Testing

This is the primary method for validating that data flows correctly between Sales (Customer Engagement) and Finance (Finance & Operations).

Purpose: Ensures seamless information flow, communication protocols, and data synchronization. Example Scenario: Creating a new Account in D365 Sales and verifying it automatically triggers the creation of a Customer in D365 Finance via Dual-write.

Box 2: User acceptance

Aligns with defined user workflows and business processes

The best testing type to ensure a Microsoft Dynamics 365 solution aligns with defined user workflows and business processes is User Acceptance Testing (UAT).

While other testing phases (like Functional or Integration) verify that the system works technically, UAT is the final "road test" where the business confirms that the solution actually supports their day-to-day operations.

QUESTION 14 💩💩💩

A company has a Microsoft 365 tenant in Canada and multiple Microsoft Power Platform environments in Canada and the United States.

The company plans to deploy a Microsoft Copilot Studio agent to the Canadian environment that will use:

Microsoft Dataverse data stored in Canada A connector that connects to an Azure OpenAI instance in the United States

You need to ensure that the agent adheres to data residency and data movement policies before being deployed.

What should you do?

A. Ensure that the data processed by Azure OpenAI is stored in the United States.

B. From the Microsoft Purview portal, validate the Data loss prevention settings.

C. Migrate the tenant to the United States.

D. Ensure that cross-region data movement is enabled for the Canadian environment and connector dependencies.

正确答案：✅ D. Ensure that cross-region data movement is enabled for the Canadian environment and connector dependencies.

原因

Dataverse 数据在 Canada
Azure OpenAI 在 United States
Agent 部署在 Canadian Power Platform environment
需要满足 data residency / data movement

因为 Agent 需要跨区域访问 Azure OpenAI，所以必须确保该环境允许 cross-region data movement，并且相关 connector dependencies 也符合要求。

其他选项

❌ A：只是说明数据存储位置，并不能确保跨区域数据移动符合策略。
❌ B：Purview DLP 主要用于数据治理/防泄漏，不是 Copilot Studio 跨区域数据移动的核心设置。
❌ C：没必要把整个 tenant 迁移到美国。

🎯 记忆

Copilot Studio + Dataverse Canada + Azure OpenAI US → Cross-region data movement must be enabled

答案：D

Correct Answer: D

Explanation/Reference:

Explanation:

Manage Cross-Geo Data Movement Settings

Because your agent in Canada connects to Azure OpenAI in USA, data will leave your primary geography.

Explicit Opt-in:

You must navigate to the Power Platform Admin Center and explicitly allow data movement across regions for generative AI features.

Regional Scope:

If this toggle is off, Copilot features requiring the external Azure OpenAI instance will be disabled to protect residency.

Security Assurance: Microsoft does not log or retain prompt/response data during this cross-region transit for the purpose of training foundation models.

QUESTION 15 💩💩💩

A company has a Microsoft Copilot Studio agent for customer support. You are reviewing and validating the following prompts:

A prompt that has instructions to "help the customer as best you can"
A prompt that helps retrieve product information from a knowledge base
You need to ensure that the agent delivers consistent and accurate responses.

What should you do for each prompt? To answer, select the appropriate options in the answer area. NOTE: Each correct selection is worth one point.

Hot Area:

A prompt that has instructions to "help the customer as best you can":

Add filler words to make the prompt sound more natural and conversational.
Keep the prompt vague to enable model flexibility.
Rewrite the prompt with clear and task-specific instructions.

A prompt that helps retrieve product information from a knowledge base:

Add several open-ended questions to give the model broader context.
Use responses with only reference sources and limit the response scope.
Remove the knowledge source so that the model responds freely with general product information.

答案：

Requirement	Correct Answer	原因
A prompt that has instructions to "help the customer as best you can"	✅ Rewrite the prompt with clear and task-specific instructions.	Prompt 应明确、具体，避免模糊指令，提高回答一致性。
A prompt that helps retrieve product information from a knowledge base	✅ Use responses with only reference sources and limit the response scope.	限制回答仅基于知识库，可减少幻觉（Hallucination），提高准确性。

考试记忆

Prompt Engineering：明确、具体、任务导向（Clear & Task-specific）
RAG / Knowledge Base：Only use reference sources，限制回答范围，减少生成错误信息

最终答案：

第一空： ✅ Rewrite the prompt with clear and task-specific instructions.
第二空： ✅ Use responses with only reference sources and limit the response scope.

Box 1: Rewrite the prompt with clear and task-specific instructions.

To ensure a Microsoft Copilot Studio agent delivers high-quality support, you must move beyond vague instructions like "help as best you can." Generative AI requires constraints, structured data, and persona definitions to remain consistent.

Here is how to optimize your agent for accuracy and reliability:

Refine the System Prompt A vague prompt leads to "hallucinations" (made-up facts). Replace the general instruction with a Persona and Boundary script.

Define the Persona: "You are a professional Support Specialist for [Company Name]. Your tone is empathetic, concise, and helpful."

Set Boundaries: "Only answer questions based on the provided uploaded documents. If the answer is not in the documentation, politely state you don't know and offer to connect to a human."

Establish Steps: "Always greet the user, identify their core issue, and provide a step-by-step solution."

Box 2: Use responses with only reference sources and limit the response scope.

To ensure your Microsoft Copilot Studio agent remains grounded, accurate, and avoids "hallucinations," you must configure the System Prompt to enforce strict grounding. This tells the AI to act only as a retrieval tool rather than a general-purpose assistant.

Restrict Search Scope:

In the Generative Answers node, ensure "Search only selected sources" is toggled on.

Upload specific PDFs, Manuals, or SharePoint folders rather than using the entire public web

QUESTION 16 💩💩

You are designing a testing solution for Microsoft Copilot Studio agents.

You need to validate prompt engineering best practices to ensure that the agents generate accurate and contextually relevant responses.

Which prompt validation techniques and metrics should you include in the solution? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Answer Area

Prompt validation techniques:

Exclude domain-specific terminology from the prompts.
Use prompts that have varied phrasing.
Use only simple, one-word prompts.

Metrics:

The number of words generated per response
Response relevance and accuracy
The response generation time

答案：

Prompt validation techniques: ✅ Use prompts that have varied phrasing.
Metrics: ✅ Response relevance and accuracy.

原因

Varied phrasing 可以测试 Agent 对不同自然语言表达方式的鲁棒性，避免只对某一种固定问法有效。

Response relevance and accuracy 直接衡量回答是否：

与用户问题相关
内容准确
符合上下文

其他选项：

❌ Exclude domain-specific terminology → 企业 AI 通常需要正确处理领域术语。
❌ Only simple one-word prompts → 无法充分测试复杂真实场景。
❌ Number of words → 长度不代表质量。
❌ Response generation time → 属于性能指标，不是本题要求的 prompt quality 指标。

🎯 最终

Prompt validation → Use prompts that have varied phrasing

Metric → Response relevance and accuracy

Box 1: Use prompts that have varied phrasing

Here is why this technique is superior to the others for validating prompt engineering best practices:

Robustness and Generalization: Using varied phrasing (synonyms, different sentence structures) tests whether the agent understands the underlying intent rather than just matching keywords.

This ensures the agent is not brittle and can handle real-world user queries that vary in phrasing.

Best Practice Validation: Microsoft encourages testing agent responses by iterating and refining prompts. Testing with variations helps identify if a prompt is too rigid or if it correctly captures the context.

For a comprehensive validation strategy, this technique is typically paired with Batch Testing, where you upload a set of varied test cases to evaluate the AI's consistency and accuracy.

Incorrect:

Exclude domain-specific terminology This is incorrect. Effective prompting often requires using domain-specific terms to ensure the AI understands the specialized context.
Use only simple, one-word prompts This is ineffective. Simple prompts lack the context required for high-quality, accurate responses.

Box 2: Response Relevance and accuracy

By using Response Relevance and Accuracy as your primary metrics, you ensure the agent remains grounded in its knowledge sources while remaining flexible to user intent.

QUESTION 17 💩

A company has two Microsoft Power Platform environments named Dev1 and Prod1. A Microsoft Copilot Studio agent named Agent1 is built into a solution in the Dev1 environment.

You plan to deploy Agent1 to Prod1.

You need to make Agent1 available to the users in Prod1. The solution must minimize administrative effort.

What should you do?

A. Share Agent1 with the users in Prod1.

B. Export the solution as an unmanaged solution and import the solution into Prod1.

C. Export the solution as a managed solution and import the solution into Prod1.

D. Create a new Copilot Studio agent in Prod1 by replicating the configuration of Agent1.

正确答案与详细解析

正确答案：

C. Export the solution as a managed solution and import the solution into Prod1.（将解决方案导出为托管解决方案，并将其导入 Prod1 中。）

解析与考点分析

为什么选 C？

Power Platform ALM（应用生命周期管理）最佳实践：
- 在 Microsoft Power Platform 体系中，将应用、Flow 或 Copilot Studio Agent 从开发环境（Dev1）部署到生产环境（Prod1）时，标准的官方规范是导出为托管解决方案（Managed Solution）。
满足“最小化管理工作量（Minimize administrative effort）”：
- 托管解决方案（Managed Solution）是用于非开发环境（测试/生产环境）的标准部署包。
- 导入为托管解决方案后，可以防止生产环境中被意外直接修改（锁定生产环境），同时后续可以通过覆盖或更新版本轻松进行全自动升级与卸载，大大降低了长期的系统运维和管理成本。

为什么不选其他选项？

A. Share Agent1 with the users in Prod1（与 Prod1 中的用户共享 Agent1）： Agent1 当前存在于 Dev1 环境中，仅跨环境“共享”无法将 Agent 真正部署并运行在 Prod1 生产环境的数据源和隔离策略下。
B. Export the solution as an unmanaged solution and import the solution into Prod1（导出为非托管解决方案导入 Prod1）：非托管解决方案（Unmanaged Solution）只能用于开发环境。如果导入生产环境，会导致生产环境中产生非托管图层（Unmanaged layers），后续无法进行干净的解决方案升级和卸载，极大地增加管理工作量。
D. Create a new Copilot Studio agent in Prod1 by replicating the configuration of Agent1（在 Prod1 中通过复制配置新建）：手动重新创建和配置 Agent 属于极度繁琐且容易出错的人工重复劳动，严重违背了“最小化管理工作量”的要求。

Correct Answer: C

Explanation/Reference:

Explanation:

To deploy a Microsoft Copilot Studio agent to a second environment while minimizing administrative work, you should use managed solutions and Power Platform Pipelines (or manual export/import if pipelines are not configured).

Follow these steps to move and share the agent effectively:

Prepare the Solution (Source Environment)

Since your agent is already in a solution, ensure all its dependencies are included to avoid "missing component" errors during import.

Export and Import

To minimize manual labor, use a Managed Solution for the target environment. Managed solutions are easier to update or remove later and prevent accidental direct edits in production.

QUESTION 18 💩

A company has a Microsoft Power Platform environment that contains Microsoft Dataverse data.

You create a Microsoft Copilot Studio agent named Agent1 that processes the Dataverse data.

You discover that Agent1 fails to return relevant or accurate results.

You need to improve the quality and reliability of data grounding.

What should you do?

A. Retrain Agent1.

B. Verify and cleanse the Dataverse data.

C. Use an adaptive card in Agent1.

D. Add example user inputs to the training data of Agent1.

✅ 答案：B. Verify and cleanse the Dataverse data.

原因： Agent 的 grounding 依赖底层 Dataverse 数据。数据存在错误、重复、缺失或不一致，会直接导致回答不准确。

A ❌ Copilot Studio 不需要通过 retraining 来解决 Dataverse 数据质量问题
B ✅ Verify and cleanse the Dataverse data
C ❌ Adaptive Card 主要用于交互界面
D ❌ Example inputs 不能解决 grounding data 本身的问题

🧠 记忆：

Grounding 不准确 → 先检查和清洗数据

Correct Answer: B

Explanation/Reference:

To improve the reliability and accuracy of a Microsoft Copilot Studio agent grounded in Dataverse, you must focus on Dataverse data preparation, which includes cleaning, structuring, and enriching the data to ensure the agent retrieves relevant information. When Copilot fails to return accurate results, it is often due to poor data quality (garbage in, garbage out), excessive, unstructured data, or misconfigured search indexing.

QUESTION 19 💩💩💩

A company plans to deploy a Microsoft Copilot Studio agent to enhance customer support.

The company stores customer data across ServiceNow, Microsoft Dynamics 365 Finance, Dynamics 365 Supply Chain Management, and Excel files in SharePoint Online.

You need to recommend a solution to ensure that the agent can deliver accurate and timely responses.

What should you recommend?

A. Implement a model router for query handling.

B. Create custom prompts.

C. Implement Microsoft Power Platform connectors.

D. Enable incremental indexing in Azure AI Search.

正确答案与详细解析

正确答案：

C. Implement Microsoft Power Platform connectors.（实现 Microsoft Power Platform 连接器。）

解析与考点分析

为什么选 C？

业务痛点：公司的客户数据散落在多个不同的异构系统和数据源中，包括：
- 第三方 SaaS 系统：ServiceNow
- Dynamics 365 企业应用：Dynamics 365 Finance 和 Dynamics 365 Supply ChainManagement
- 文件与文档：存储在 SharePoint Online 中的 Excel 文件
技术解决方案： Microsoft Copilot Studio 建立在 Microsoft Power Platform 之上。为了让 Copilot Studio Agent 能够实时跨这些不同的系统检索和读取数据，最标准、最直接且低代码的方式就是使用 Microsoft Power Platform Connectors（连接器）。
Power Platform 拥有 1,000+ 现成的内置连接器（涵盖 ServiceNow、Dynamics 365 各模块、SharePoint/Excel 等）。通过配置这些连接器（作为知识源 Knowledge Sources 或 Plugin Actions），Agent 可以直接调用 API 获取最新的实时数据，从而确保回答的准确性与及时性（accurate and timely responses）。

为什么不选其他选项？

A. Implement a model router for query handling（实现用于查询处理的模型路由器）：模型路由（Model Routing）主要用于在不同的 LLM 大模型（如 GPT-3.5、GPT-4、Claude 等）之间根据查询复杂度分发请求以优化成本和时延，它无法解决从 ServiceNow 或 Dynamics 365 中检索企业业务数据的问题。
B. Create custom prompts（创建自定义提示词）： Prompt（提示词）只能指导大模型如何思考或格式化输出，如果模型本身没有连接到包含客户数据的后台系统，Prompt 无法凭空产生准确的企业内部数据。
D. Enable incremental indexing in Azure AI Search（在 Azure AI Search 中启用增量索引）：增量索引是 Azure AI Search 针对非结构化文档检索的优化手段。虽然 Azure AI Search 可以作为 RAG 的向量数据库，但对于 Dynamics 365 Finance/SCM 和 ServiceNow 这类包含大量实时事务型数据（Transactional Data）和复杂业务逻辑的系统，直接通过连接器（Connectors）调用系统 API 才是更合适且能够保证实时准确回答的标准做法。

QUESTION 20

A manufacturing company wants to deploy an agent that will automate supplier invoice processing.

You are designing a solution to evaluate the financial implications of the deployment.

The company is especially concerned about budget overruns.

You need to ensure that the solution considers the total cost of ownership (TCO), the expected savings from using automation, and whether to extend the existing AI capabilities.

What should you include in the design?

A. a break-even analysis only

B. adopting prebuilt agents to reduce the deployment time

C. training a custom model

D. a return on AI investment (ROAI) analysis

正确答案：

D. a return on AI investment (ROAI) analysis（AI 投资回报率分析）

解析与考点分析

为什么选 D？

业务场景与核心考量：
- 公司计划部署 AI Agent 来自动化处理供应商发票。
- 关注焦点：评估该部署的财务影响（Financial implications），并非常担心预算超支（Budget overruns）。
- 评估维度要求：必须综合考量总体拥有成本（TCO）、自动化带来的预期成本节省（Expected savings from using automation），以及是否需要扩展现有的 AI 能力。
方法论匹配：
- ROAI (Return on AI Investment，AI 投资回报率分析)：是一种专门针对 AI 项目的全面财务评估方法。
- 它将 AI 项目的总投入成本（TCO，包括 API 调用费、模型训练费、Copilot 订阅费及运维成本）与预期收益/节省（通过自动化发票处理节省的人工工时、减少的人为错误成本等）进行对比计算： \text{ROAI} = \frac{\text{预期自动化节省成本} - \text{TCO总体拥有成本}}{\text{TCO总体拥有成本}} \times 100\%
- 进行全面的 ROAI 分析能够帮助管理层在项目启动前精准进行预算建模，从而直接防止预算超支。

为什么不选其他选项？

A. a break-even analysis only（仅进行盈亏平衡分析）：盈亏平衡分析只告诉你在什么时间点“收回本钱”，无法全面评估 TCO 构成、投资回报率（ROI）以及是否扩展现有 AI 能力的财务可行性。
B. adopting prebuilt agents to reduce the deployment time（采用预置 Agent 以减少部署时间）：这是实施策略选项，而非用于评估财务影响和防止预算超支的分析评估方法。
C. training a custom model（训练自定义模型）：自定义模型训练通常需要极高的算力和数据标注成本，不仅无法解决预算超支担忧，反而可能增加财务风险。

Correct Answer: D

Explanation/Reference:

Explanation:

This design plan outlines the financial and operational framework for deploying a Microsoft-based AI agent for supplier invoice processing. By leveraging current industry benchmarks and Microsoft’s own agentic AI frameworks, this plan focuses on maximizing Return on AI Investment (ROAI) while mitigating the risk of budget overruns.

QUESTION 21 💩💩💩

A company has a Microsoft Power Platform solution that contains the following components:

Microsoft Dataverse tables
A Microsoft Power BI workspace named WS1
A canvas app named App1 that uses Dataverse
A Power BI semantic model that connects to Dataverse by using DirectQuery

You plan to use generative AI to provide answers to queries based on a subset of corporate data.

You need to ensure that the data is available as a grounding data source for AI systems.

What should you do?

A. Populate a Dataverse table.

B. Share WS1.

C. Endorse the semantic model.

D. Export the semantic model.

✅ C. Endorse the semantic model.（对语义模型进行背书/认证。）

解析与考点分析

为什么选 C？

业务场景：公司有一个通过 DirectQuery 连接到 Dataverse 的 Power BI 语义模型（Semantic Model），希望使用生成式 AI（如 Power BI Copilot / Microsoft Copilot / Copilot Studio）基于企业数据子集提供查询解答，并确保该数据能够作为 AI 系统的接地数据源（Grounding data source）
微软架构与最佳实践：
- 在 Power BI / Microsoft Fabric / Copilot 生态系统中，为了让 AI 系统识别、信任并优先使用某个语义模型作为可靠的接地数据源（Grounding Source），必须对该语义模型进行背书（Endorse）（即设置为 Promoted 提升或 Certified 认证）。
- 对语义模型进行背书可以向 AI 系统和用户表明该数据源经过严格治理、数据准确且属于企业权威数据源（Single Source of Truth），从而授权生成式 AI 基于该模型进行数据回答与分析。

为什么不选其他选项？

A. Populate a Dataverse table（填充 Dataverse 表）： Dataverse 已经是底层的元数据存储，但题目中的AI 问答是基于已建立好的 Power BI 语义模型（包含度量值、关系和计算）。仅填充数据表不能直接将语义模型注册/暴露为受信任的 AI 接地数据源。
B. Share WS1（共享 WS1 工作区）：共享工作区只是赋予用户对工作区内资源的访问权限，无法将数据模型提升为 AI系统认可的受信任接地数据源。
D. Export the semantic model（导出语义模型）：导出模型（如导出为文件）会脱离现有的云端 DirectQuery实时连接，断开数据更新，无法作为 AI 系统实时的接地数据源。

Correct Answer: C

Explanation/Reference:

Explanation:

To ensure data is available as a grounding source for generative AI systems within this Power Platform solution, you should endorse the semantic model.

While populating a Dataverse table and sharing the workspace are necessary for general data accessibility, endorsing the Power BI semantic model (through certification or promotion) specifically helps generative AI systems like Power BI Copilot and Microsoft 365 Copilot prioritize and identify it as a high-quality, trusted source for grounding answers.

Why Endorsement is Key for Grounding

Generative AI systems use grounding to reduce "hallucinations" by anchoring their responses in real corporate data. In the Power BI ecosystem:

Discovery: Endorsement makes the semantic model more discoverable across the organization.

Prioritization: Copilot often prioritizes endorsed (Promoted or Certified) models when searching for relevant data to answer a natural language query.

Trust: It signals to both the AI and the end-user that the data in this DirectQuery-connected model is verified and authoritative for business queries.

QUESTION 22 💩💩

A company plans to implement an AI business solution for a consumer goods company.

You need to create agents that meet the following requirements:

Orchestrate the sales order fulfillment and shipping of goods to customers.
Analyze historical data and trends to replenish stock.

Which type of agent should you use for each requirement?

To answer, drag the appropriate agent types to the correct requirements. Each agent type may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

Select and Place

Agent types

Autonomous
Prompt-and-response
Task

Answer Area

Orchestrate the sales order fulfillment and shipping:
Analyze historical data and trends:

答案：

Orchestrate the sales order fulfillment and shipping: ✅ Autonomous
Analyze historical data and trends: ✅ Prompt-and-response

记忆

Autonomous agent → 能自主执行、协调多个步骤的业务流程 → 订单履约和运输
Prompt-and-response agent → 根据用户请求分析/生成信息 → 分析历史数据和趋势

最终：Autonomous + Prompt-and-response

Box 1: Autonomous

Orchestrate the sales order fulfillment and shipping of goods to customers.

An Autonomous Agent is the best choice for orchestrating sales order fulfillment and shipping in a consumer goods (CPG) environment.

While prompt-and-response agents are good for simple queries and task-based agents are good for linear workflows, an autonomous agent can manage the "many-to-many" variables inherent in supply chains without constant human intervention.

Why Autonomous is the Best Fit An autonomous agent acts as a reasoning engine rather than a static script. In fulfillment, it excels because it can:

Handle Exceptions: If a warehouse is out of a specific SKU, the agent can independently look for the next closest facility or suggest a substitute product.

Multi-Step Reasoning: It can check credit limits, verify inventory, calculate shipping dates, and trigger a logistics pickup in one cohesive flow.

Proactive Problem Solving: If a weather delay affects a shipping route, the agent can reroute the order and notify the customer before a human even spots the issue.

Box 2: Autonomous

Analyze historical data and trends to replenish stock.

For a consumer goods company managing stock replenishment, an Autonomous Agent is the best fit.

Stock replenishment requires a system that proactively monitors inventory, analyzes trends in real-time, and takes action (like drafting purchase orders) without waiting for a human to ask. While "Task" and "Prompt-and-response" agents have roles in analysis, they lack the "always-on" orchestration needed for a supply chain.

Why an Autonomous Agent Wins Proactive Replenishment Unlike prompt-based systems, an autonomous agent in Microsoft Copilot Studio can be triggered by business events—such as a low-stock alert in your ERP—rather than a user's prompt. It can:

Continuously Monitor: Watch inventory levels across multiple warehouses 24/7.

Auto-Draft Orders: Apply vendor rules and lead times to create draft purchase orders for approval.

Sense Disruptions: Adjust supply chain settings before a problem escalates based on real-time data.

QUESTION 23

A company has an AI agent that automates the review of customer feedback stored in a cloud database.

You plan to generate monthly reports from the agent's output to provide insights into customer sentiment and guide product development and marketing.

You need to ensure that the data ingested by the agent is clean and suitable for the intended use.

What should you do to prepare the data?

A. Create a workflow in Microsoft Power Automate.

B. Identify and address biased data.

C. Create an agent flow in Microsoft Copilot Studio.

D. Sort the database by customer last name.

Correct Answer: B

✅ Correct Answer: B. Identify and address biased data.

The key phrase is:

“ensure that the data ingested by the agent is clean and suitable for the intended use.”

This is primarily a data quality and responsible AI requirement.

Why B is correct

Before using customer feedback to generate sentiment reports and influence product development and marketing, you need to examine the data for bias and address it.

For example:

Certain customer groups may be overrepresented.
Feedback may disproportionately come from dissatisfied customers.
Some languages, regions, or demographics may be underrepresented.
Historical data may contain systematic biases.

If biased data is fed into the AI agent, the resulting sentiment analysis and business recommendations can also be biased.

Therefore:

Identify → assess → mitigate biased data → use the cleaned dataset for AI processing.

Why the others are incorrect

Option	Why incorrect
A. Create a workflow in Power Automate	Power Automate is for workflow/process automation. It doesn't inherently ensure data quality or remove bias.
B. Identify and address biased data	✅ Directly addresses data quality, fairness, and suitability for AI use.
C. Create an agent flow in Copilot Studio	Agent flows automate tasks within an agent; they don't solve the underlying data-quality issue.
D. Sort by customer last name	Sorting changes the order of records, not their quality, accuracy, or suitability.

🎯 Exam takeaway

For Azure AI / responsible AI questions, watch for wording such as:

clean and suitable data
data quality
representative data
fairness
bias
underrepresented groups

These often point toward:

Identify and address biased data.

Answer: B.

QUESTION 24 💩💩💩

A company is designing a Microsoft Power Platform solution to reduce the manual steps of a business process by deploying an existing AI model.

You need to calculate the return on AI investment (ROAI) by identifying the metadata and telemetry of the solution.

What should you use?

A. Microsoft Power Platform admin center

B. Success by Design

C. the Business value toolkit

D. Microsoft Cloud Adoption Framework for Azure

正确答案：✅ C. the Business value toolkit

calculate ROAI

business process

metadata and telemetry

business value

Business value toolkit 用于评估 AI/Power Platform 方案的业务价值，包括 ROI/ROAI、成本、效率提升以及 telemetry 等数据。

其他选项：

❌ Power Platform admin center → 环境、治理、容量、管理。
❌ Success by Design → Dynamics 365 项目实施方法论。
❌ Cloud Adoption Framework → Azure 云迁移和 adoption 方法论。

🎯 记忆

ROAI / Business value → Business value toolkit

答案：C

Correct Answer: C

Explanation/Reference:

Explanation:

The Business Value Toolkit is the correct resource to use for calculating Return on AI Investment (ROAI) for a Microsoft Power Platform solution.

Calculating ROAI with the Business Value Toolkit The Business Value Toolkit helps organizations move beyond simple automation by providing structured templates and analytics to justify AI investments.

Metadata Identification: It captures environment-specific data, such as the number of automated flows, the type of AI models used (e.g., AI Builder), and the business units involved.

Telemetry Integration: It leverages usage data from the Power Platform admin center and Application Insights to track real-time performance, such as execution frequency and success rates.

ROI Metrics: It transforms technical telemetry into financial outcomes, such as:

Time saved: Hours recovered from manual data entry or processing.
Error reduction: Improvements in accuracy compared to manual steps.
Cost avoidance: Savings from reduced reliance on specialized manual labor.

Incorrect:

[Not D] While the Cloud Adoption Framework (CAF) for Azure provides broad strategic guidance for AI adoption, the Business Value Toolkit is specifically designed to quantify the impact of low-code and AI solutions using solution metadata and telemetry.

QUESTION 25 💩💩💩

A company has a Microsoft Power Platform environment.

You need to build two agents named Agent1 and Agent2. The solution must meet the following requirements:

Agent1 must be extendable by using the Semantic Kernel and must connect to multiple business apps and APIs.
Agent2 must connect directly to data stored in Microsoft Dataverse and must be embeddable in a Microsoft Power Apps canvas app.

What should you use to build each agent? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Agent1:

Microsoft Foundry
Azure Logic Apps
Copilot in Power Apps
Microsoft Copilot Studio

Agent2:

Microsoft Foundry
Azure Logic Apps
Copilot in Power Apps
Microsoft Copilot Studio

QUESTION 答案 (Answer Area)

Agent1: 正确选项： Microsoft Foundry
Agent2: 正确选项： Microsoft Copilot Studio

核心考点速记（备考复习）

Agent1 (Microsoft Foundry)：
- 关键特征：使用 Semantic Kernel SDK（专业代码 Pro-code 扩展）构建，需要接入多个复杂的企业 API 与业务应用。
- 平台选择： Microsoft Foundry（Azure AI Foundry）是专业开发者使用 Semantic Kernel 框架构建和托管高级 Agent 的官方平台。
Agent2 (Microsoft Copilot Studio)：
- 关键特征：接入 Microsoft Dataverse 数据，且需要嵌入到 Power Apps 画布应用（Canvas App）中。
- 平台选择： Microsoft Copilot Studio 是 Power Platform 生态下的低代码 Agent 构建工具，原生支持 Dataverse 知识库与 Canvas App 内置 Copilot 控件的嵌入。

QUESTION 26 💩💩💩

A company has an Azure environment that supports multiple business units.

The company plans to implement an AI solution that will perform sentiment analysis on customer product reviews.

You need to evaluate the potential cost of the solution to support return on AI investment (ROAI) analysis.

What should you use?

A. Cost Management + Billing

B. Microsoft Fabric SKU Estimator

C. Total Cost of Ownership (TCO) Calculator

D. Azure Reservations

正确答案：✅ A. Cost Management + Billing

已有 Azure environment
要实施 AI solution
需要评估实际 Azure 成本
用于 ROAI（Return on AI Investment）分析

Cost Management + Billing 可以分析 Azure 资源的实际/预计成本，并按订阅、资源、业务单位等维度进行成本分析。

其他选项

❌ B. Microsoft Fabric SKU Estimator → 用于 Fabric 容量/成本估算。
❌ C. TCO Calculator → 主要用于比较 on-premises vs. Azure 的总体成本，不适合已经存在的 Azure AI 方案成本分析。
❌ D. Azure Reservations → 用于降低长期资源使用成本，不是 ROAI 成本分析工具。

🎯 答案

A. Cost Management + Billing

QUESTION 27

You need to recommend a Microsoft Power Platform solution for customer support.

The solution must include AI capabilities in Microsoft Power Automate and must meet the following requirements:

Use a centralized workspace for AI models.
Generate short overviews from large amounts of unstructured text, such as case notes or transcripts, without requiring additional training or coding.

What should you include in the recommendation for each requirement? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Answer Area

Use a centralized workspace:

An Microsoft Foundry hub
Azure OpenAI Foundry
Microsoft Copilot Studio
Microsoft Dataverse

Generate short overviews:

An AI Builder prebuilt model
An AI Builder prebuilt prompt
Azure OpenAI
GitHub Copilot
Microsoft Copilot in Power Automate

答案：

Requirement	Correct Answer	原因
Use a centralized workspace	✅ A Microsoft Foundry hub	Microsoft Foundry Hub 提供集中管理 AI 模型、项目和资源的工作区。
Generate short overviews	✅ An AI Builder prebuilt prompt	AI Builder 的 Prebuilt Prompt（如 Summarize Text）可直接对大量非结构化文本（案例记录、聊天记录等）生成摘要，无需训练或编写代码。

为什么其他选项不对？

❌ Azure OpenAI Foundry：不是用于集中管理 Power Platform AI 工作区。
❌ Microsoft Copilot Studio：用于构建 AI Agent，不是 AI 模型管理中心。
❌ Microsoft Dataverse：数据存储平台，不是 AI 工作区。
❌ AI Builder prebuilt model：主要用于文档处理、预测、分类等模型，不是文本摘要。
❌ Azure OpenAI：可以实现摘要，但通常需要自己设计 Prompt 或开发，不符合无需训练或编码。
❌ GitHub Copilot：开发辅助工具。
❌ Microsoft Copilot in Power Automate：帮助创建 Flow，不是用于文本摘要能力。

考试记忆

集中管理 AI 资源 → ✅ Microsoft Foundry Hub
无需训练、直接文本摘要 → ✅ AI Builder Prebuilt Prompt

Box 1: Azure AI Foundry

Box 2: An AI Builder prebuilt model

Box 1: Azure AI Foundry Use a centralized workspace for AI models.

Integrating Azure AI Foundry with Microsoft Power Automate provides a powerful, centralized environment for managing the AI models that drive your customer support project. This combination allows you to leverage enterprise-grade generative AI while maintaining governance through a unified hub.

Core AI Capabilities in Power Automate Power Automate can use AI Foundry models to transform standard support flows into intelligent, context- aware processes.

Intelligent Email Responders: Automatically analyze incoming support emails to determine sentiment, urgency, and intent.

Ticket Summarization: Use GPT models to condense long customer conversation histories for human agents.

Contextual Routing: Extract key entities (like product names or order IDs) to route tickets to the correct specialized team.

Multi-Agent Orchestration: Coordinate complex support tasks—like checking a database, drafting a reply, and updating a CRM—using centralized AI logic.

Incorrect:

Microsoft Foundry Hub While Microsoft Foundry Hubs (part of Azure AI Foundry) are used for large-scale, professional-grade AI development, AI Builder is the specific low-code tool designed to integrate seamlessly into Microsoft Power Automate.

Box 2: An AI Builder prebuilt model

Generate short overviews from large amounts of unstructured text, such as case notes or transcripts, without requiring additional training or coding.

To generate short overviews from large amounts of unstructured text in Power Automate without additional training, you should use the AI Builder "Summarize text" prebuilt model.

Key Features of the Summarize Model Zero Training Required: Works immediately out of the box with Microsoft’s pre-trained logic.

Unstructured Data: Handles case notes, long emails, and meeting transcripts effectively.

Low Code: Can be added to a Power Automate flow as a single "action" step.

Customizable Length: Allows you to specify if you want a brief summary or a more detailed overview.

QUESTION 28 💩💩💩

A company uses Microsoft Dynamics 365 Finance for accounts payable and customer debt recovery.

You are designing an AI finance process that meets the following requirements:

Provides AI-driven details to help staff identify overdue vendor invoices and outstanding balances
Helps staff reduce how long it takes to review overdue invoices and payment history

You need to recommend which Microsoft Copilot features to include in the design.

What should you recommend for each requirement? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Answer Area

Help identify vendor overdue invoices and outstanding balances:

Agent management
AI Summaries with Copilot
The Account Reconciliation agent
The Supplier Communication agent

Reduce how long it takes to review overdue invoices and payment history:

Analyze demand plans with Copilot
Collections coordinator summary
The Account Reconciliation agent
The Supplier Communication agent

✅ 答案

Help identify vendor overdue invoices and outstanding balances: → AI Summaries with Copilot
Reduce how long it takes to review overdue invoices and payment history: → Collections coordinator summary

🧠 记忆：

Vendor invoices / balances → AI Summaries

Collections / payment history → Collections coordinator summary

Help identify vendor overdue invoices and outstanding balances:
正确选项： AI Summaries with Copilot
Reduce how long it takes to review overdue invoices and payment history:
正确选项： Collections coordinator summary

核心考点速记（备考复习）

AI Summaries with Copilot（用于应付账款 AP）：在 Dynamics 365 Finance 的发票自动化与供应商工作区中，Copilot 生成的 AI 文本摘要可帮助财务人员快速提取并识别供应商逾期发票和未结余额（Vendor overdue invoices and balances）。
Collections coordinator summary（用于催收与债务回收）：在 Dynamics 365 Finance 的信用与催收模块中，专门用于自动汇总客户的逾期发票、付款历史及信用记录，显著缩短催收人员审查逾期发票和付款历史的时间。

Box 1: The Account Reconciliation Agent

Provides AI-driven details to help staff identify overdue vendor invoices and outstanding balances In Microsoft Dynamics 365 Finance, the Account Reconciliation agent and Copilot for Finance provide a powerful AI-driven ecosystem to manage overdue vendor invoices and customer debt recovery. These agents shift your team from reactive manual checking to proactive, exception-based management.

The Account Reconciliation Agent This agent automates the high-volume task of matching subledger transactions to the general ledger, specifically for Accounts Payable (AP) and Accounts Receivable (AR).

Automated Matching: Continuously monitors and reconciles transactions throughout the month rather than waiting for period-end.

Exception Identification: Automatically flags discrepancies, such as voucher amount mismatches or unmatched entries, for staff review.

AI-Suggested Actions: Using Microsoft Copilot, the agent suggests the most likely resolution for each exception, such as posting a correction or reaching out to a vendor.

Efficiency: Reduces the time spent on manual spreadsheet-based reconciliations, allowing staff to focus on resolving the "why" behind overdue balances

Box 2: Collections coordinator summary

Helps staff reduce how long it takes to review overdue invoices and payment history

To reduce the time spent reviewing overdue invoices and payment history in Microsoft Dynamics 365 Finance, you need the Collections Coordinator Summary feature powered by Copilot.

This AI-driven process automates the synthesis of customer financial data, allowing staff to move directly from review to action without manual data digging.

QUESTION 29 - Duplicated

A company has an AI agent that automates the review of customer feedback stored in a cloud database.

You plan to generate monthly reports from the agent's output to provide insights into customer sentiment and guide product development and marketing.

You need to ensure that the data ingested by the agent is clean and suitable for the intended use.

What should you do to prepare the data?

A. Ensure that the size of the database does not exceed 100 GB.

B. Translate the data into a single language.

C. Identify and address biased data.

D. Sort the database by customer last name.

正确答案：✅ C. Identify and address biased data.

原因

题目要求确保 AI ingest 的数据：

clean and suitable for the intended use

而这个 Agent 要分析 customer sentiment，如果历史反馈数据存在偏差（例如某类客户、地区、产品被过度代表），会直接导致 sentiment analysis 和后续报告产生偏差。

所以需要在数据准备阶段：

Identify and address biased data

其他选项

❌ A. Database ≤ 100 GB → 数据库大小与数据质量无关。
❌ B. Translate into one language → 多语言数据不一定需要翻译；重点是数据质量和偏差。
❌ D. Sort by customer last name → 排序不会改善 AI 数据质量。

🎯 考试记忆

AI data preparation → identify/address bias, missing data, duplicates, inconsistencies

答案：C

QUESTION 30

A company has an Azure environment that supports multiple business units.

The company plans to implement an AI solution that will perform sentiment analysis on customer product reviews.

You need to evaluate the potential cost of the solution to support return on AI investment (ROAI) analysis.

What should you use?

A. Total Cost of Ownership (TCO) Calculator

B. Azure Reservations

C. Azure pricing calculator

D. Azure Monitor

正确答案：

C. Azure pricing calculator（Azure 价格计算器）

解析与考点分析

为什么选 C？

业务场景：公司计划（Plans）实现一个新的 AI 解决方案（对客户产品评论进行情感分析），需要在部署前评估该方案在 Azur中的预估/潜在成本（Potential cost），用于 AI 投资回报率（ROAI）分析。
工具区别与选择：
- Azure pricing calculator（Azure 价格计算器）：用于配置并计算全新规划的 Azure 服务（如 Azure AI Language 情感分析 API、Azure OpenAI 等）根据预计用量（如每月调用的 API 次数、Token 数量）所产生的未来预估成本。这是为新 AI 解决方案计算潜在支出并带入 ROAI 财务模型的标准工具。

为什么不选其他选项？

A. Total Cost of Ownership (TCO) Calculator（总体拥有成本计算器）： TCO 计算器用于对比“将现有的本地数据中心（On-premises）基础架构迁移到 Azure”的长期成本差异（对比硬件、机房电力、IT 运维成本等），不适用于单纯估算某个全新 Azure AI 服务的用量成本。
B. Azure Reservations（Azure 预留）： 是一种承诺使用 1 年或 3 年以获取折扣的计费优惠模式，并非成本估算与 ROAI 评估工具。
D. Azure Monitor（Azure 监控）：用于监控已部署资源的性能和日志，无法对未部署的全新 AI 方案进行事前成本预估。

QUESTION 31 - Duplicated

A company has an Azure environment that supports multiple business units.

The company plans to implement an AI solution that will perform sentiment analysis on customer product reviews.

You need to evaluate the potential cost of the solution to support return on AI investment (ROAI) analysis.

What should you use?

A. Azure savings plans

B. Microsoft Fabric SKU Estimator

C. Cost Management + Billing

D. Azure Monitor

正确答案：✅ C. Cost Management + Billing

原因

题目关键词是 evaluate potential cost + ROAI analysis，需要从 Azure 环境的成本角度进行分析。

C. Cost Management + Billing ✅ 用于分析 Azure 资源成本、按订阅/资源等维度查看成本，并支持成本预测和 ROI 分析。

其他选项：

❌ A. Azure savings plans → 用于降低长期计算成本，不是成本分析工具。
❌ B. Microsoft Fabric SKU Estimator → 针对 Fabric 容量成本估算。
❌ D. Azure Monitor → 监控性能和资源指标，不是成本分析工具。

答案：C. Cost Management + Billing

Incorrect:

Azure Monitor
Azure Reservations
Azure savings plans

Is used to reduce costs for compute services via long-term commitment.

It is a way to lower your bill, but not a tool for evaluating or reporting on investment performance

Microsoft Fabric SKU Estimator

Is used to estimate capacity needs for Fabric-specific data projects. It is eseful only if your sentiment analysis solution is built specifically on the Microsoft Fabric platform

QUESTION 32 - Duplicated

A company has an Azure environment that supports multiple business units.

The company plans to implement an AI solution that will perform sentiment analysis on customer product reviews.

You need to evaluate the potential cost of the solution to support return on AI investment (ROAI) analysis.

What should you use?

A. Azure Reservations

B. Microsoft Fabric SKU Estimator

C. Anomaly Detection in Azure Cost Management

D. Azure pricing calculator

D. Azure pricing calculator（Azure 价格计算器）

核心考点速记（备考复习）

Azure 价格计算器（Azure pricing calculator）：专门用于在项目部署前，根据预计的服务用量（如情感分析 API 的每月调用次数）估算全新的 Azure 云服务（如 Azure AI Services）的潜在成本，从而为 ROAI（AI 投资回报率）分析提供精确的支出数据。
区别辨析：
- TCO Calculator：用于“本地数据中心迁移到云端”的总体成本对比。
- Cost Management：用于监控“已部署资源”的实际历史账单与异常费用。

Correct Answer: D

✅ 答案：D. Azure pricing calculator

原因：题目是要在 AI 方案实施前，估算未来成本，用于 ROAI 分析。

A. Azure Reservations ❌ → 购买长期资源使用折扣
B. Microsoft Fabric SKU Estimator ❌ → 估算 Fabric 容量成本
C. Cost Management Anomaly Detection ❌ → 检测已经产生的异常费用
D. Azure pricing calculator ✅ → 根据预计用量估算 Azure 服务成本

🧠 记忆：

未来成本估算 → Pricing Calculator 已产生费用监控 → Cost Management 本地 vs Azure 总成本 → TCO Calculator

答案：D ✅

Design AI-powered business solutions

Testlet 1

Overview Contoso, Ltd. is a high-tech manufacturing company that uses Microsoft Dynamics 365 Finance. Dynamics 365 Supply Chain Management, and Dynamics 365 Commerce for its North American operations. The company designs and develops innovative products that have many patents and proprietary technologies. The patents and engineering designs are closely guarded secrets.

Contoso executives want to integrate and adopt AI solutions to help scale the company in preparation for an anticipated period of rapid growth.

The company has multiple legal entities and Azure subscriptions that will be used in the adopted AI solutions.

Requirements AI Adoption The following executives will have specific responsibilities in the overall AI adoption:

Chief Technology Officer (CTO): Select one Dynamics 365 Finance, Dynamics 365 Supply Chain Management or Dynamics 365 Commerce prebuilt AI agent and one custom Microsoft Copilot Studio AI agent to prioritize and deploy during the initial AI adoption phase.
Chief Information Officer (CIO): Ensure that appropriate security labels are assigned to the data used by the AI agents.
Chief Financial Officer (CFO): Analyze the return on investment (ROI) for the AI agents being deployed. Chief Information Security Officer (CISO): Discover and inventory AI resources for auditing.
Chief Executive Officer (CEO): Ensure that all solutions adhere to industry-standard responsible AI practices.

All AI initiatives and agents will have a detailed business use case, a defined audience profile, and an estimated ROI that will compare the cost savings of the current process against the estimated costs of using the new AI solutions.

The company's research and development (R&D) department already has a custom Model Context Protocol (MCP) server that contains comprehensive product specifications and compliance data.

Prebuilt AI Agent

The CTO has NOT yet selected which prebuilt AI agent to use in Dynamics 365 Supply Chain Management. The CTO wants to view available agent templates to identify which agent will add the most business value.

Depending on which high-priority AI agents are identified, its agent capabilities must be previewed in a discovery meeting with the relevant business operation stakeholders.

Custom AI Agent

Contoso has identified the following custom AI agent requirements:

The custom AI agent will use data from Dynamics 365 Supply Chain Management to answer questions for the manufacturing team as a low-code solution.
The custom AI agent will be accessible from within Microsoft Teams.
The custom AI agent must be designed to eventually connect to other agents that can be selected based on their description.
The topics used in the custom AI agent will be selected based NOT on a trigger phrase, but on a description of the purpose of the query, to make the interactions more conversational.
The custom AI agent must be able to answer questions about product specifications by using existing technologies. The product specifications are maintained by the R&D department.
The custom AI agent must be integrated with and accessible from Dynamics 365 Supply Chain Management.
The custom AI agent must be able to use Dynamics 365 Supply Chain Management business logic that is stored outside of the application.

Analysis, Reporting, and Troubleshooting

Analysis, Reporting, and Troubleshooting

Contoso has identified the following analysis, reporting, and troubleshooting requirements:

The CISO will audit all the AI solutions monthly for compliance and security.
The CFO will analyze all the AI solutions quarterly to compare the estimated ROI against actual measured efficiencies and adoption. The CFO will use the Copilot Studio agent usage estimator to perform this analysis.
The CISO wants to identify how much sensitive data was accessed for a given AI agent run and who accessed the data. Too much sensitive data accessed by a single user might indicate a high security risk.
The CTO wants to track user feedback on the quality of the AI agent responses during user interactions with the agents. Consistently poor feedback will trigger an escalated reengineering discussion.
The CEO wants a quarterly assessment of all the required metrics for their specific responsibilities. The tools used for the assessments must be Microsoft-recommended and must verify reliability, interpretability, fairness, and compliance.
The CFO wants to identify how many interactions with the AI agents are abandoned on a given day as compared to resolved conversations. Too many abandoned sessions might indicate that Copilot Studio credits are being used inefficiently by end users.

QUESTION 1 💩💩💩

What should you recommend to assist the CTO with the prebuilt agent selection process?

A. Agent management

B. Copilot Studio

C. Lifecycle Services (LCS)

D. Immersive Home

正确答案： A. Agent management（Agent 管理） ✅

解析与考点分析

为什么选 A？

对照案例需求（Contoso, Ltd. 案例）：
- 案例中明确提到 CTO 的职责与痛点： "The CTO has NOT yet selected which prebuilt AI agent to use in Dynamics 365 Supply Chain Management. The CTO wants to view available agent templates to identify which agent will add the most business value." （CTO 尚未选择在 Dynamics 365 Supply Chain Management 中使用哪种预置 AI Agent。CTO 希望查看可用的 Agent 模板，以确定哪个 Agent 能增加最大的业务价值。）
微软产品功能定位：
- 在 Dynamics 365 Finance and Operations（包括 Supply Chain Management 供应链管理）应用体系中，微软引入了 Agent management（Agent 管理）工作区/中心。
- Agent management 是专门供 CTO、系统管理员和业务分析师浏览、评估、预览和启用 Dynamics 365 中开箱即用的预置 AI Agent 模板（Prebuilt Agent Templates）（如 Supplier Communication Agent、Demand Analysis Agent 等）的集中化管理平台。

为什么不选其他选项？

B. Copilot Studio： Copilot Studio 主要用于从零构建或扩展自定义 AI Agent（Custom AI Agent），而不是在 Dynamics 365 内部浏览和选择开箱即用的预置 Agent 模板（案例中 CTO 的另一项任务“自定义 Agent”才使用 Copilot Studio）。
C. Lifecycle Services (LCS)：是 Dynamics 365 F&O 旧版的环境运维与应用生命周期管理门户，不用于浏览和体验具体的业务 AI Agent 模板。
D. Immersive Home：属于 Power BI / 应用主页集成展示界面，与 Dynamics 365 的预置 Agent 模板选型和预览管理无关。

Correct Answer: A Section: (none)

Explanation/Reference:

Explanation:

Agent management is the correct tool to use for viewing available agent templates and deciding which AI agent to use within Dynamics 365 Supply Chain Management.

Available Agent Discovery Tools When working within Dynamics 365 Finance and Operations apps (including Supply Chain Management), the specifically designed interface for these tasks is Agent management.

Agent Management: A dedicated preview feature in Finance and Operations apps that allows users to discover, configure, and manage autonomous AI-powered agents.

Incorrect:

[Not B] Copilot Studio: While this is the platform used to build and customize agents and track detailed analytics (like thumbs-up/down feedback), it is a standalone low-code tool rather than the primary discovery pane inside the SCM interface.

[Not C] Lifecycle Services (LCS): Primarily used for environment management, deployments, and updates rather than the day-to-day discovery of specific AI agent templates.

[Not D] Immersive Home: Typically refers to a broader Microsoft 365 or entry-level experience, not the specialized ERP agent management portal.

QUESTION 2 💩💩💩

What should you include in the custom AI agent design to meet the R&D product specifications and the compliance information requirements? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Answer Area

To expose the data to the agent, create:

an Azure AI Bot Service channel
a custom connector
a custom OData entity
the Semantic Kernel

Add to the agent:

an event trigger
the MCP server
a REST API
a tool

答案：

To expose the data to the agent, create: ✅ a custom connector
Add to the agent: ✅ the MCP server

原因

题目明确说 R&D 已经有一个 custom MCP server，里面包含：

Product specifications
Compliance information

因此 Agent 应该直接使用现有的 MCP server。

而要让企业数据/API 暴露给 Copilot Studio Agent，通常使用 Custom Connector。

最终答案：

Create → Custom connector
Add → MCP server

Testlet 2

Background

Fabrikam, Inc., is a global consumer goods company that is undergoing a digital transformation initiative to migrate its entire infrastructure to the Microsoft cloud. As a key element of this cloud migration, the company will implement Microsoft Dynamics 365 Sales, moving away from the current on-premises proprietary technologies used by its business-to-business (B2B) sales team.

As part of the cloud migration, Fabrikam will adopt an AI-first approach to its business solutions and implement AI solutions, wherever possible, to streamline operations.

Problem Statements

Fabrikam's infrastructure currently relies on various on-premises systems that require sales executives to use corporate computers with physical keyboards to access business information during customer interactions. Mobile phones cannot be used for these purposes, as the systems depend on keyboard input. As a result, the sales executives spend a lot of time using keyboards to search for data on several disparate systems and file servers, rather than focusing on the customers. This affects the customer experience.

Fabrikam stakeholders are concerned that users will be hesitant to adopt AI. If the AI initiatives are NOT adopted, cost savings will never be realized. Additionally, funding for future AI initiatives will depend on demonstrating an increase in AI adoption month over month. As the AI agent initiative for the sales team will be the first for Fabrikam, the rapid adoption of the agent is a high priority.

Planned Initiatives

General Fabrikam management has prioritized AI-driven projects to improve efficiency, customer engagement, and responsible AI adoption. The current application infrastructure is on-premises and must be migrated to the cloud to support the adoption of these technologies.

Infrastructure Migration

Fabrikam plans to migrate from its current on-premises infrastructure to a completely cloud-based topology; this will include user authentication, the security framework, and, primarily, the adoption of the services by end users.

All the data from the different systems will be consolidated into a single data source - a common data model that will use a Microsoft Dataverse environment as a single source of truth (SSOT) for the sales team.

Sales Cycle Enablement

To achieve the company's objectives, Fabrikam intends to implement the following strategies to enhance the sales cycle:
Use low-code development to create a single AI agent that has Dataverse as its core component. Ensure that sales managers can access unanswered correspondence from prospects and intervene as appropriate.
Replace the previous proprietary software with Dynamics 365 Sales to track sales cycles and customer interactions.
Have the sales executives use Dynamics 365 Sales to track interactions for open opportunities and send follow-up communications to prospects.
Have the sales executives use handsfree headsets to interact with an AI agent when they have questions about internal policies or customer data.

Requirements

Infrastructure Migration

Fabrikam has identified the following infrastructure migration requirements:

Azure must be used for all future infrastructure workloads.

The company must follow Microsoft-recommended methodologies for infrastructure migration to the cloud.

Any created AI agents must have their return on investment (ROI) calculated to ensure that the solution will save the company money.

Sales Cycle Enablement

Fabrikam has identified the following requirements for sales cycle enablement:
The final AI agent must follow Microsoft recommendations for a conversational user experience.
A designated checklist must be reviewed to ensure that the AI agent follows Microsoft deployment recommendations for a compliant solution.
Detailed telemetry must be logged for the first created AI agent to help troubleshoot and optimize the agent during the initial AI agent adoption process.
Unexpected AI agent actions must end in an escalation to a live representative. For example, a sales executive must be rerouted to a representative if the agent cannot answer a question after two failed attempts.
The return on investment (ROI) of switching from the current process to the future process is required for stakeholder sign off.
The sales team must use Dynamics 365 Sales to correspond with prospects more quickly and efficiently than currently.
Sales managers must report on the adoption of the AI agent to key Fabrikam stakeholders on a monthly basis.
Any sensitive information, such as user IDs and names, shared via the AI agent must be tracked for future auditing.

QUESTION 1 💩💩💩

Which template should you use for the AI agent to meet the requirements for the sales executives?

A. IT Helpdesk in Microsoft Copilot Studio

B. AI agents in Microsoft Foundry

C. Voice in Microsoft Copilot Studio

D. AI chat in Microsoft Foundry

正确答案：✅ C. Voice in Microsoft Copilot Studio

原因

题目最关键的需求是：

sales executives use handsfree headsets to interact with an AI agent

也就是需要 Voice 交互。

Voice in Microsoft Copilot Studio 专门用于构建支持语音交互的 Copilot/Agent，符合销售人员使用免提耳机进行查询的场景。

其他选项

❌ A. IT Helpdesk → 面向 IT 支持场景。
❌ B. AI agents in Microsoft Foundry → 用于开发更复杂的 AI Agent，但不是本题要求的低代码语音模板。
❌ D. AI chat in Microsoft Foundry → 面向文本聊天，不符合 hands-free voice requirement。

🎯 最终答案

C. Voice in Microsoft Copilot Studio

Correct Answer: C

Sales Cycle Enablement To achieve the company's objectives, Fabrikam intends to implement the following strategies to enhance the sales cycle:

Have the sales executives use Dynamics 365 Sales to track interactions for open opportunities and send follow-up communications to prospects.

*-> Have the sales executives use handsfree headsets to interact with an AI agent when they have questions about internal policies or customer data.

To enable handsfree headset interaction with an AI agent for internal policies and customer data in Dynamics 365 Sales, you should use Microsoft Copilot Studio to create a custom agent template.

Here is the breakdown of the recommended approach and templates:

Recommended Template Voice-enabled agent template (in Copilot Studio): This template provides the foundational, pre-configured setup for Interactive Voice Response (IVR) capabilities, allowing for natural language voice inputs and text-to-speech output.

Internal Data Knowledge Source: Within this agent, you will connect to Dataverse (for customer data) and configure Knowledge Sources (for internal policy documents).

QUESTION 2 💩💩💩

Which tool should you use for the prospect communication requirements in Dynamics 365 Sales?

A. Azure AI Search

B. Copilot email assist

C. the Voice template Microsoft Copilot Studio

D. Deep Research in Microsoft Foundry Agent Service

✅ 答案：B. Copilot email assist

原因：

题目要求的是：

Dynamics 365 Sales 中与 prospects 更快速、更高效地进行 correspondence

The sales team must use Dynamics 365 Sales to correspond with prospects more quickly and efficiently than currently.

Copilot email assist 专门帮助销售人员生成和处理销售邮件，例如：

生成邮件内容
根据上下文提供回复建议
加快与 prospects 的沟通

其他选项：

A. Azure AI Search ❌ → 搜索和 grounding
B. Copilot email assist ✅ → Dynamics 365 Sales 邮件沟通
C. Voice template ❌ → 语音交互，不是 prospect email communication
D. Deep Research ❌ → 深度研究，不是销售邮件效率工具

🧠 记忆：

Dynamics 365 Sales + prospect correspondence → Copilot email assist

答案：B ✅

Correct Answer: B Section: (none)

In Dynamics 365 Sales, the Copilot email assist feature acts as a powerful accelerator for handling prospects by automating the most time-consuming parts of communication.

Here is how it helps you move faster and more efficiently:

Instant Drafting: You can generate professional-looking email drafts in seconds by choosing a predefined category (like "reply to an inquiry" or "follow up") or by entering your own custom prompt.

Context-Aware Personalization: The AI uses data directly from your CRM—such as past interactions, deal status, and customer notes—to ensure every message is tailored specifically to that prospect's needs. Tone & Style Adjustments: You can quickly refine the length and tone (e.g., formal, friendly, or urgent) of a draft to better resonate with a particular recipient.

Summarization: When dealing with long email chains, Copilot provides a concise summary of the conversation history, allowing you to catch up instantly without reading through every old message. Actionable Reminders: It monitors your inbox to identify pending action items or key customer requests you may have missed, ensuring no prospect falls through the cracks.

Seamless Integration: These tools are available directly within the Dynamics 365 Email Rich Text Editor and across Microsoft 365 apps like Outlook and Teams, keeping you in your flow of work.

QUESTION 3 💩💩

Which components should you use to meet the sales cycle enablement requirements? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Answer Area

For Al agent creation:

Microsoft Foundry
Dynamics 365 Sales
Microsoft Copilot Studio
the Power Platform admin center

For unexpected Al agent actions:

a custom connector
an event trigger
a Fallback topic
a REST API

答案：

For AI agent creation: ✅ Microsoft Copilot Studio
For unexpected AI agent actions: ✅ a Fallback topic

解释

1. AI agent creation → Copilot Studio

题目要求：

low-code
Dataverse 为核心
Agent 用于销售团队
conversational experience

这些都是 Microsoft Copilot Studio 的典型场景。

2. Unexpected AI agent actions → Fallback topic

题目明确要求 Agent 无法回答时进行处理，例如：

连续两次回答失败 → 转人工代表

Fallback topic 用于处理无法匹配/无法处理的用户请求，并可以设计后续的 escalation。

🎯 最终答案

Requirement	Answer
AI agent creation	Microsoft Copilot Studio
Unexpected AI agent actions	Fallback topic

Explanation/Reference:

Box 1: Microsoft Copilot Studio

Box 2: a Fallback topic

Box 1: Microsoft Copilot Studio For AI agent creation

Scenario:

A designated checklist must be reviewed to ensure that the AI agent follows Microsoft deployment recommendations for a compliant solution.

The Microsoft tool that should be used to create the AI agent is Microsoft Copilot Studio.

It is a low-code platform specifically designed for building and managing intelligent agents that integrate with Microsoft Dataverse as a core component for storing data and providing business context. To ensure the agent follows Microsoft's deployment recommendations, users should consult the implementation checklist provided within the Microsoft Copilot Studio documentation.

Key Features of Microsoft Copilot Studio:

Low-Code Interface: Uses a graphical builder or natural language to design agent behaviors and conversation flows.

Dataverse Integration: Leverages Dataverse as the enterprise data platform to ground agents in organizational knowledge and store shared context.

Governance and Compliance: Includes built-in responsible AI features and aligns with enterprise security standards.

Multi-Channel Deployment: Allows publishing agents across Microsoft Teams, websites, and mobile app

Box 2: a Fallback topic

For unexpected AI agent actions.

Scenario:

Unexpected AI agent actions must end in an escalation to a live representative. For example, a sales executive must be rerouted to a representative if the agent cannot answer a question after two failed attempts.

In a low-code environment using Microsoft Copilot Studio with Dataverse, you can ensure unexpected AI actions lead to a live representative by configuring specific system topics.

Configure the System Fallback Topic

The Fallback system topic triggers when the AI agent cannot match a user's intent to any existing topic or knowledge source with sufficient confidence.

Enable Fallback: In Copilot Studio, navigate to Settings > General Settings > System fallback and click Add.

Action: Edit the Fallback topic to include a Go to topic node that redirects to the Escalate system topic.

Implement the Escalate System Topic The Escalate topic is the primary mechanism for handoffs.
Handle Errors and Infinite Loops

Question Set 3 - Design AI-powered business solutions

QUESTION 1 💩💩💩

You need to design a Microsoft 365 Copilot solution to optimize employee productivity. The solution must meet the following requirements:

Ensure that the employees can query content stored in a subset of Microsoft SharePoint Online sites and in Teams by using natural language-based prompt actions.

Ensure that employees receive contextually relevant responses in Microsoft 365 Copilot.

What should you include in the design?

A. Build a Microsoft Power Automate desktop flow to read the SharePoint content and post the responses to Teams.

B. Modify SharePoint settings.

C. Create a custom REST API that crawls the SharePoint content.

D. Configure Microsoft Graph access

答案：✅ 4. Configure Microsoft Graph access

解析

题目要求：

使用自然语言查询 SharePoint Online 和 Teams 内容。
在 Microsoft 365 Copilot 中获得与上下文相关的回答。

Microsoft 365 Copilot 的知识来源就是 Microsoft Graph。

通过 Microsoft Graph，Copilot 可以安全地访问：

SharePoint Online
Microsoft Teams
Outlook
OneDrive
Microsoft 365 其他数据

并根据用户权限返回上下文相关的内容。

因此，应配置 Microsoft Graph access。

为什么其他选项不对？

❌ 1. Power Automate desktop flow: 用于 RPA 自动化，不是 Microsoft 365 Copilot 的知识来源。
❌ 2. Modify SharePoint settings: 修改 SharePoint 设置不能让 Copilot 具备自然语言检索能力。
❌ 3. Create a custom REST API Microsoft 365 Copilot 已通过 Microsoft Graph 访问 Microsoft 365 数据，无需自行开发爬虫 API。

考试记忆

Microsoft 365 Copilot 查询 M365 数据 → ✅ Microsoft Graph
Copilot Studio 查询企业外部数据 → Knowledge、Connector、MCP
Azure AI Search → 企业文档检索（RAG）

最终答案：✅ 4. Configure Microsoft Graph access

Correct Answer: D

Explanation/Reference:

Explanation:

To enable Microsoft 365 Copilot to query a specific subset of SharePoint Online and Teams content using natural language, you can implement a combination of Restricted SharePoint Search and Microsoft Graph Connectors.

Restricting Content Access You can limit the scope of data Copilot searches by using features that control which sites are indexed or accessible.
Configuring Microsoft Graph Access Microsoft Graph acts as the bridge connecting Copilot to your organizational data. To integrate specific sources.
Enabling Prompt Actions & Context Declarative Agents: You can create specialized Copilot Agents grounded in specific SharePoint knowledge sources. These agents use natural language instructions to focus on a subset of data for more contextually relevant responses.

Direct Referencing: Users can improve response relevance by explicitly naming files, folders, or Teams channels in their natural language prompts (e.g., "Summarize notes from the 'Product Launch' channel").

QUESTION 2 💩💩💩

A company uses Microsoft Dynamics 365 Finance to manage accounts payable.

You are designing an AI invoice processing solution.

You need to recommend the prerequisites to configure a prebuilt copilot for accounts payable.

What should you recommend?

A. From Microsoft Copilot Studio, create an accounts payable agent.

B. Extend Microsoft 365 Copilot for Sales to an accounts payable agent.

C. Build an AI tool in Microsoft Foundry.

D. From the Power Platform admin center, assign the Finance and Operations AI security role to users.

正确答案： ✅

D. From the Power Platform admin center, assign the Finance and Operations AI security role to users.（在 Power Platform 管理中心，将 Finance and Operations AI 安全角色分配给用户。）

为什么选 D？

业务场景：公司使用 Dynamics 365 Finance（财务模块）管理应付账款（Accounts Payable），并希望配置该模块中预置的 Copilot（Prebuilt Copilot）来处理发票。
前置条件与权限控制（Prerequisites）：
- Dynamics 365 Finance（属于 Finance & Operations 应用程序体系）内置了预置的生成式 AI / Copilot 功能（如发票自动化处理、应付账款匹配等）。
- 要让用户能够使用这些预置（Prebuilt）的 Copilot 功能，管理员必须在 Power Platform 管理中心或 Dynamics 365 安全配置中，为相关用户分配 Finance and Operations AI 安全角色（Finance and Operations AI security role）及对应的 AI 访问权限。

为什么不选其他选项？

A. From Microsoft Copilot Studio, create an accounts payable agent（从 Copilot Studio 创建应付账款 Agent）：这是从零开始开发自定义（Custom） AI Agent 的步骤，而题目明确要求配置现有的预置（Prebuilt） Copilot。
B. Extend Microsoft 365 Copilot for Sales to an accounts payable agent（将 Copilot for Sales 扩展为应付账款 Agent）： Copilot for Sales 是专门针对 CRM（销售与 Outlook）场景的，与 Dynamics 365 Finance 的应付账款（AP）无直接关系。
C. Build an AI tool in Microsoft Foundry（在 Microsoft Foundry 中构建 AI 工具）：Azure AI Foundry（前身为 Azure AI Studio）用于专业开发者构建自定义大模型和 RAG 应用，不属于 Dynamics 365 预置 Copilot 的低代码/无代码前置配置。

Explanation/Reference:

Users must be assigned the Finance and Operations AI and AIB Roles in the Dataverse environment.

Note:

To configure the prebuilt Copilot for Accounts Payable in Microsoft Dynamics 365 Finance, you must satisfy several foundational and technical requirements.

System & Environment Prerequisites
Technical Configuration & Data Settings
Licensing & Security Roles Licenses *-> Dataverse Roles: Users must be assigned the Finance and Operations AI and AIB Roles in the Dataverse environment.
Feature Activation

QUESTION 3 💩💩💩

A company plans to deploy a Microsoft Dynamics 365 Contact Center agent.

You need to ensure that the agent can transfer the conversation to a live customer service representative.

Which two components should you include in the solution? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. Microsoft Foundry

B. Microsoft Copilot Studio

C. Microsoft 365 Agents Toolkit

D. an Azure AI Bot Service skill

E. Customer engagement hub

正确答案：✅ B、E

B. Microsoft Copilot Studio → 创建和配置 Agent，包括 转人工（human handoff） 能力。
E. Customer engagement hub → 提供 Customer Service / Contact Center 的人工客服工作区，用于接管转来的会话。

考试记忆：

AI Agent → Copilot Studio → Human Agent → Customer engagement hub

所以答案：B + E。

QUESTION 4. 💩💩💩

A company uses Microsoft Dynamics 365 Supply Chain Management.

You are designing an AI supply chain process that meets the following requirements:

Provides managers with AI-driven insights that surface key information from customer orders
Helps planners use AI to anticipate future product needs more accurately
You need to recommend which Microsoft Copilot features to include in the design.

What should you recommend for each requirement? To answer, select the appropriate options in the answer area.

Answer Area

Provide AI-driven insights from customer orders:

Al Summaries with Copilot
Generative insights for Demand planning
The Customer credit and collections workspace
Workload insights with Copilot

Anticipate future product needs:

Generative insights for Demand planning
Microsoft Power BI
Product information management
The Supplier Communications Agent

答案：

Provide AI-driven insights from customer orders: ✅ AI Summaries with Copilot
Anticipate future product needs: ✅ Generative insights for Demand planning

记忆

Customer orders → AI Summaries with Copilot：从订单等业务数据中快速提取关键摘要和洞察。
Future product needs / Demand forecast → Generative insights for Demand planning：帮助规划人员分析需求、预测未来产品需求。

最终答案：

① AI Summaries with Copilot ② Generative insights for Demand planning

Box 1: AI summaries with Copilot

Box 2: Generative insights for Demand planning

Box 1: AI summaries with Copilot

Provides managers with AI-driven insights that surface key information from customer orders

To provide managers with AI-driven insights from customer orders in Dynamics 365 Supply Chain Management, use the AI summaries with Copilot feature, specifically leveraging the embedded Customer Summary capabilities. This tool provides instant, personalized overviews of critical data, such as backordered items, credit limits, and risk indicators, directly within the workspace.

Box 2: Generative insights for Demand planning

Helps planners use AI to anticipate future product needs more accurately

The Copilot Generative insights for Demand planning feature in Microsoft Dynamics 365 Supply Chain Management enables demand planners to move from reactive, manual forecasting to proactive, AI-driven demand anticipation. By analyzing, clustering, and interpreting vast datasets, Copilot helps identify patterns such as seasonality and signal correlations to improve accuracy.

QUESTION 5 💩💩💩

A company has a Microsoft 365 E5 subscription and uses Microsoft Copilot Studio.

The company has a Microsoft SharePoint Online library that contains 10,000 policy PDFs from various departments.

The library contains a populated column named Department for each PDF.

You need to design a Copilot Studio agent that will use the SharePoint library as a knowledge source. The solution must meet the following requirements:

Enable the agent to answer user questions about company policies.
Ensure that the agent can identify which departments and policies are connected.

What should you include in the design for each requirement? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Enable the agent to answer questions about company policies:

Build a custom model in Microsoft Foundry.
From Copilot Studio, add SharePoint as a knowledge source.
Import the PDFs into Microsoft Dataverse.
Use AI Builder to process and feed SharePoint content.

Identify which departments and policies are connected:

Apply Microsoft Purview sensitivity labels.
Create a Microsoft Dataverse table for the departments.
From Copilot Studio, configure the SharePoint tool.
Upgrade to SharePoint Premium.

这是该道热点选择题（Hot Area）的正确答案与详细解析：

QUESTION 答案 (Answer Area)

Enable the agent to answer questions about company policies:

✅ 正确选项： From Copilot Studio, add SharePoint as a knowledge source.（在 Copilot Studio 中，将 SharePoint 添加为知识源。）

Identify which departments and policies are connected:

✅ 正确选项： From Copilot Studio, configure the SharePoint tool.（在 Copilot Studio 中，配置 SharePoint 工具。）

解析与考点分析

第一部分分析：使 Agent 能够回答有关公司政策的问题

考点： Copilot Studio 原生知识源（Knowledge Sources）与 RAG 支持。
解析：
- Microsoft Copilot Studio 提供了开箱即用的知识源（Knowledge Sources）功能。开发者可以直接添加 SharePoint Online 文档库作为知识源。
- 系统会自动对文档库中的 10,000 份 PDF 政策文件建立索引，利用生成式回答（Generative Answers）回答用户关于公司政策的各种提问，无需编写代码，也不需要将 PDF 额外导入到 Dataverse 中或使用 AI Builder 预处理。

第二部分分析：识别哪些部门和政策相关联

考点： SharePoint 元数据（Metadata）与 Copilot Studio 工具/知识库配置。
解析：
- 背景信息：题目明确说明 SharePoint 文档库中已经包含一个填充好的列 Department（部门）。
- 解决方案：在 Copilot Studio 中配置 SharePoint 工具/知识库设置（configure the SharePoint tool），可以配置搜索查询参数和元数据列映射（Metadata Filtering）。这样 Agent 在检索 PDF 时能够读取并理解每个文件关联的 Department 列元数据，从而精准识别政策文件与具体部门之间的关联。

干扰项排除说明

Import the PDFs into Microsoft Dataverse / Build a custom model in Microsoft Foundry：增加了不必要的架构复杂度和开发/迁移成本， Copilot Studio 本身已原生支持直接接入 SharePoint。
Apply Microsoft Purview sensitivity labels：用于敏感数据的安全加密与信息保护，不用于关联和提取业务元数据（如部门列）。
Upgrade to SharePoint Premium： SharePoint Premium（前身为 Syntex）用于利用 AI 自动提取无结构文档中的元数据，但题目中明确提到 Department 列已经被填充（already populated），因此无需额外升级付费。

Box 1: From Copilot Studio, add SharePoint as a knowledge source

Enable the agent to answer user questions about company policies.

To enable a Microsoft Copilot Studio agent to answer questions from a large SharePoint Online library using the Department column as a filter, you must configure Generative Answers with specific metadata filtering.

*-> 1. Configure SharePoint as a Knowledge Source Add the Source: In Copilot Studio, go to the Knowledge tab and select Add knowledge. Choose SharePoint and provide the URL of your site or specific document library.

Authentication: Ensure Manual Authentication with Microsoft Entra ID is configured. The agent requires the Sites.Read.All and Files.Read.All scopes to access documents on behalf of the user.

Indexing: Note that indexing tens of thousands of files can take significant time (days or weeks).

Enable Metadata Filtering (Department Column) To ensure the agent only searches policies relevant to a user's department, you must use Advanced Settings to filter the search results based on your custom column

Box 2: From Copilot Studio, configure the SharePoint tool.

Ensure that the agent can identify which departments and policies are connected.

The SharePoint tool can be configured directly from Microsoft Copilot Studio to act as a knowledge source, allowing your AI agent to read files, documents, and lists from SharePoint to provide answers.

Note:

To ensure your Copilot Studio agent effectively identifies and connects your "Department" metadata with specific policies across tens of thousands of documents, you must address both search indexing and agent instructions.

To ensure that your Microsoft Copilot Studio agent can identify and filter documents based on the Department column from your SharePoint library, you should configure the Advanced settings of your SharePoint knowledge source to use KQL (Keyword Query Language) filters.

Incorrect:

Upgrade to SharePoint Premium Configuring a Microsoft Copilot Studio agent to filter a SharePoint knowledge source using KQL (Keyword Query Language) for a Department column typically requires advanced filtering capabilities, but "SharePoint Premium" is not a strict prerequisite for the KQL configuration itself. Instead, the primary requirement is a Microsoft 365 Copilot or Copilot Studio license to access these advanced features.

QUESTION 6 💩💩

You need to design a Microsoft Copilot Studio agent that meets the following requirements:

Supports interactive speech responses
Optimizes decision-making and the accuracy of responses

What should you include in the design for each requirement? To answer, drag the appropriate options to the correct requirements. Each option may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

Options

A deep reasoning model
Azure Language in Foundry Tools
Azure AI Speech
Copilot Studio voice features
Speech Synthesis Markup Language (SSML)

Answer Area

Supports interactive speech responses:
Optimizes decision-making and response accuracy:

✅ 答案

Supports interactive speech responses: → Copilot Studio voice features
Optimizes decision-making and response accuracy: → A deep reasoning model

🧠 记忆：

Voice → Copilot Studio voice features

Reasoning / accuracy → Deep reasoning model

Requirement 1

Supports interactive speech responses

关键词：

Supports interactive speech responses

这里是在设计 Microsoft Copilot Studio agent。

在 Copilot Studio 中应该加入什么功能？

新版 Copilot Studio 已经内建 Voice 能力（Voice-enabled agent）。

Microsoft 官方推荐：

Copilot Studio voice features 它封装了 Azure AI Speech 的能力，让 Agent 可以直接进行语音对话。

因此考试通常会优先选：

✅ Copilot Studio voice features

而不是 Azure AI Speech。

如果题目问：

哪个 Azure 服务提供 Speech-to-Text？才会选 Azure AI Speech。

Requirement 2

Optimizes decision-making and response accuracy

关键词：

decision-making
response accuracy

这是推理能力。

微软最新 Copilot Studio：

可以选择：

GPT-4.1
GPT-5
Deep reasoning models

Reasoning Model 专门负责：

多步推理
规划
更准确回答
更好的决策

因此：

✅ A deep reasoning model

为什么其他选项不对？

Azure AI Speech ❌

提供：

STT
TTS

属于底层 Azure 服务。

题目是在设计 Copilot Studio Agent，优先使用： Copilot Studio voice features

Azure Language in Foundry Tools ❌

负责：

NER
Sentiment
Classification

不是推理。

SSML ❌ 只是：控制

Voice
Pitch
Pause
Pronunciation

不能提供 Interactive Speech，也不会提高推理。

最终答案

Requirement	Correct Answer
Supports interactive speech responses	✅ Copilot Studio voice features
Optimizes decision-making and response accuracy	✅ A deep reasoning model

如果选项同时出现：

Azure AI Speech
Copilot Studio voice features

通常按下面原则选择：

设计 Copilot Studio Agent 功能 → ✅ Copilot Studio voice features
设计 Azure 架构 / Speech Service → ✅ Azure AI Speech

因为 Copilot Studio Voice Features 底层就是基于 Azure AI Speech，但考试更偏向选择更贴近设计层级的组件。

最终答案：

Supports interactive speech responses: ✅ Copilot Studio voice features
Optimizes decision-making and response accuracy: ✅ A deep reasoning model

Box 1: Copilot Studio voice features Supports interactive speech responses

Box 2: A deep reasoning model Optimizes decision-making and the accuracy of responses

Explanation/Reference:

Explanation:

Box 1: Copilot Studio voice features Supports interactive speech responses

To support interactive speech responses in a Microsoft Copilot Studio agent, you must incorporate specialized voice features and capabilities. These features enable the agent to move beyond chat-based interactions to handle voice-based input (speech-to-text) and output (text-to-speech).

Box 2: A deep reasoning model Optimizes decision-making and the accuracy of responses

Integrating a deep reasoning model (specifically Azure OpenAI o3) into Microsoft Copilot Studio optimizes decision-making and accuracy by enabling multi-step analysis, logical deduction, and the handling of complex, multi-turn conversations. This capability, enabled via the "reason" keyword in instructions, allows agents to analyze large datasets and provide context-aware, transparent answers, ideal for scenarios like financial analysis or compliance

QUESTION 7 💩💩

You are designing a low-code AI business solution by using Microsoft Copilot Studio.

The solution must include an agent that automates tasks by simulating user interactions across third-party apps and websites, such as clicking buttons, entering text, and extracting information from screens.

You need to recommend what to include in the agent.

What should you recommend?

A. Model Context Protocol (MCP)

B. a natural language understanding + (NLU+) model in Copilot Studio

C. Computer Use in Copilot Studio

D. Copilot skills

题目关键词

automates tasks by simulating user interactions

across third-party apps and websites

clicking buttons / entering text / extracting information from screens

这描述的正是微软最新推出的 Computer Use 功能。

✅ C. Computer Use in Copilot Studio

C. Computer Use in Copilot Studio

解析与考点分析

为什么选 C？

需求分析：题目要求智能体（Agent）能够通过模拟用户在第三方应用和网站上的交互动作来进行任务自动化，具体操作包括：
- 模拟点击按钮（Clicking buttons）
- 模拟输入文本（Entering text）
- 从屏幕中提取信息（Extracting information from screens）
概念匹配：
- Computer Use in Copilot Studio（计算机使用功能）：是 Microsoft 在 AI Agent 领域引入的核心能力。它赋予大语言模型（LLM）“看懂屏幕”和“像人类一样操作键盘鼠标”的能力。大模型通过分析屏幕图像/UI 画面，自主定位按钮位置并执行点击、打字和屏幕数据提取，非常适合缺乏 API 接口的第三方网站和旧版桌面应用的 UI 自动化。

为什么不选其他选项？

A. Model Context Protocol (MCP)： MCP 是一种基于 API/上下文通信的标准协议，用于连接 Agent 和后端数据服务，无法直接进行屏幕视觉识别和模拟鼠标/键盘的 UI 点击操作。
B. a natural language understanding + (NLU+) model in Copilot Studio： NLU+模型主要用于解析用户输入的自然语言意图（Intent）和提取实体（Entities），不能模拟屏幕交互。
D. Copilot skills（Copilot 技能）：属于传统 Bot Framework 或 API 驱动的扩展能力，并非通过视觉/屏幕模拟交互来实现自动化。

Correct Answer: C

Explanation/Reference:

Explanation:

In Microsoft Copilot Studio, you can use the computer use tool (currently in preview) to build low-code agents that automate tasks by simulating human interactions on a Windows computer.

Key Capabilities of Computer Use UI Interaction: Agents can interact with any system featuring a graphical user interface (GUI), including third-party websites and desktop applications.

*-> Human-like Actions: The tool enables the agent to click buttons, select menus, enter text into fields, and scroll, mimicking a human user.

No API Required: It is particularly useful for automating legacy systems or applications that lack direct API connectors.

Vision-Based Reasoning: Powered by Computer-Using Agents (CUA), the tool uses AI to "see" the screen, allowing it to adapt in real time if button positions or layouts change.

*-> Low-Code Setup: You define what the agent should do using natural language instructions rather than complex coding.

Reference:

QUESTION 8 💩💩

You need to recommend a solution to integrate a Microsoft Copilot agent with a Microsoft Dynamics 365 Contact Center chat channel.

The agent must respond to customer questions and hand off the conversation to a live customer service representative when the customer requests an escalation.

What should you recommend?

A. Build an agent flow.

B. Configure the Conversation Start topic.

C. Configure a skill.

D. Call a Microsoft Power Automate connector.

E. Configure the Escalate topic.

✅ 答案：E. Configure the Escalate topic.

原因：

题目关键是：

客户要求 escalation → 转交给 live customer service representative

在 Copilot Studio 中，Escalate topic 专门用于处理人工转接（handoff to a human agent）。

A. Agent flow ❌ 用于自动化流程
B. Conversation Start topic ❌ 用于对话开始
C. Skill ❌ 用于扩展 Agent 能力
D. Power Automate connector ❌ 用于连接外部服务
E. Escalate topic ✅ 用于转人工

🧠 记忆：

Customer requests human → Escalate topic

Correct Answer: E Section: (none)

Explanation/Reference:

Explanation:

To hand off a conversation from a Microsoft Copilot agent to a live representative in Dynamics 365 Contact Center, you must configure the Escalate system topic to include a Transfer conversation node. This ensures that when a customer requests an escalation, the agent seamlessly passes the full chat history and context to the live agent.

Steps to Configure the Escalate Topic Open the Escalate Topic: In Microsoft Copilot Studio, navigate to Topics, select the System tab, and open the Escalate topic.

Add a Transfer Node:

Delete any default "no one available" messages if they do not fit your workflow.

Select the Add node icon (+), go to Topic Management, and select Transfer conversation.

Include Context (Optional): You can add a Private message to agent within the node. This message is only visible to the live representative and helps them quickly understand the customer's needs.

Save and Publish: Save your changes and publish the agent to apply the new escalation logic.

QUESTION 9 💩💩

A company has a customer order system that creates sales orders manually.

You need to design an AI solution to automate the following tasks as part of the system:

Save the order details to a database.
Update the order status in the database.
Extract the order details from an order file.

Prepare and send a confirmation email to customers.

The solution must minimize development effort and support intelligent automation and solution integration.

What should you include in the design?

A. a workflow in Azure Logic Apps

B. a multi-agent solution that uses the Semantic Kernel SDK

C. a multi-agent solution that uses Microsoft Foundry Agent Service

D. a Microsoft Copilot Studio agent that uses Microsoft Power Automate workflows

正确答案与详细解析

正确答案：

D. a Microsoft Copilot Studio agent that uses Microsoft Power Automate workflows（使用 Microsoft Power Automate 工作流的 Microsoft Copilot Studio Agent）

解析与考点分析

为什么选 D？

业务需求分解：
- 从订单文件中提取订单明细（Extract order details）：使用 Power Automate 中内置的 AI Builder（文档处理 / 表单识别模型）可以无需代码快速提取文件内容。
- 将订单明细保存至数据库并更新状态（Save and update database）：使用 Power Automate 的内置数据库连接器（如 Dataverse 或 SQL Server Connector）进行操作。
- 准备并向客户发送确认电子邮件（Prepare and send confirmation email）：使用 Power Automate 的 Office 365 Outlook 连接器自动化发送邮件。
- 流程编排与智能交互：使用 Copilot Studio Agent 负责触发和调度整个流程。
核心限制条件：
- 题目明确要求：“最小化开发工作量（minimize development effort）” 并支持 “智能自动化与解决方案集成（intelligent automation and solution integration）”。
- Copilot Studio + Power Automate 是微软 Power Platform 体系下的低代码/无代码（Low-code/No-code）智能自动化标准组合，提供数百种现成的连接器与 AI 模块，能以最少的人工编码成本快速构建端到端的业务流。

为什么不选其他选项？

B. a multi-agent solution that uses the Semantic Kernel SDK（使用 Semantic Kernel SDK 的多 Agent 方案）：
C. a multi-agent solution that uses Microsoft Foundry Agent Service（使用 Microsoft Foundry Agent Service 的多 Agent 方案）：
- 这两个选项都属于专业代码（Pro-code）开发方案。需要编写代码（C#/Python）、维护 API 端点和自定义集成，极大地增加了开发和维护成本，直接违背了“最小化开发工作量”的要求。
A. a workflow in Azure Logic Apps（Azure Logic Apps 工作流）：属于 Azure 开发者级别的集成工具，虽然类似于 Power Automate，但缺乏 Copilot Studio 的智能 Agent 编排交互能力以及 Power Platform 内置开箱即用的 AI 文档处理集成。

Correct Answer: D Section: (none)

Explanation/Reference:

Explanation:

To create this system, you'll need to integrate Microsoft Copilot Studio with Power Automate and AI Builder. This combination allows your agent to not only "talk" but also "act" by executing complex backend workflows.

Solution Architecture Front-end: A Copilot Studio agent serves as the interface, receiving order requests or files from customers.

Brain (Logic): Power Automate cloud flows act as the "skills" for your agent, handling all database and email operations.

Data Extraction: AI Builder (Document Processing) extracts specific fields (like Item ID, Quantity, or Customer Name) from uploaded order files.

Storage: Microsoft Dataverse or SQL Server functions as the database to save and update order records

QUESTION 10

You are designing an AI strategy for Microsoft Dynamics 365 finance and operations apps. You are evaluating the use of Microsoft Copilot Studio to provide in-app help and guidance based on generative AI general knowledge.

You need to recommend which knowledge sources to include in the generative help and guidance agent.

The solution must minimize the risk of generating inaccurate responses.

What should you recommend? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Hot Area:

Answer Area

Custom knowledge sources:
- [ ] Must be uploaded to the agent (必须上传到智能体)
- [ ] Must be excluded from the agent (必须从智能体中排除)
- [ ] Are not supported (不支持)
AI general knowledge:
- [ ] Must be enabled for the agent (必须为智能体启用)
- [ ] Must be disabled for the agent (必须为智能体禁用)
- [ ] Is not supported (不支持)

答案：

Requirement	Correct Answer	原因
Custom knowledge sources	✅ Must be uploaded to the agent	上传企业内部知识（如文档、Dataverse、SharePoint 等），让 Agent 基于可信数据回答问题。
AI general knowledge	✅ Must be disabled for the agent	题目要求降低生成不准确回答的风险，禁用 AI General Knowledge 可减少幻觉（Hallucination），让回答仅基于企业知识。

考试关键点

企业知识问答 → 上传 Custom knowledge sources
减少幻觉 / 提高准确性 → Disable AI general knowledge

最终答案：

Custom knowledge sources: ✅ Must be uploaded to the agent
AI general knowledge: ✅ Must be disabled for the agent

Explanation/Reference:

Explanation:

Box 1: Must be uploaded to the agent Custom knowledge sources

Box 2: Must be enabled for the agent AI general knowledge

To implement a generative AI agent for in-app help in Dynamics 365 Finance and Operations while minimizing inaccuracies, you must configure the agent in Microsoft Copilot Studio by uploading specific knowledge sources and enabling general AI knowled

Upload Custom Knowledge Sources To ensure the agent provides accurate, organization-specific guidance, upload your internal documentation directly to the agent
Enable General AI Knowledge To allow the agent to use its own broad generative AI knowledge for general inquiries:

Open Microsoft Copilot Studio and select the Dataverse environment linked to your Finance and Operations apps.

Navigate to Agents and open the specific agent named Copilot for finance and operations apps.

On the Overview tab, find the Knowledge section and set Allow the AI to use its own general knowledge to Enabled.

Publish the changes to make this capability available in the D365 F&O sidecar.

QUESTION 11 💩💩

You need to design a multi-agent solution that will include a custom agent.

The solution must meet the following requirements:

Define the rules and constraints that the agent must follow.

Automate a backend process that involves data movement between services and runs independently of the agent's reasoning steps.

What should you include in the design for each requirement? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Answer Area

Define rules and constraints:
- [ ] Agent flows
- [ ] Conversation topics
- [ ] Microsoft Power Automate cloud flow
Automate a backend process:
- [ ] Conversation topics
- [ ] Microsoft Power Automate cloud flow
- [ ] Microsoft Power Pages

这是该道热点选择题（Hot Area）的正确答案与详细解析：

QUESTION 答案 (Answer Area)

1. Define rules and constraints:

正确选项： Conversation topics（对话主题）

2. Automate a backend process:

正确选项： Microsoft Power Automate cloud flow（Microsoft Power Automate 云端流）

解析与考点分析

第一部分分析：定义 Agent 必须遵循的规则和约束

考点： Copilot Studio Agent 的对话控制与边界约束机制。

解析：

在 Microsoft Copilot Studio 中，Conversation topics（对话主题） 是定义特定业务场景下对话逻辑、条件分支节点、变量规则以及确定性限制（Guardrails & Constraints）的核心组件。即使在使用生成式 AI 编排（Generative Orchestration）时，通过配置 Topic（主题），开发者依然能够明确限定 Agent 在处理特定业务时的触发规则和行为边界，防止 Agent 偏离预定业务规则。

第二部分分析：自动化跨服务数据迁移且独立于 Agent 推理步骤的后台流程

考点： 后台集成与流程自动化工具。

解析：

Microsoft Power Automate cloud flow（云端流）： 专为跨系统/服务（如 Dataverse、SharePoint、SQL、第三方 API）的数据迁移、异步处理和后台工作流自动化而设计。
题目明确要求“独立于 Agent 的推理步骤（runs independently of the agent's reasoning steps）”，这意味着该流程属于确定性的后台数据同步或批处理，不需要大语言模型（LLM）参与思考，因此使用 Power Automate 云端流 是标准且高效的解决方案。
Power Pages 是面向外部用户的网站构建工具，与后台流程自动化无关。

Box 1: Conversation topics

Box 2: Microsoft Power Automate cloud flow

Box 1: Conversation topics Define the rules and constraints that the agent must follow.

In a Microsoft AI project using Microsoft Copilot Studio, you define the rules, constraints, and dialogue paths for a custom agent primarily through Topics. Topics act as the agent's "competencies," determining how a conversation plays out based on specific user intents.

To use topics for defining agent behavior and constraints, follow these core principles:

Define Conversation Paths: Use topics to map out discrete, structured paths for the agent to follow. This allows you to enforce specific procedures, such as user verification, before providing sensitive information.

Set Triggers: Each topic begins with a trigger (phrases, keywords, or events) that signals when the agent should switch to a specific set of rules or logic.

Enforce Logic via Nodes: Within a topic, use nodes—such as questions, conditions, and actions—to define the logic, variables, and branching paths the agent must follow.

Modularize with "Bite-size" Topics: Break down complex agent logic into smaller, manageable topics. You can use the Redirect node to pass the conversation (and its constraints) from one topic to another.

Handle Errors and Fallbacks: Use System Topics (like On Error or Escalate) to define how the agent should behave when it hits a constraint it cannot resolve or fails to understand a query.

Box 2: Microsoft Power Automate cloud flow Automate a backend process that involves data movement between services and runs independently of the agent's reasoning steps.

In a Microsoft multi-agent project, you can use Power Automate cloud flows to handle backend data movement independently of an agent's reasoning steps. This approach is often referred to as classic orchestration.

By offloading structured, rule-based tasks to a cloud flow, you ensure that high-volume data operations or multi-service integrations remain deterministic and reliable, while your custom agents focus on dynamic, probabilistic decision-making.

QUESTION 12 💩💩

A company uses Microsoft 365 and Dynamics 365.

You need to recommend a solution to automatically summarize email threads, generate suggested replies in Microsoft Outlook, and provide meeting preparation summaries that include relevant customer relationship management (CRM) data.

Solution: You recommend Microsoft 365 Copilot for Sales.

Does this meet the goal?

A. Yes

B. No

正确答案与详细解析

正确答案：

A. Yes（是的，满足目标）

解析与考点分析

为什么选 A（Yes）？

对照题目提出的三项核心功能需求与 Microsoft 365 Copilot for Sales 的原生功能：

自动总结 Outlook 邮件会话（Summarize email threads in Microsoft Outlook）：
- Copilot for Sales 原生支持： 在 Outlook 中打开包含客户交易相关信息的邮件时，Copilot for Sales 会在侧边栏自动提取并生成邮件会话摘要（Email Summaries）。
在 Outlook 中生成建议回复（Generate suggested replies in Microsoft Outlook）：
- Copilot for Sales 原生支持： 结合 Dynamics 365 Sales 中的 CRM 数据（如产品报价、商机状态、客户历史），在 Outlook 中自动撰写包含专业 CRM 上下文的建议回复邮件草稿（Email Replies）。
提供包含相关 CRM 数据的会议准备摘要（Provide meeting preparation summaries that include relevant CRM data）：
- Copilot for Sales 原生支持： 在 Outlook 日程和 Teams 中，Copilot for Sales 会为即将到来的客户会议生成会议准备简报（Meeting Preparation Summaries），汇总该客户在 Dynamics 365 中的最新商机记录、最近联系历史和关键关注点。

因此，推荐 Microsoft 365 Copilot for Sales 完全且精准地满足了所有需求。

Correct Answer: A

Explanation/Reference:

Explanation:

Correct:

You recommend Microsoft 365 Copilot for Sales.

Incorrect:

You recommend a classic Microsoft Dataverse workflow.
You recommend a Microsoft 365 Copilot agent template.

Note:

In the described scenario, Microsoft 365 Copilot for Sales acts as the primary bridge between your productivity tools and CRM data.

It integrates directly into Microsoft Outlook and Teams to surface realtime insights from Dynamics 365 Sales or Salesforce.

Key capabilities for this specific workflow include:

Automated Email Summarization: Copilot scans long email threads in Outlook to extract key points, highlights, and BANT (Budget, Authority, Need, Timeline) data. If the sender is an external contact recognized in your CRM, the summary is automatically enriched with relevant account and opportunity data.

Suggested Email Replies: When replying to customer emails, Copilot generates drafts based on the context of the conversation and existing CRM data. You can use predefined response categories (e.g., "Reply to an inquiry," "Offer a proposal") or custom prompts to include specific opportunity details in the draft.

Meeting Preparation Summaries: Before a scheduled meeting, Copilot for Sales provides a "preparation card" in Teams or Outlook. This summary includes:

CRM Data: Matched opportunity and account attributes.
Contextual History: Summaries of past email exchanges and the last three seller notes.
Strategic Insights: Key risks, follow-up actions, and discussion points from previous interactions.

QUESTION 13

A company uses Microsoft 365 and Dynamics 365.

You need to recommend a solution to automatically summarize email threads, generate suggested replies in Microsoft Outlook, and provide meeting preparation summaries that include relevant customer relationship management (CRM) data.

Solution: You recommend a classic Microsoft Dataverse workflow.

Does this meet the goal?

A. Yes

B. No

正确答案与详细解析

正确答案：

B. No（不满足目标）

解析与考点分析

为什么选 B（No）？

题目需求：需要自动总结邮件会话（Summarize email threads）、生成智能建议回复（Generate suggested replies）以及提供包含 CRM 数据的生成式会议准备摘要（Meeting preparation summaries）。这些都需要生成式 AI（Generative AI / LLM）和与 Outlook 的原生 UI 交互能力。
解决方案分析：
- Classic Microsoft Dataverse workflow（传统的 Dataverse 工作流）：是 Dynamics 365 / Dataverse 中用于后台确定性数据操作的传统旧版工作流引擎（如：当状态变更时更新某个字段、按照固定模板发送一条通知邮件等）。
- 传统 Dataverse 工作流不具备生成式 AI 大模型能力，无法阅读并理解自然语言文本以生成总结，也无法在 Outlook 客户端界面中为销售人员提供智能回复和会议准备摘要。

正确答案应为上题中的 Microsoft 365 Copilot for Sales。

Correct Answer: B

Explanation:

Correct:

You recommend Microsoft 365 Copilot for Sales.. ✅

Incorrect:

You recommend a classic Microsoft Dataverse workflow. ❌
You recommend a Microsoft 365 Copilot agent template.. ❌

Note:

In the described scenario, Microsoft 365 Copilot for Sales acts as the primary bridge between your productivity tools and CRM data. It integrates directly into Microsoft Outlook and Teams to surface realtime insights from Dynamics 365 Sales or Salesforce.

Key capabilities for this specific workflow include:

Automated Email Summarization: Copilot scans long email threads in Outlook to extract key points, highlights, and BANT (Budget, Authority, Need, Timeline) data. If the sender is an external contact recognized in your CRM, the summary is automatically enriched with relevant account and opportunity data.

Suggested Email Replies: When replying to customer emails, Copilot generates drafts based on the context of the conversation and existing CRM data. You can use predefined response categories (e.g., "Reply to an inquiry," "Offer a proposal") or custom prompts to include specific opportunity details in the draft.

Meeting Preparation Summaries: Before a scheduled meeting, Copilot for Sales provides a "preparation card" in Teams or Outlook. This summary includes:

CRM Data: Matched opportunity and account attributes.
Contextual History: Summaries of past email exchanges and the last three seller notes.
Strategic Insights: Key risks, follow-up actions, and discussion points from previous interactions.

QUESTION 14

A company uses Microsoft 365 and Dynamics 365.

You need to recommend a solution to automatically summarize email threads, generate suggested replies

in Microsoft Outlook, and provide meeting preparation summaries that include relevant customer relationship management (CRM) data.

Solution: You recommend a Microsoft 365 Copilot agent template.

Does this meet the goal?

A. Yes

B. No

正确答案与详细解析

正确答案：

B. No（不满足目标）

解析与考点分析

为什么选 B（No）？

核心需求：自动在 Outlook 中基于 Dynamics 365（CRM）数据总结邮件会话、生成建议回复以及准备会议摘要。
解决方案分析：
- Microsoft 365 Copilot agent template（Microsoft 365 Copilot 智能体模板）：是用于在 Copilot Studio 中构建自定义声明式 Copilot（Declarative Copilot）或扩展插件的起点模板。
- 模板本身并不包含现成的 Dynamics 365 CRM 业务逻辑、数据集集成以及 Outlook 内置的销售场景工作流。如果使用智能体模板，开发人员需要进行大量自定义编码与数据连接开发，而无法直接实现开箱即用的 CRM 邮件总结和会议准备功能。
正确方案：微软针对此场景提供的开箱即用的官方产品是 Microsoft 365 Copilot for Sales（即本系列连续题的第一题方案）。

QUESTION 15 💩💩💩

You need to design an application lifecycle management (ALM) process for a Microsoft Power Platform environment that contains a solution named Solution1.

Solution1 must include a custom connector for Copilot in Microsoft Dynamics 365 Customer Service. Solution1 must meet the following requirements:

Ensure that the custom connector can be deployed consistently across environments as part of the ALM process.

Allow the custom connector to be edited only in the development environment.

What should you include in the design?

A. Add the custom connector to GitHub.

B. Share the custom connector.

C. Create the custom connector in the default solution.

D. Add the custom connector to Solution1.

正确答案：

D. Add the custom connector to Solution1.（将自定义连接器添加到 Solution1 中。）✅

解析与考点分析

为什么选 D？

Power Platform ALM 核心机制：
- 在 Microsoft Power Platform 中，解决方案（Solution）是包装、承载和跨环境迁移组件（包括应用、工作流、Dataverse 表以及自定义连接器 Custom Connectors）的唯一标准载体。
满足题目两大要求：
- 要求一：能够跨环境一致地部署（Deployed consistently across environments）：将自定义连接器（Custom Connector）添加到 Solution1 中，使其成为“解决方案感知（Solution-aware）”组件。后续只需将 Solution1 导出并导入到目标环境（测试/生产环境），即可实现跨环境的一致自动化部署。
- 要求二：仅允许在开发环境中编辑（Edited only in the development environment）：当包含该自定义连接器的 Solution1 以托管解决方案（Managed Solution）的形式部署到测试和生产环境时，该连接器在目标环境中将自动变为只读状态，确保只能在开发环境（Dev）中的未托管 Solution1 中进行修改。

为什么不选其他选项？

A. Add the custom connector to GitHub（将自定义连接器添加到 GitHub）： GitHub 是源码托管仓库。虽 ALM 流水线可以使用 GitHub，但如果连接器本身没有被打包进 Power Platform 的 Solution1 中，流水线也无法将其作为解决方案的一部分在环境间迁移。
B. Share the custom connector（共享自定义连接器）：共享只是赋予当前环境中其他用户使用或编辑该连接器的权限，无法实现跨环境的 ALM 部署和托管锁定。
C. Create the custom connector in the default solution（在默认解决方案中创建自定义连接器）：绝对不能在 Default Solution 中创建业务组件！默认解决方案（Default Solution）无法作为独立的干净包导出并部署到其他环境，这是 ALM 中的严重违规操作。

Correct Answer: D Section: (none)

Explanation/Reference:

Explanation:

To implement a consistent and secure Application Lifecycle Management (ALM) process for your custom connector, you should add the custom connector to the solution.

Adding the custom connector as a component of a solution is the standard method for ensuring it can be packaged and deployed across environments.

QUESTION 16 💩💩

A company uses a Microsoft Copilot Studio agent to automate tasks in a web app.

During testing, you discover that the automation sometimes fails because of frequent changes to the app's user interface.

You need to recommend a solution to ensure that the agent successfully automates the tasks.

The solution must minimize changes to the agent.

What should you include in the recommendation?

A. Computer Use in Copilot Studio

B. custom models in Azure AI Studio

C. conversation topics in Copilot Studio

D. an agent flow in Copilot Studio

正确答案与详细解析

正确答案：

A. Computer Use in Copilot Studio

解析与考点分析

为什么选 A？

业务痛点：传统的前端自动化（如基于 DOM 元素、XPath 或特定选择器的 RPA/Web 自动化）在目标 Web 应用的 UI 界面频繁发生变更（frequent changes to the app's user interface）时极易失效，需要不断人工维护和重新录制。
解决方案：
- Computer Use（计算机使用/视觉 UI 交互能力）：是生成式 AI Agent 的一项突破性能力（允许大语言模型结合多模态视觉能力，像人类一样通过“看”屏幕截图来识别按钮、文本框并操作鼠标/键盘）。
- 即使 Web 应用的前端代码、样式（CSS）或元素 ID 频繁改变，Computer Use 依然能够通过视觉语义识别目标位置并完成自动化任务，从而无需频繁去修改 Agent 的底层规则或重新配置选择器，完美满足“确保任务成功且最小化对 Agent 的修改（minimize changes to the agent）”这一要求。

为什么不选其他选项？

B. custom models in Azure AI Studio（Azure AI Studio 中的自定义模型）：需要自行训练或微调模型，开发和维护成本极高，无法“最小化对 Agent 的修改”。
C. conversation topics in Copilot Studio（Copilot Studio 中的对话主题）：用于定义对话分支和提示词逻辑，无法解决 Web 应用 UI 元素改变导致的自动化操作失败问题。
D. an agent flow in Copilot Studio（Copilot Studio 中的 Agent 流 / 自动化流）：如果使用标准的 Flow 或 Desktop Flow，其底层依然依赖确定的 API 或 DOM 选择器。UI 改变后 Flow 依然会报错，导致需要频繁手动维护。

Correct Answer: A

Explanation/Reference:

Explanation:

The Computer Use tool in Microsoft Copilot Studio is specifically designed to remedy automation failures caused by UI changes with minimal manual intervention.

Unlike traditional Robotic Process Automation (RPA), which relies on rigid selectors or fixed coordinates that break when a layout shifts, Computer Use is powered by Computer-Using Agents (CUA). This AI model combines computer vision with advanced reasoning to "see" the screen like a human, allowing it to adapt to interface changes in real-time.

Why Computer Use Fixes UI Fragility Visual Reasoning: Instead of looking for a specific HTML tag that might change, the agent looks for the visual concept of a button (e.g., "Submit" or a "plus icon").

Real-Time Adaptation: If a button moves from the left side of the screen to the right, the agent simply scans the screen and finds its new location without requiring you to update any code or scripts.

Natural Language Instructions: You describe the task in plain English (e.g., "Log into the portal and download the latest invoice"). If the portal changes its navigation menu, the AI reasons through the new layout to find the invoice link.

Self-Healing Properties: Because the agent makes decisions at each step based on the current visual state, it effectively "self-heals" during execution, overcoming obstacles that would typically crash a legacy bot.

QUESTION 17 💩💩💩

A company processes invoices stored across multiple systems in multiple formats. You need to implement an AI solution to automate the invoice processing. The solution must meet the following requirements:

Automate multi-step invoice processing tasks, including document analysis, data validation, and approval routing.

Enable users to interact directly via Microsoft Teams to review and approve invoices.

Minimize development efforts to define and customize approval workflows.

What should you include in the solution?

A. Azure Document Intelligence in Foundry Tools and Azure Logic Apps

B. a SharePoint agent

C. Microsoft Copilot Studio and AI Builder

D. Azure OpenAI and Azure Functions

正确答案：✅ C. Microsoft Copilot Studio and AI Builder

原因

题目三个关键词：

Invoice document analysis → AI Builder 可提取发票字段。
Teams 直接交互、审核和批准 → Copilot Studio Agent 可以集成到 Teams。
Minimize development efforts → 使用 Power Platform 的 low-code 能力，减少自定义代码。

其他选项

❌ A. Document Intelligence + Logic Apps → 可以处理文档和工作流，但 Teams 对话式审批以及低代码 Agent 能力不如 Copilot Studio + AI Builder。
❌ B. SharePoint agent → 主要用于 SharePoint 内容问答，不适合完整的发票处理流程。
❌ D. Azure OpenAI + Azure Functions → 需要更多自定义开发，不符合 minimize development efforts。

最终答案：C. Microsoft Copilot Studio and AI Builder

Correct Answer: C

Explanation/Reference:

Explanation:

To automate invoice processing using Microsoft Copilot Studio and AI Builder, you can build an "Intelligent Invoicing Agent" that serves as the central orchestration layer. This solution combines AI-driven data extraction with low-code approval workflows accessible directly within Microsoft Teams.

Solution Architecture The most efficient approach uses Copilot Studio as the "brain" and AI Builder as the "eyes" to read documents.

Extraction Layer (AI Builder): Uses the Prebuilt Invoice Model to automatically identify and extract data like vendor name, date, and total amount without manual training.

Orchestration Layer (Copilot Studio): Hosts "Agent Flows" (advanced workflows) that manage the end-to-end logic, from receiving a file to routing it for approval.

Interaction Layer (Microsoft Teams): Users upload invoices or receive approval cards directly in a chat window, eliminating the need to switch between different accounting apps.

QUESTION 18 💩💩

You need to design a Microsoft Copilot Studio agent for customer support.

The agent must securely retrieve product warranty data from a REST API. The solution must minimize development effort.

What should you include in the design?

A. Export the agent as a managed solution and customize the agent in Power Apps.

B. Create a custom connector in Copilot Studio and use the connector to call the API.

C. Use a Microsoft Power Automate desktop flow to screen scrape the warranty data.

D. Add the warranty data to the Fallback topic.

正确答案：✅ B. Create a custom connector in Copilot Studio and use the connector to call the API.

原因

题目关键词：

REST API
securely retrieve data
minimize development effort

Custom Connector 可以把现有 REST API 封装成 Copilot Studio 可调用的 action/tool，无需自己开发大量 API integration code

其他选项：

❌ A → Managed solution 与调用 REST API 无关。
❌ C → Desktop flow / screen scraping 复杂且不适合直接调用 REST API。
❌ D → Fallback topic 用于处理无法理解的用户输入，不适合访问外部 API。

🎯 记忆

Existing REST API + Copilot Studio + Low-code → Custom Connector

答案：B

Correct Answer: B

Explanation/Reference:

Explanation:

To securely retrieve data from a REST API for a customer support agent in Microsoft Copilot Studio while minimizing development, creating a custom connector is a highly efficient approach. This method acts as a no-code/low-code wrapper around your API, allowing the agent to perform actions or fetch real-time information.

Implementation Steps

Create the Custom Connector

You can build this within Power Apps or Power Automate under the Data or Custom Connectors menu.

Add the Connector as an Action

Once the connector is created and published, you can add it directly to your agent.

QUESTION 19 💩

A company has an ecommerce support portal that uses Microsoft Dataverse.

You are designing a Microsoft Copilot Studio agent for the portal. The agent must meet the following requirements:

Respond with a default help message when the user input is unclear.
Initiate external processes, such as retrieving the order status, when users make specific requests.
Generative orchestration will be enabled for the solution.

You need to recommend a feature for each requirement.

What should you recommend? To answer, drag the appropriate features to the correct requirements. Each feature may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

NOTE: Each correct selection is worth one point.

Answer Area

Define rules and constraints:
- [ ] Agent flows
- [ ] Conversation topics
- [ ] Microsoft Power Automate cloud flow
Automate a backend process:
- [ ] Conversation topics
- [ ] Microsoft Power Automate cloud flow
- [ ] Microsoft Power Pages

✅ 答案

Define rules and constraints: → Conversation topics
Automate a backend process: → Microsoft Power Automate cloud flow

🧠 记忆

Rules / constraints → Topics

Backend automation → Power Automate cloud flow

QUESTION 答案 (Answer Area)

Define rules and constraints (或：当用户输入不明确时回复默认帮助消息): 👉 Conversation topics（对话主题）
Automate a backend process (或：触发检索订单状态等外部后台流程): 👉 Microsoft Power Automate cloud flow（Microsoft Power Automate 云端流）

解析与考点分析

Define rules and constraints / 默认为不明确输入回复帮助消息
对应选择： Conversation topics
解析：
- 在 Microsoft Copilot Studio 中，Conversation topics（对话主题）用于定义对话分支、响应规则以及在特殊情况下的行为约束。
- 当用户输入不明确（Unclear input）或触发未知意图时， Copilot Studio 会自动路由到系统内置的降级主题——Fallback Topic（降级主题，属于 Conversation topics 的一部分），从而输出默认的帮助信息并引导用户。
Automate a backend process / 触发检索订单状态等外部流程
对应选择： Microsoft Power Automate cloud flow
解析：
- 当用户提出具体请求（例如：“查询我的订单状态”或“取消预约”）时，智能体需要调用外部系统或 API 执行数据检索与后台操作。
- Microsoft Power Automate cloud flow（云端流）：是 Copilot Studio 触发外部后台流程（如连接 Dataverse、第三方 ERP、发送通知或调用 Webhook 检索订单状态）的标准工具与插件载体。

QUESTION 20 💩💩💩💩💩

A company uses Microsoft Dynamics 365 to manage service operations.

Dispatchers coordinate service requests, and technicians perform scheduled on-site work.

You need to design a solution that will use Microsoft Copilot to improve the efficiency of the service operations.

The solution must meet the following requirements:

Provide AI-driven assistance to help staff organize and resolve work orders.
Deliver contextual AI support to frontline workers as they prepare for and complete customer appointments.

Which two components should you include in the design? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. Copilot Service workspace

B. Copilot in Outlook

C. Dynamics 365 Customer Service

D. Copilot in Customer Service

E. Copilot in Field Service

F. the Dynamics 365 Field Service mobile app

正确答案与详细解析

正确答案：

E. Copilot in Field Service
F. the Dynamics 365 Field Service mobile app

解析与考点分析

为什么选 E 和 F？

题目描述的是典型的 Dynamics 365 Field Service（现场服务）业务场景，涉及两个核心角色：调度员（Dispatchers）和一线技术人员（Frontline Workers / Technicians）。

对照题目的两大核心需求：

需求一：“提供 AI 驱动的帮助，协助员工组织和解决工单（organize and resolve work orders）”
- 对应选项 E (Copilot in Field Service)：专为 Field Service 调度员和管理者设计。Copilot 可以自动汇总工单内容、提取客户需求、智能推荐调度排程，并自动生成工单总结，极大地提升调度员管理和组织工单的效率。
需求二：“在一线员工准备和完成客户预约时，为其提供上下文 AI 支持（Deliver contextual AI support to frontline workers...）”
- 对应选项 F (the Dynamics 365 Field Service mobile app)：一线技术人员在前往客户现场（On-site）执行任务时，使用的是 Dynamics 365 Field Service 移动应用。移动应用中嵌入了 Copilot AI 功能，技术人员可以在现场查看 AI 汇总的工单历史、快速撰写客户跟进更新、获取操作步骤指导，从而高效完成现场预约任务。

为什么不选其他选项？

A, C, D (Dynamics 365 Customer Service 相关组件)：适用于呼叫中心/客服代表（Customer Service Agents）处理在线咨询和热线工单，不适用于现场上门服务（Scheduled on-site work）、调度员排程以及一线现场技术人员场景。
B (Copilot in Outlook)：属于通用的 Outlook 邮件处理助手，无法为现场技术人员在移动端执行上门服务预约时提供专用的现场工作流上下文支持。

QUESTION 21

A company uses Microsoft Foundry agents.

You need to ensure that an agent can dynamically use external tools at runtime without updating the agent.

What should you include in the solution?

A. a Microsoft Foundry hub

B. a Model Context Protocol (MCP) server

C. Azure AI Search

D. Microsoft Copilot Studio

正确答案：

B. a Model Context Protocol (MCP) server

解析与考点分析

为什么选 B？

概念与核心特性：
Model Context Protocol (MCP，模型上下文协议)： 是当前 AI 领域用于连接 AI Agent 与外部数据源/工具的开放标准协议（在 Azure AI Foundry 中被原生支持）。
MCP 服务器（MCP Server）： 充当 Agent 与外部系统之间的适配器。它将外部工具、数据库和 API 统一封装并公开。
满足题目需求：
题目要求 Agent 能够在运行时动态使用外部工具，且无需更新 Agent 本身（dynamically use external tools at runtime without updating the agent）。
通过接入 MCP 服务器，当后台新增、修改或扩展外部工具时，Agent 会在运行时通过 MCP 协议自动动态发现（Dynamic Discovery）并调用这些新工具，而不需要对 Agent 的代码或模型配置进行重新部署和更新。

为什么不选其他选项？

A. a Microsoft Foundry hub（Microsoft Foundry 枢纽）： 是 Azure AI Foundry 的顶层管理资源，用于组织团队资源、共享连接和安全治理，不提供运行时动态工具发现和调用的协议能力。
C. Azure AI Search： 是检索增强生成（RAG）中的向量数据库和全文搜索服务，属于数据源，而非用于动态执行外部工具的协议/服务器。
D. Microsoft Copilot Studio： 是 Power Platform 体系下的低代码 SaaS 智能体构建工具，与 Azure AI Foundry 下开发的代码级/Pro-code Agent 属于不同的平台体系。

Correct Answer: B

Explanation/Reference:

n Microsoft Foundry, the Model Context Protocol (MCP) is the architectural standard used to decouple an agent's logic from its toolset, allowing it to discover and use external tools at runtime without requiring code updates or re-deployment.

Why an MCP Server is Required Traditionally, integrating a tool required "hardcoding" its schema into the agent's definition. MCP replaces this with a dynamic discovery model:

Decoupling: The agent is only configured with the MCP server endpoint rather than individual tool definitions.

Runtime Discovery: When the agent starts a session, it queries the MCP server for a manifest of available tools, their parameters, and descriptions.

Zero-Update Scaling: You can add new capabilities to the MCP server (e.g., a new database query or API connection), and the agent will automatically "see" and use them in the next turn without any changes to the agent itself.

QUESTION 22 💩💩💩

A company has a Microsoft Dynamics 365 Sales environment that has Microsoft Copilot enabled.

You need to customize Copilot by tailoring how opportunity summaries are generated or how they are presented to users.

Solution: You add fields to the opportunity summary.

Does this meet the goal?

A. Yes

B. No

答案：✅ A. Yes

解析

在 Dynamics 365 Sales 中，Copilot 提供 Opportunity Summary（商机摘要）。

管理员可以自定义摘要内容，例如：

添加或移除字段
调整摘要中显示的信息
根据业务需求展示特定字段

因此，添加字段到 Opportunity Summary 就是在定制 Copilot 生成或呈现商机摘要，符合题目要求。

考试记忆

自定义 Opportunity Summary → ✅ 可以通过添加/配置字段实现。
无需修改模型或编写代码。

最终答案：✅ A. Yes

Correct Answer: A

Explanation/Reference:

Correct:

You add fields to the opportunity summary.

Incorrect:

You build Microsoft Power Automate flows to trigger customized Copilot summaries.
You configure AI Builder lead scoring models to influence opportunity summaries.

Note:

To customize Microsoft Copilot opportunity summaries and incorporate AI Builder lead scoring data, you must configure the specific fields Copilot uses to ground its generative AI outputs.

By default, Copilot for Sales generates summaries using a set of predefined fields. To "influence" these summaries with scoring data, you need to add the predictive score and grade fields to the Opportunity summary configuration.

Step-by-Step Configuration Switch Area: In the Sales Hub app, select the Change area menu in the bottom-left corner and choose App Settings.

Navigate to Copilot: Under General Settings, select Copilot.

Select Entity: Choose the Opportunities tab.

Add Fields:

Click Add fields.
Select the checkboxes for the fields you want to include.
You can select out-of-the-box, custom, and related table fields.
Save: Click Add and then Save your changes to update the summary configuration.

QUESTION 23

A company has a Microsoft Dynamics 365 Sales environment that has Microsoft Copilot enabled.

You need to customize Copilot by tailoring how opportunity summaries are generated or how they are presented to users.

Solution: You build Microsoft Power Automate flows to trigger customized Copilot summaries.

Does this meet the goal?

A. Yes

B. No

答案：❌ B. No

解析

题目要求：

Customize Copilot by tailoring how opportunity summaries are generated or presented to users.

在 Dynamics 365 Sales Copilot 中，自定义 Opportunity Summary 通常通过：

配置 Copilot 设置
添加/调整摘要字段
调整 Copilot 相关配置

实现。

而 Power Automate flows：

用于业务流程自动化
触发操作、发送通知、更新数据等
不能修改 Copilot 内置摘要生成逻辑或呈现方式

所以：

❌ 创建 Power Automate Flow 不能达到定制 Copilot Opportunity Summary 的目标。

最终答案：B. No

Correct Answer: B

Explanation/Reference:

Explanation:

Correct:

You add fields to the opportunity summary.

Incorrect:

You build Microsoft Power Automate flows to trigger customized Copilot summaries.
You configure AI Builder lead scoring models to influence opportunity summaries.

Note:

To customize Microsoft Copilot opportunity summaries and incorporate AI Builder lead scoring data, you must configure the specific fields Copilot uses to ground its generative AI outputs.

By default, Copilot for Sales generates summaries using a set of predefined fields. To "influence" these summaries with scoring data, you need to add the predictive score and grade fields to the Opportunity summary configuration.

Step-by-Step Configuration Switch Area: In the Sales Hub app, select the Change area menu in the bottom-left corner and choose App Settings.

Navigate to Copilot: Under General Settings, select Copilot.

Select Entity: Choose the Opportunities tab.

Add Fields:

Click Add fields.
Select the checkboxes for the fields you want to include.
You can select out-of-the-box, custom, and related table fields.
Save: Click Add and then Save your changes to update the summary configuration.

QUESTION 24

A company has a Microsoft Dynamics 365 Sales environment that has Microsoft Copilot enabled.

You need to customize Copilot by tailoring how opportunity summaries are generated or how they are presented to users.

Solution: You configure AI Builder lead scoring models to influence opportunity summaries.

Does this meet the goal?

A. Yes

B. No

正确答案：❌ B. No

原因

AI Builder lead scoring models 用于对 Lead（潜在客户）进行评分/预测转化可能性，不会用来配置或影响 Dynamics 365 Sales Copilot 的 opportunity summaries。

如果要定制 Opportunity Summary，应该通过 Opportunity Summary 的字段配置/相关 Copilot customization capabilities 来调整内容。

🎯 记忆

Lead scoring → Lead 优先级/评分
Opportunity summary → Opportunity summary configuration

答案：B. No

Correct Answer: B

Correct:

You add fields to the opportunity summary.

Incorrect:

You build Microsoft Power Automate flows to trigger customized Copilot summaries.
You configure AI Builder lead scoring models to influence opportunity summaries

Note: To customize Microsoft Copilot opportunity summaries and incorporate AI Builder lead scoring data, you must configure the specific fields Copilot uses to ground its generative AI outputs.

By default, Copilot for Sales generates summaries using a set of predefined fields. To "influence" these summaries with scoring data, you need to add the predictive score and grade fields to the Opportunity summary configuration.

Step-by-Step Configuration Switch Area: In the Sales Hub app, select the Change area menu in the bottom-left corner and choose App Settings.

Navigate to Copilot: Under General Settings, select Copilot.

Select Entity: Choose the Opportunities tab.

Add Fields:

Click Add fields.
Select the checkboxes for the fields you want to include.
You can select out-of-the-box, custom, and related table fields.
Save: Click Add and then Save your changes to update the summary configuration.

QUESTION 25

A company has a Microsoft Dynamics 365 Sales environment that has Microsoft Copilot enabled.

You need to customize Copilot by tailoring how opportunity summaries are generated or how they are presented to users.

Solution: You add the opportunity summary widget to the Opportunity form.

Does this meet the goal?

A. Yes

B. No

正确答案：❌ B. No

原因

把 Opportunity summary widget 添加到 Opportunity form，主要是把已有的摘要组件显示在表单上，并不能真正定制：

摘要是如何生成的
摘要包含哪些内容
摘要如何呈现/定制

所以不能满足题目要求的 “customize Copilot by tailoring how opportunity summaries are generated or presented”。

答案：B. No

Correct Answer: B

Explanation/Reference:

Explanation:

Correct:

You add fields to the opportunity summary.

Incorrect:

You add the opportunity summary widget to the Opportunity form.
You build Microsoft Power Automate flows to trigger customized Copilot summaries.
You configure AI Builder lead scoring models to influence opportunity summaries

Note:

To customize Microsoft Copilot opportunity summaries and incorporate AI Builder lead scoring data, you must configure the specific fields Copilot uses to ground its generative AI outputs.

By default, Copilot for Sales generates summaries using a set of predefined fields. To "influence" these summaries with scoring data, you need to add the predictive score and grade fields to the Opportunity summary configuration.

Step-by-Step Configuration Switch Area: In the Sales Hub app, select the Change area menu in the bottom-left corner and choose App Settings.

Navigate to Copilot: Under General Settings, select Copilot.

Select Entity: Choose the Opportunities tab.

Add Fields:

Click Add fields.

Select the checkboxes for the fields you want to include.

You can select out-of-the-box, custom, and related table fields.

Save: Click Add and then Save your changes to update the summary configuration.

Deploy AI-powered business solutions

Testlet 1

Overview

Contoso, Ltd. is a high-tech manufacturing company that uses Microsoft Dynamics 365 Finance. Dynamics 365 Supply Chain Management, and Dynamics 365 Commerce for its North American operations. The company designs and develops innovative products that have many patents and proprietary technologies. The patents and engineering designs are closely guarded secrets.

Contoso executives want to integrate and adopt AI solutions to help scale the company in preparation for an anticipated period of rapid growth.

The company has multiple legal entities and Azure subscriptions that will be used in the adopted AI solutions.

Requirements

AI Adoption The following executives will have specific responsibilities in the overall AI adoption:
Chief Technology Officer (CTO): Select one Dynamics 365 Finance, Dynamics 365 Supply Chain Management or Dynamics 365 Commerce prebuilt AI agent and one custom Microsoft Copilot Studio AI agent to prioritize and deploy during the initial AI adoption phase.
Chief Information Officer (CIO): Ensure that appropriate security labels are assigned to the data used by the AI agents.
Chief Financial Officer (CFO): Analyze the return on investment (ROI) for the AI agents being deployed. Chief Information Security Officer (CISO): Discover and inventory AI resources for auditing.
Chief Executive Officer (CEO): Ensure that all solutions adhere to industry-standard responsible AI practices

All AI initiatives and agents will have a detailed business use case, a defined audience profile, and an estimated ROI that will compare the cost savings of the current process against the estimated costs of using the new AI solutions.

The company's research and development (R&D) department already has a custom Model Context Protocol (MCP) server that contains comprehensive product specifications and compliance data

Prebuilt AI Agent

The CTO has NOT yet selected which prebuilt AI agent to use in Dynamics 365 Supply Chain Management. The CTO wants to view available agent templates to identify which agent will add the most business value.

Depending on which high-priority AI agents are identified, its agent capabilities must be previewed in a discovery meeting with the relevant business operation stakeholders.

Custom AI Agent

Contoso has identified the following custom AI agent requirements:

The custom AI agent will use data from Dynamics 365 Supply Chain Management to answer questions for the manufacturing team as a low-code solution.
The custom AI agent will be accessible from within Microsoft Teams.
The custom AI agent must be designed to eventually connect to other agents that can be selected based on their description.
The topics used in the custom AI agent will be selected based NOT on a trigger phrase, but on a description of the purpose of the query, to make the interactions more conversational.
The custom AI agent must be able to answer questions about product specifications by using existing technologies. The product specifications are maintained by the R&D department.
The custom AI agent must be integrated with and accessible from Dynamics 365 Supply Chain Management.
The custom AI agent must be able to use Dynamics 365 Supply Chain Management business logic that is stored outside of the application.

Analysis, Reporting, and Troubleshooting

Contoso has identified the following analysis, reporting, and troubleshooting requirements:

The CFO will analyze all the AI solutions quarterly to compare the estimated ROI against actual measured efficiencies and adoption. The CFO will use the Copilot Studio agent usage estimator to perform this analysis.
The CISO wants to identify how much sensitive data was accessed for a given AI agent run and who accessed the data. Too much sensitive data accessed by a single user might indicate a high security risk.
The CTO wants to track user feedback on the quality of the AI agent responses during user interactions with the agents. Consistently poor feedback will trigger an escalated reengineering discussion.
The CEO wants a quarterly assessment of all the required metrics for their specific responsibilities. The tools used for the assessments must be Microsoft-recommended and must verify reliability, interpretability, fairness, and compliance.
The CFO wants to identify how many interactions with the AI agents are abandoned on a given day as compared to resolved conversations. Too many abandoned sessions might indicate that Copilot Studio credits are being used inefficiently by end users.

QUESTION 1 💩💩💩💩💩

Which two components for the custom AI agent should you include in the application lifecycle management (AIM) process? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. an Azure package

B. a ZIP package

C. a Microsoft Power Platform solution

D. a Cloud Scale Unit (CSU) package

E. an X++model

正确答案：✅ C、B

C. Microsoft Power Platform solution → Copilot Studio Agent 的 ALM 核心载体，用于跨 Dev/Test/Prod 部署。
B. ZIP package → 用于打包/导入导出 Agent 相关组件。

记忆

Copilot Studio Agent ALM → Power Platform Solution + ZIP package

所以答案：B + C。

Correct Answer: BC Section: (none)

Explanation/Reference:

[C] A Microsoft Power Platform solution is the standard container for transporting and managing customizations across the Power Suite (Copilot Studio, Power Automate, and Power Apps). It perfectly aligns with the specific requirements.
[B] Based on the application lifecycle management (ALM) process for a Microsoft Power Platform solution that integrates with Dynamics 365 Supply Chain Management, you should use a ZIP package.

Recommended ALM Component In the Microsoft Power Platform ecosystem, the standard unit of deployment for ALM is a Solution, which is exported and transported as a ZIP package.

ZIP package: This is the standard format for Power Platform solutions containing Copilot Studio agents, Power Automate flows, and custom connectors used to bridge Dynamics 365 and Teams.

Incorrect:

[Not A] Azure package: Typically refers to Azure-native resources; while used for custom coding, the Power Platform solution itself is handled via ZIP exports.

[Not D] Cloud Scale Unit (CSU) package: Specifically for commerce or edge workloads, not for deploying general AI agents.

[Not E] X++ model is used for core code changes inside Dynamics 365 Finance & Operations, but not for the lowcode Power Platform/Copilot Studio orchestration described.

QUESTION 2

Which tools should you recommend to assist the CISO and the CIO with their specific responsibilities? To answer, drag the appropriate tools to the correct executives. Each tool may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

Tools (待拖拽选项)
Azure Blob Storage
Azure Resource Graph Explorer
Copilot Studio
Microsoft Purview

Answer Area (答题区)

CISO (首席信息安全官):
- [ 拖拽目标位置 1 ]
CIO (首席信息官):
- [ 拖拽目标位置 2 ]

答案：

CISO: ✅ Azure Resource Graph Explorer
CIO: ✅ Microsoft Purview

原因

CISO → Azure Resource Graph Explorer

题目中 CISO 的职责是 discover and inventory AI resources for auditing。

Resource Graph Explorer 可以查询和盘点 Azure 资源，适合做 AI resource inventory。

CIO → Microsoft Purview

CIO 的职责是确保数据有适当的 security labels / classification。

Microsoft Purview 提供数据发现、分类和敏感度标签（sensitivity labels）能力。

🎯 记忆

CISO → Discover / Inventory resources → Azure Resource Graph Explorer

CIO → Data classification / Sensitivity labels → Microsoft Purview

最终：CISO = Azure Resource Graph Explorer；CIO = Microsoft Purview

这是该道针对 Contoso, Ltd. 案例分析的拖拽匹配题（Select and Place）的正确答案与详细解析：

QUESTION 答案 (Answer Area)

CISO (首席信息安全官): 👉 Azure Resource Graph Explorer (或 Microsoft Purview)
CIO (首席信息官): 👉 Microsoft Purview

在 Contoso, Ltd. 案例中，CISO 与 CIO 拥有非常明确的职责与工具对应关系：

CIO (Chief Information Officer - 首席信息官)

案例核心职责要求：

"Chief Information Officer (CIO): Ensure that appropriate security label are assigned to the data used by the AI agents." （CIO 职责：确保为 AI Agent 使用的数据分配适当的安全/敏感性标签。）

对应工具： Microsoft Purview
解析：在微软生态中，为数据资产创建、分配和强制执行敏感性标签（Sensitivity Labels / Security Labels）的官方核心工具是 Microsoft Purview Information Protection (MPIP)。

CISO (Chief Information Security Officer - 首席信息安全官)

案例核心职责要求： "Chief Information Security Officer (CISO): Discover and inventory AI resources for auditing." （CISO 职责：发现并清查（盘点）AI 资源以供审计。）
对应工具： Azure Resource Graph Explorer
解析：案例中提到 Contoso 拥有多个法律实体和多个 Azure 订阅（Multiple Azure subscriptions）。要在跨订阅的云端环境中高效地搜索、发现和盘点（Discover and inventory）所有已部署的 AI 资源（如 Azure OpenAI 实例、AI Services 资源等），微软推荐的标准工具是 Azure Resource Graph Explorer（支持使用 KQL 快速跨订阅查询和盘点云资源）。

(注：CISO 在后续审计敏感数据访问时也会用到 Microsoft Purview，如果该题每个框只填一个主选工具，CISO 的首选核心工具即为 Azure Resource Graph Explorer；若允许选多个，Purview 同样适用。)

QUESTION 3

What should you recommend to assist the CEO with their specific responsibilities?

A. the Microsoft Service Trust Portal

B. Microsoft Foundry Tools

C. Microsoft Purview

D. the Responsible AI dashboard

E. Compliance Center

Correct Answer: D

Explanation/Reference:

Explanation:

Scenario: The CEO wants a quarterly assessment of all the required metrics for their specific responsibilities. The tools used for the assessments must be Microsoft-recommended and must verify reliability, interpretability, fairness, and compliance.

To ensure your Microsoft Dynamics 365 AI agent meets quarterly compliance and ethical standards, the Responsible AI (RAI) Dashboard in Azure Machine Learning is the primary recommended tool.

Microsoft recommends aligning quarterly assessments with their Responsible AI Standard v2, which focuses on six core principles: fairness, reliability, safety, privacy, security, inclusiveness, transparency, and accountability.

QUESTION 4

Which Copilot Studio analytics metrics should you recommend to assist the executives with their specific responsibilities? To answer, drag the appropriate metrics to the correct executives. Each metric may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

NOTE: Each correct selection is worth one point.

Select and Place:

Metrics (Options):

[ Effectiveness ]
[ Tool use ]
[ Satisfaction ]
[ Use ]

Answer Area:

CFO concerns about Copilot Studio credit usage: [ ________ ]
CTO concerns about poor feedback on AI agent responses: [ ________ ]

Box 1: Effectiveness

Box 2: Satisfaction

正确答案

Requirement	答案	原因
CFO – Copilot Studio credit usage	✅ Effectiveness	Effectiveness 关注 Session Outcomes，包括 Resolved / Abandoned / Escalated，正好对应题目要求的 abandoned vs. resolved。
CTO – Poor feedback on AI responses	✅ Satisfaction	Satisfaction 关注用户反馈，包括 thumbs up/down（Reactions），用于判断回答质量。

为什么 CFO 不是 Use？

题目虽然提到：

Copilot Studio credits are being used inefficiently

但 CFO 实际要测量的指标是：

abandoned sessions vs. resolved conversations

这是 Effectiveness → Session Outcomes 的指标。

而 Use 更偏向 Agent 的使用情况/使用量，例如使用次数、sessions 等，并不能直接回答 Resolved vs. Abandoned。

考试记忆

Resolved / Abandoned / Escalated → ⭐ Effectiveness
Thumbs up / Thumbs down / User feedback → ⭐ Satisfaction
Usage / adoption / how much used → Use
Tool invocation / tool performance → Tool use

所以这道题最终：

Box 1: Effectiveness ✅

Box 2: Satisfaction ✅

Box 1: Effectiveness

CFO concerns about Copilot Studio credit usage

Scenario: The CFO wants to identify how many interactions with the AI agents are abandoned on a given day as compared to resolved conversations. Too many abandoned sessions might indicate that Copilot Studio credits are being used inefficiently by end users.

To identify how many interactions with AI agents are abandoned compared to resolved, you should use the Effectiveness metrics.

In Microsoft Copilot Studio, Effectiveness specifically tracks Session Outcomes, which categorize every engaged conversation into one of three primary states:

Resolved: The user's issue was addressed, and the session ended successfully.

Abandoned: The user stopped interacting without reaching a resolution or escalation, or the session timed out.

Escalated: The conversation was handed over to a human agent.

Box 2: Satisfaction

CTO concerns about poor feedback on AI agent responses

Scenario: The CTO wants to track user feedback on the quality of the AI agent responses during user interactions with the agents. Consistently poor feedback will trigger an escalated reengineering discussion.

AI Overview To track user feedback on the quality of AI agent responses in Microsoft Copilot Studio (and Dynamics 365), you should use the Satisfaction analytics metric. Specifically, the Reactions chart within the Satisfaction section tracks user-provided feedback, which measures user perception of response quality.

Satisfaction (Reactions): This section captures direct user feedback (thumbs up/down) on individual answers and provides a ratio of positive to negative feedback, including specific user comments for detailed analysis.

Deploy AI-powered business solutions Testlet 2

Background

Fabrikam, Inc., is a global consumer goods company that is undergoing a digital transformation initiative to migrate its entire infrastructure to the Microsoft cloud. As a key element of this cloud migration, the company will implement Microsoft Dynamics 365 Sales, moving away from the current on-premises proprietary technologies used by its business-to-business (B2B) sales team.

As part of the cloud migration, Fabrikam will adopt an AI-first approach to its business solutions and implement AI solutions, wherever possible, to streamline operations.

Problem Statements

Fabrikam's infrastructure currently relies on various on-premises systems that require sales executives to use corporate computers with physical keyboards to access business information during customer interactions. Mobile phones cannot be used for these purposes, as the systems depend on keyboard input. As a result, the sales executives spend a lot of time using keyboards to search for data on several disparate systems and file servers, rather than focusing on the customers. This affects the customer experience.

Fabrikam stakeholders are concerned that users will be hesitant to adopt AI. If the AI initiatives are NOT adopted, cost savings will never be realized. Additionally, funding for future AI initiatives will depend on demonstrating an increase in AI adoption month over month. As the AI agent initiative for the sales team will be the first for Fabrikam, the rapid adoption of the agent is a high priority.

Planned Initiatives General

Fabrikam management has prioritized AI-driven projects to improve efficiency, customer engagement, and responsible AI adoption. The current application infrastructure is on-premises and must be migrated to the cloud to support the adoption of these technologies.

Infrastructure Migration

Fabrikam plans to migrate from its current on-premises infrastructure to a completely cloud-based topology; this will include user authentication, the security framework, and, primarily, the adoption of the services by end users.

All the data from the different systems will be consolidated into a single data source - a common data model that will use a Microsoft Dataverse environment as a single source of truth (SSOT) for the sales team.

Sales Cycle Enablement

To achieve the company's objectives, Fabrikam intends to implement the following strategies to enhance the sales cycle:

Use low-code development to create a single AI agent that has Dataverse as its core component. Ensure that sales managers can access unanswered correspondence from prospects and intervene as appropriate.

Replace the previous proprietary software with Dynamics 365 Sales to track sales cycles and customer interactions.

Have the sales executives use Dynamics 365 Sales to track interactions for open opportunities and send follow-up communications to prospects.

Have the sales executives use handsfree headsets to interact with an AI agent when they have questions about internal policies or customer data.

Requirements

Infrastructure Migration

Fabrikam has identified the following infrastructure migration requirements:

Azure must be used for all future infrastructure workloads.
The company must follow Microsoft-recommended methodologies for infrastructure migration to the cloud.
Any created AI agents must have their return on investment (ROI) calculated to ensure that the solution will save the company money.

Sales Cycle Enablement

Fabrikam has identified the following requirements for sales cycle enablement:

The final AI agent must follow Microsoft recommendations for a conversational user experience.
A designated checklist must be reviewed to ensure that the AI agent follows Microsoft deployment recommendations for a compliant solution.
Detailed telemetry must be logged for the first created AI agent to help troubleshoot and optimize the agent during the initial AI agent adoption process.
Unexpected AI agent actions must end in an escalation to a live representative. For example, a sales executive must be rerouted to a representative if the agent cannot answer a question after two failed attempts.
The return on investment (ROI) of switching from the current process to the future process is required for stakeholder sign off.
The sales team must use Dynamics 365 Sales to correspond with prospects more quickly and efficiently than currently.
Sales managers must report on the adoption of the AI agent to key Fabrikam stakeholders on a monthly basis.
Any sensitive information, such as user IDs and names, shared via the AI agent must be tracked for future auditing.

Question 1 💩💩💩

Which tool should you recommend to address the sensitive information concerns in the sales process?

A. the Analytics tab in Microsoft Copilot Studio

B. Model Context Protocol (MCP)

C. Application Insights

D. Microsoft Foundry Tracing UI

E. Monitoring in Microsoft Foundry

✅ 答案：C. Application Insights

原因：

题目明确要求：

Any sensitive information, such as user IDs and names, shared via the AI agent must be tracked for future auditing.

Application Insights 可以记录 Copilot Studio Agent 的 telemetry，并且可以记录包括 userid、name、text、speak 等敏感 activity properties。([Microsoft Learn][1])

因此最符合题目要求的是：

C. Application Insights ✅

🧠 考试记忆：

User ID / Name + telemetry + auditing → Application Insights

另外，这道题和前面的 “Detailed telemetry must be logged” 也是同一个方向： Copilot Studio → Application Insights → detailed telemetry。([Microsoft Learn][2])

正确答案与详细解析

正确答案：

C. Application Insights

解析与考点分析

为什么选 C？

案例需求对照（Fabrikam, Inc. 案例）：
- 案例中明确提出了以下两条针对 AI Agent 遥测与审计的核心需求： "Detailed telemetry must be logged for the first created AI agent to help troubleshoot and optimize the agent..." （必须为第一个 AI Agent 记录详细的遥测数据，以协助排查故障和优化...） "Any sensitive information, such as user IDs and names, shared via the AI agent must be tracked for future auditing." （通过 AI Agent 共享的任何敏感信息（如用户 ID 和姓名）都必须被追踪并记录，以供未来审计。）
微软 Copilot Studio 架构规范：
- Fabrikam 构建的是基于 Copilot Studio 的低代码 AI Agent。
- Copilot Studio 内置的 Analytics 选项卡只提供通用的宏观 KPI 图表（如 CSAT 评分、解决率），不会记录原始的用户敏感信息和底层 Trace 轨迹。
- 要记录详细的遥测数据（Detailed telemetry）、用户 ID（User IDs）、对话原生日志并进行安全审计，Copilot Studio 官方推荐的标准配置是接入 Azure Application Insights。连接后，所有的交互事件、用户 ID、输入文本及错误日志都会安全地存储在底层日志数据库中，以供未来的故障排查与安全审计。

为什么不选其他选项？

A. the Analytics tab in Microsoft Copilot Studio：只展示宏观的统计仪表板，出于隐私保护，默认不提供详细的用户敏感数据追踪和底层原始日志审计。
B. Model Context Protocol (MCP)：是一种用于连接外部数据与工具的通信协议，不具备遥测和审计日志收集功能。
D & E. Microsoft Foundry Tracing UI / Monitoring in Microsoft Foundry：用于 Azure AI Foundry（专业代码 Pro-code）中的大模型监控，而案例中 Fabrikam 构建的是基于 Copilot Studio 的低代码 Agent。

QUESTION 2 💩💩💩

Which existing tool and data should you use to gather the required metrics for stakeholder signoff for the AI agents? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Tool:
- [ ] Microsoft Foundry
- [ ] Azure Resource Monitor (ARM)
- [ ] Dynamics 365 Sales
- [ ] Microsoft Copilot Studio
Data required for the tool:
- [ ] the cumulative time spent on the task over the past year
- [ ] the current cost to complete the tasks per instance
- [ ] the current time to complete the task today per instance
- [ ] the current cost of the Dynamics 365 Sales licenses

QUESTION 答案 (Answer Area)

Tool（工具）: 👉 Microsoft Copilot Studio
Data required for the tool（该工具所需的数据）: 👉 the current time to complete the task today per instance（目前完成单次任务所需的时间）

解析与考点分析

Tool（工具）：Microsoft Copilot Studio
案例背景需求： "Any created AI agents must have their return on investment (ROI) calculated to ensure that the solution will save the company money." "The return on investment (ROI) of switching from the current process to the future process is required for stakeholder sign off." （必须为创建的 AI Agent 计算投资回报率 ROI，且必须获得干系人对 ROI 的签字确认。）
解析：
- 案例中，Fabrikam 使用低代码构建 AI Agent。在微软 AI 体系中，Microsoft Copilot Studio 内置了专门的 Agent Usage Estimator & ROI Calculator（Agent 用量估算与 ROI 计算器工具），专门用于帮助企业评估和计算从旧的人工/传统流程切换到 Copilot AI Agent 后的预期成本节省和投资回报率（ROI），以获取干系人的签字认可。
Data required for the tool（该工具所需的数据）：the current time to complete the task today per instance
估算模型考点：
- 在使用 Copilot Studio 估算工具计算 ROI 时，输入模型的基线核心数据是：单次任务在现有流程中的平均耗时（the current time to complete the task today per instance）。
- 计算逻辑：通过对比“目前单次任务耗时（例如：每次人工查询分散系统需耗时 10 分钟）”与“AI Agent 自动化处理后的耗时（如 30 秒）”，结合预期的月度任务调用量和人工工时成本，估算工具能够精准计算出 AI Agent 每年能为公司节省的总工时、总成本以及消耗的 Copilot Credits，从而得出具体的 ROI 数据供干系人签署。

干扰项排除说明

Tool 干扰项：
- Microsoft Foundry / ARM / Dynamics 365 Sales：均不包含针对 Copilot Studio AI Agent 用量与 ROI 估算的专属计算工具。
Data 干扰项：
- the cumulative time spent over the past year（过去一年的累计总时间）：估算器需要的是单次任务的基线单位耗时（Per instance），而非过去一年的宏观累计时间。
- the current cost of Dynamics 365 Sales licenses（D365 许可证成本）：许可证属于基础设施成本，并非衡量单个业务任务自动化时间节省与 ROI 的直接基线变量。

QUESTION 3

Which tool should you recommend to help secure funding for future AI agent development?

A. Evaluations in Microsoft Foundry

B. the Azure Cost Optimization workbook

C. Azure Operator Insights

D. the Analytics tab in Microsoft Copilot Studio

E. Direct Preference Optimization (DPO)

正确答案与详细解析

正确答案：

D. the Analytics tab in Microsoft Copilot Studio（Microsoft Copilot Studio 中的“分析”选项卡）

解析与考点分析

为什么选 D？

对照案例关键需求（Fabrikam, Inc. 案例）：
- 案例中明确指出了获得后续资金支持的关键条件： "Funding for future AI initiatives will depend on demonstrating an increase in AI adoption month over month." （未来的 AI 项目资金将取决于能否展示 AI 采用率（Adoption）按月增长。）
- 案例中的销售管理层职责： "Sales managers must report on the adoption of the AI agent to key Fabrikam stakeholders on a monthly basis." （销售经理必须每月向关键干系人汇报 AI Agent 的采用率。）
工具功能匹配：
- 在 Microsoft Copilot Studio 中，Analytics（分析）选项卡提供了专门的 Usage（使用情况）和 Adoption（采用率）仪表板。
- 它能够自动统计并可视化呈现 Agent 的月度活跃用户数（MAU）、会话总量、使用趋势增长曲线以及用户参与度。
- 销售经理可以直接导出或截图 Analytics tab 中的月度增长图表，向管理层证明“AI 采用率正在逐月稳步增长”，从而为未来的 AI Agent 开发成功申请并确保资金支持（secure funding）。

为什么不选其他选项？

A. Evaluations in Microsoft Foundry：用于评估专业代码模型输出的质量（如连贯性、幻觉、安全性），不用于展示端到端用户的月度业务采用率（Adoption）指标。
B. the Azure Cost Optimization workbook：用于分析 Azure 云基础设施的成本与节省建议，无法呈现 Copilot Studio Agent 的用户采用率趋势。
C. Azure Operator Insights：针对电信运营商大数据的分析服务，与本案例完全无关。
E. Direct Preference Optimization (DPO)：一种大语言模型微调（Fine-tuning）中的偏好优化算法，属于技术训练手段，而非业务汇报与采用率监控工具。

Correct Answer: D

Explanation/Reference:

Explanation:

Scenario Requirements Any created AI agents must have their return on investment (ROI) calculated to ensure that the solution will save the company money.

-In a Microsoft AI migration, calculating the Return on Investment (ROI) for AI agents is essential for justifying costs and securing future funding. You can utilize the Analytics tab in Microsoft Copilot Studio to track these financial and performance metrics directly.

Calculating ROI with Copilot Studio Analytics The Analytics tab provides a dedicated Savings tile (ROI) that allows you to quantify the impact of your agents:

Define Savings Parameters: You can input estimated time saved (in seconds, minutes, or hours) and money saved per successful agent run.

Real-Time Tracking: Total savings are calculated automatically for your selected period based on successful runs of resolved conversations.

Retroactive Application: If you update your savings estimates, the system can apply these changes to previous runs to provide an accurate historical view.

Granular Insights: Savings can be defined at the overall agent-run level or for specific tools used within a run.

Deploy AI-powered business solutions - Question Set 3

QUESTION 1 💩💩

You are designing a Microsoft Copilot Studio agent that uses a custom Microsoft Foundry model to generate responses.

You need to ensure that the agent can securely connect to and invoke the custom model during user interactions.

What should you include in the design?

A. Configure the agent to use classic orchestration.

B. Create a connection to Microsoft Foundry in the agent.

C. Add the Microsoft Foundry model as a Copilot Studio skill.

D. Create a custom engine agent.

—--

✅ 答案：B. Create a connection to Microsoft Foundry in the agent.

原因：

Copilot Studio Agent 需要通过 connection 安全访问 Microsoft Foundry 中的自定义模型。
A Classic orchestration：不是连接模型的方式。
C Skill：不是用于连接 Foundry custom model。
D Custom engine agent：不符合题目要求。

🧠 记忆：

Copilot Studio → Microsoft Foundry model → Connection

Correct Answer: B

Section: (none)

Explanation/Reference:

Explanation:

To create a connection to Microsoft Foundry within a Microsoft Copilot Studio agent and securely invoke a custom model, follow these steps:

Connect via the "External Agent" Method This method allows your Copilot Studio agent to orchestrate and call a specific Microsoft Foundry agent directly.
Connect via "Bring Your Own Model" (BYOM) If you want to use a specific model from the Foundry model catalog for individual prompt actions.
Security and Governance Connecting to Microsoft Foundry ensures enterprise-grade security through:

Entra ID Integration: Agents are assigned a unique Microsoft Entra Agent ID, allowing IT teams to enforce Conditional Access and Identity Governance.

Unified Management: Both your Copilot Studio and Foundry agents can be managed centrally in the Microsoft Entra admin center.

QUESTION 2 💩

You are designing an AI business solution that contains the following components:

A Microsoft Power Automate workflow
A Microsoft Copilot Studio agent
A Microsoft Dataverse database
A Microsoft Power Apps app

As part of the application lifecycle management (ALM) process, you plan to package the components, so that they can be deployed to other environments as a group.

You need to recommend a solution that supports versioning, dependencies, and deployments.

What should you include in the recommendation?

A. GitHub Actions

B. Azure DevOps

C. Microsoft Power Platform solutions

正确答案：✅ C. Microsoft Power Platform solutions

原因

题目关键词：

Power Automate
Copilot Studio agent
Dataverse
Power Apps

打包成一个整体

Versioning
Dependencies
Deploy to other environments

这些正是 Power Platform Solutions 的用途。

Solution 可以把多个 Power Platform 组件作为一个逻辑单元管理，并支持：

版本管理
组件依赖关系
Dev → Test → Production 部署
Managed / Unmanaged Solutions

其他选项

❌ GitHub Actions → CI/CD 自动化工具，不是 Power Platform 组件的打包容器。
❌ Azure DevOps → 可以用于 ALM Pipeline，但不是组件本身的打包和依赖管理机制。

🎯 最终答案

C. Microsoft Power Platform solutions

记忆：

Power Apps + Power Automate + Copilot Studio + Dataverse → Solution

Solution + CI/CD → Power Platform Pipelines / Azure DevOps / GitHub Actions

Correct Answer: C Section: (none)

Explanation/Reference:

Explanation:

To package your AI business solution components—including the Power Automate workflow, Copilot Studio agent, Dataverse database, and Power Apps app—you should use Microsoft Power Platform solutions.

While GitHub Actions and Azure DevOps are powerful tools for automating the deployment process (CI/ CD), Solutions are the actual "containers" or packaging mechanism required to bundle these specific Power Platform assets together so they can be moved as a single unit between environments.

QUESTION 3

A company has Microsoft 365 Copilot agents.

You need to design a security solution for the agents. The solution must meet the following requirements:

Identify and mitigate potential risks that relate to AI use.
Protect AI apps and the sensitive data processed or generated by the agents.
Support responsible AI governance by retaining and logging interactions, detecting policy violations, and investigating incidents.

Which two components should you include in the design? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. Microsoft Purview

B. Azure AI Content Safety

C. role-based access control (RBAC) in Microsoft Foundry

D. Microsoft Defender

正确答案与详细解析

正确答案：

A. Microsoft Purview
D. Microsoft Defender

为什么选 A 和 D？

在微软的 AI 安全与治理统一框架（Security for AI）中，Microsoft Purview 与 Microsoft Defender 构成了针对 Microsoft 365 Copilot 及 AI Agent 的双壁核心安全解决方案：

Microsoft Purview（数据治理、合规与日志追踪，选 A）：
- 满足需求：保留和记录交互（Retaining and logging interactions）、检测策略违规（Detecting policy violations）、保护敏感数据（Protect sensitive data）。
- 功能作用：
  - Purview AI Hub / Audit（审计）：自动记录并保留用户与 AI Copilot/Agent 交互的提示词（Prompts）和生成回答（Responses）日志，供合规审查与事件调查（eDiscovery）。
  - Data Loss Prevention (DLP) & Information Protection：识别并保护 Agent 处理或生成的敏感数据（如防止敏感数据泄露到 Agent 回答中），检测违反企业策略的操作。
Microsoft Defender（AI 应用威胁防护与风险缓解，选 D）：
- 满足需求： 识别并缓解与 AI 相关的潜在风险（Identify and mitigate AI risks）、保护 AI应用程序（Protect AI apps）、安全事件调查（Investigating incidents）。
- 功能作用：
  - Defender for Cloud / Defender for Cloud Apps：提供针对 AI 应用和 AI 工作负载的全面安全防护（AI Security Posture Management, AISPM）。
  - 监控针对 AI 智能体和应用的恶意威胁、影子 AI（Shadow AI）使用风险、越狱（Jailbreak）攻击以及安全事件响应与调查。

为什么不选其他选项？

B. Azure AI Content Safety（Azure AI 内容安全）：侧重于生成式 AI 文本/图像内容的中台实时审核（如过滤暴恐、仇恨言论或实时越狱检测），但不提供日志长期保留、数据防泄漏（DLP）、合规审查或跨系统安全事件调查。
C. role-based access control (RBAC) in Microsoft Foundry（Microsoft Foundry 中的 RBAC）：仅用于控制开发者在 Azure AI Foundry 平台内部的底层项目权限，无法提供全局的 M365 Copilot Agent 交互日志保留、敏感数据治理或合规事件调查。

Section: (none)

Explanation/Reference:

Explanation:

To secure Microsoft 365 Copilot agents, you can implement a unified solution using Microsoft Purview for data governance and Microsoft Defender for real-time threat protection. This combination identifies risks, protects sensitive data, and ensures responsible AI governance through continuous monitoring and incident investigation.

QUESTION 4 💩

You are creating validation criteria for a custom generative AI model that produces business reports based on internal enterprise data.

You need to assess whether the model's outputs are appropriate and meaningful for the business reports.

Which metric should you use?

A. the number of active users interacting with the model

B. alignment of the output to domain-specific tasks

C. the average system resource usage during inference

D. the model training duration

正确答案：

B. alignment of the output to domain-specific tasks（输出与特定领域任务的对齐度）

解析与考点分析

为什么选 B？

1 题目要求： 需要评估自定义生成式 AI 模型输出的商业报告是否合适且有意义（appropriate and meaningful for the business reports）。

2 生成式 AI 的领域评估标准：

企业内部的商业报告通常包含特定行业的术语、业务逻辑、合规要求和特定的格式（即领域特定知识，Domain-specific knowledge）。
评估生成的报告是否“合适”且“有意义”，核心指标就是评估模型的输出是否与特定领域任务的要求相匹配/对齐（alignment to domain-specific tasks），包括内容的准确性、专业相关性以及是否符合业务规范。

为什么不选其他选项？

A. the number of active users interacting with the model（与模型交互的活跃用户数量）： 这是产品采用率/用户活跃度指标（Product Adoption/Usage Metric），无法反映模型生成的报告内容本身是否准确或有意义。
C. the average system resource usage during inference（推理期间的平均系统资源使用率）： 这是系统性能/运维指标（Infrastructure Performance Metric），关注的是 CPU/GPU/内存消耗和成本，与生成文本的质量无关。
D. the model training duration（模型训练时长）： 这是模型训练阶段的技术指标（Training Metric），反映的是算力消耗和训练效率，无法衡量训练出来的模型输出质量。

QUESTION 5

A company has Microsoft Foundry agents that generate responses by using Azure OpenAI resources. The agents are deployed to both the United States and Europe.

A company mandate states that the agents and their grounding data must adhere to data residency and movement regulations.

You need to recommend a governance solution for the agents.

What should you include in the recommendation?

A. Microsoft Defender for Cloud

B. Azure Policy

C. Azure Monitor

D. Microsoft Purview

正确答案与详细解析

正确答案：

D. Microsoft Purview

解析与考点分析

为什么选 D？

核心需求：公司部署在跨国区域（美国和欧洲）的 AI Agent 及其接地数据（Grounding data）必须符合数据驻留（Data residency）和跨国数据流动法规（Data movement regulations）（如 GDPR、欧洲数据边界 EU Data Boundary 等）。
产品功能匹配：
- Microsoft Purview：是微软企业级数据治理、风险管理与合规控制的核心平台。
- 针对 AI 与 Grounding Data 的治理：
  - Purview AI Hub & Data Map：专门用于分类、跟踪和监控 AI 智能体调用的接地数据资产，跟踪数据血缘与数据流动路径。
  - 数据驻留与跨国传输合规：检测并限制敏感数据的跨国界流动（Data movement），确保接地数据在处理和回答生成过程中严格遵循当地的数据驻留法规。

为什么不选其他选项？

A. Microsoft Defender for Cloud（云安全态势管理）：侧重于基础设施与 AI 应用的云安全威胁防护、漏洞扫描与攻击防御，不负责数据层面的法律合规与跨国数据流动治理。
B. Azure Policy（Azure 策略）：侧重于基础设施资源（Resource Level）的部署位置限制（如只能在特定 Azure 区域创建 Azure OpenAI 资源），但无法治理和管控数据内容本身及其流向（Grounding Data Movement）。
C. Azure Monitor（Azure 监控）：用于收集系统运行日志、性能指标与告警，不具备数据治理和合规控制功能。

✅ 正确答案：D. Microsoft Purview

关键在题目新增的 requirement：

the agents and their grounding data must adhere to data residency and movement regulations

这里的核心不是简单的 Azure resource configuration，而是：

Data residency：数据必须存放在哪个 geographic region
Data movement：数据是否可以跨 region / jurisdiction 移动
Grounding data：Agent 使用的数据本身
Regulations：数据相关的法规和合规要求

这些属于 data governance / compliance 范畴，因此 Microsoft Purview 更合适。

为什么不是 Azure Policy？

Azure Policy 非常适合：

“资源只能部署在 approved regions。”

例如：

Azure Policy
    ↓
Allowed locations
    ↓
Azure OpenAI resource
    ↓
East US / West Europe only

这是资源部署治理。

但本题要求的是：

Agent
  ↓
Grounding data
  ↓
Data residency
  ↓
Data movement
  ↓
Regulatory compliance

关注的是 data 本身的治理和合规，所以选择 Microsoft Purview。

四个选项怎么区分

Service	主要用途	本题
Microsoft Defender for Cloud	Security posture / workload protection	❌
Azure Policy	Resource governance / allowed locations / configuration compliance	❌
Azure Monitor	Metrics / logs / alerts / monitoring	❌
Microsoft Purview	Data governance / data residency / data compliance / data movement	✅

Section: (none)

Data Loss Prevention (DLP): By integrating Purview DLP policies, organizations can monitor, block, or warn when sensitive data is used in AI prompts or responses in real-time.

Data Residency Compliance: For strict European residency (e.g., GDPR), Azure OpenAI resources should be deployed using Data Zone (DZ) SKUs (such as in Sweden Central or Germany West Central), which contractually guarantee that both data storage and processing remain within the specified geography.

Embedded Governance: Admins can enable a native integration within Microsoft AI Foundry at the subscription level. This automatically sends prompt and response data to Purview for auditing and compliance without requiring additional developer code.

QUESTION 6

A company has a Microsoft Copilot Studio agent that uses custom connectors to interact with enterprise APIs.

You need to recommend an application lifecycle management (ALM) process to ensure that the connectors are deployed consistently across development, test, and production environments and meet governance and traceability requirements.

What should you recommend?

A. Deploy the APIs as Azure Functions.

B. Manage the connectors as solution components and deploy the components by using ALM pipelines.

C. Maintain connector definitions in environment variables.

D. Export and import the connectors between the environments as unmanaged solutions.

正确答案与详细解析

正确答案：

B. Manage the connectors as solution components and deploy the components by using ALM pipelines.（将连接器作为解决方案组件进行管理，并使用 ALM 管道部署这些组件。）

Development
    │
    │ Custom Connector
    ↓
Solution
    │
    ↓
ALM Pipeline
    │
    ├──────────────→ Test
    │
    └──────────────→ Production

这样可以获得：

Consistent deployment：不同环境使用同一套受控组件
Version control / traceability
Managed deployment
Governance
减少手工 export/import

解析与考点分析

为什么选 B？

业务需求：需要确保自定义连接器（Custom Connectors）在开发（Dev）、测试（Test）和生产（Prod）环境中能够一致性地部署（Deployed consistently），并且满足治理与可追溯性要求（Governance and traceability）。
微软 ALM 最佳实践：
- 解决方案组件化（Solution Components）：在 Power Platform / Copilot Studio 体系中，将自定义连接器包含在解决方案包（Solution）中，使其支持版本控制与依赖管理。
- ALM 流水线部署（ALM Pipelines）：使用自动化的 ALM 管道（如 Power Platform Pipelines、Azure DevOps 或 GitHub Actions）将解决方案（在目标环境部署为托管解决方案 Managed Solutions）自动分发到测试和生产环境。
- 自动化管道能够提供完整的审计日志、版本发布轨迹（Traceability），避免人工手动导入操作，满足企业级治理规范。

为什么不选其他选项？

A. Deploy the APIs as Azure Functions（将 API 部署为 Azure Functions）： Azure Functions 用于托管后端 API 的代码逻辑，但并不能解决 Power Platform 层面的自定义连接器打包、跨环境 ALM 部署和治理问题。
C. Maintain connector definitions in environment variables（在环境变量中维护连接器定义）：环境变量（Environment Variables）用于存储连接器的环境差异化配置（如目标 API 的 Base URL），但它本身不能替代自定义连接器组件的打包和 ALM 部署。
D. Export and import the connectors between the environments as unmanaged solutions（导出并作为非托管解决方案导入）：严重违规！在测试和生产环境中导入非托管解决方案（Unmanaged Solutions）违背了 ALM 治理规范，会导致生产环境代码混乱且缺乏版本掌控与追溯性（生产环境必须部署托管解决方案）。

Correct Answer: C

Explanation/Reference:

Explanation:

To implement a robust Application Lifecycle Management (ALM) process for Microsoft Copilot Studio agents using custom connectors, you must leverage Solutions and Environment Variables to ensure consistency and governance.

Solution-Based Management

Always develop your custom connectors and Copilot Studio agents within a Power Platform Solution. This allows you to package all related components (connectors, environment variables, and agents) as a single unit for deployment across your ALM pipeline (Dev → Test → Prod).

Implementing Environment Variables

To maintain consistent definitions and meet governance requirements, externalize environment-specific values using Environment Variables:

Host and Base URL: Use environment variables for the API endpoint (e.g., api.dev.enterprise.com vs. api.prod.enterprise.com) to avoid manual edits during deployment.

Security Credentials: For OAuth settings like Client ID and Client Secret, use environment variables with the Secret data type.

Azure Key Vault Integration: For high-security enterprise APIs, store secrets in Azure Key Vault and reference them via secret environment variables to ensure traceability and governance.

QUESTION 7 💩💩

A company plans to implement an AI solution that will contain a Microsoft Copilot Studio agent and a Microsoft Foundry agent.

The solution will be stored in a source code repository.

You need to recommend a deployment method for each agent. The solution must meet the following requirements:

A test environment must be used before a deployment to production. Production must be isolated from development and testing.

The deployment must be repeatable and fully automated.

The solution must NOT require manual intervention.

Which deployment method should you recommend for each agent? To answer, select the appropriate options in the answer area.

Answer Area

Copilot Studio:
- [ ] Export from the source code repository and import to the target environment.
- [ ] Use a Bicep file.
- [ ] Use a Microsoft Power Platform deployment pipeline.
Microsoft Foundry:
- [ ] Use a Bicep file.
- [ ] Use a Microsoft Power Platform deployment pipeline.
- [ ] Use an Azure DevOps pipeline.

正确答案：

Copilot Studio: ✅ Use a Microsoft Power Platform deployment pipeline
Microsoft Foundry: ✅ Use an Azure DevOps pipeline

原因

Agent	推荐方法	原因
Copilot Studio	Power Platform deployment pipeline	支持 Dev → Test → Prod 的环境晋级和隔离，适合自动化 Solution/Agent 部署。
Microsoft Foundry	Azure DevOps pipeline	适合将 Foundry AI 资源/模型部署纳入 CI/CD，实现可重复、自动化部署。

Microsoft Foundry → Azure DevOps pipeline ✅

Microsoft Foundry agent 属于 Azure / Microsoft Foundry 资源体系。

题目特别告诉你：

The solution will be stored in a source code repository.

而且要求：

repeatable and fully automated

因此适合使用 Azure DevOps pipeline 实现 CI/CD：

Source Code Repository
        │
        ↓
Azure DevOps Pipeline
        │
        ├── Build / Validate
        ↓
      Test
        │
        ↓
   Production

Pipeline 可以把 deployment 定义成代码，并自动执行 Dev → Test → Prod promotion。

因此：

Microsoft Foundry → Use an Azure DevOps pipeline

为什么其他选项不选？

Copilot Studio

❌ Export/import → 通常需要人工操作，不符合 fully automated。
❌ Bicep → 主要用于 Azure infrastructure，不是 Copilot Studio Agent 的主要 ALM 部署方式。

Microsoft Foundry

❌ Power Platform deployment pipeline → 面向 Power Platform。
❌ Bicep → 可以部署 Azure 资源，但题目强调源代码仓库 + repeatable automated deployment，Azure DevOps pipeline 更合适。

🎯 最终

Copilot Studio → Microsoft Power Platform deployment pipeline

Microsoft Foundry → Azure DevOps pipeline

Copilot Studio: 👉 Use a Microsoft Power Platform deployment pipeline.（使用 Microsoft Power Platform 部署管道。）
Microsoft Foundry: 👉 Use an Azure DevOps pipeline.（使用 Azure DevOps 管道。）

Explanation/Reference:

Explanation:

Box 1: Use a Microsoft Power Platform deployment pipeline

Box 2: Use a Bicep file

Box 1: Use a Microsoft Power Platform deployment pipeline

Copilot Studio agent The preferred deployment method is to use a Microsoft Power Platform deployment pipeline.

Microsoft Power Platform pipelines are specifically designed to meet your requirements for a secure, automated, and repeatable application lifecycle management (ALM) process for Copilot Studio agents: Box 2: Use a Bicep file Microsoft Foundry agent In the scenario described for a Microsoft Foundry agent, the preferred deployment method is to use a Bicep file.

This approach is best suited for your requirements because:

Infrastructure as Code (IaC): Azure Bicep allows you to define your entire environment—including the Foundry hub, projects, and model deployments—as code stored in your source repository.

Automation & Repeatability: Bicep files integrate directly with GitHub Actions or Azure Pipelines, enabling fully automated, repeatable deployments without manual intervention.

Environment Isolation: You can use Bicep to provision distinct, isolated resources for development, testing, and production by parameterizing the deployment for each environment.

Suitability: While Power Platform pipelines are used for Copilot Studio agents, Foundry-based agents are Azure resources where Bicep is the native and more powerful automation tool for managing the underlying infrastructure and model endpoints

Box 2: Use a Bicep file

Microsoft Foundry agent

In the scenario described for a Microsoft Foundry agent, the preferred deployment method is to use a Bicep file.

This approach is best suited for your requirements because:

Infrastructure as Code (IaC): Azure Bicep allows you to define your entire environment—including the Foundry hub, projects, and model deployments—as code stored in your source repository.

Automation & Repeatability: Bicep files integrate directly with GitHub Actions or Azure Pipelines, enabling fully automated, repeatable deployments without manual intervention.

Environment Isolation: You can use Bicep to provision distinct, isolated resources for development, testing, and production by parameterizing the deployment for each environment.

Suitability: While Power Platform pipelines are used for Copilot Studio agents, Foundry-based agents are Azure resources where Bicep is the native and more powerful automation tool for managing the underlying infrastructure and model endpoints.

QUESTION 8 💩💩

A company has a Microsoft Copilot Studio prompt-and-response agent. You need to ensure that the agent meets the following requirements:

Provides effective and relevant responses
Provides conversational outcomes

Which metric should you use for each requirement? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Provides effective and relevant responses:

Generated answer rate and quality
Reactions
Tool use

Provides conversational outcomes:

Satisfaction
Tool use
Topics by outcome

✅ 答案

Provides effective and relevant responses: → Generated answer rate and quality
Provides conversational outcomes: → Topics by outcome

🧠 记忆：

Answer quality → Generated answer rate and quality

Conversation outcome → Topics by outcome

Provides effective and relevant responses:
- 正确选项： Generated answer rate and quality（生成答案的比例与质量）
Provides conversational outcomes:
- 正确选项： Topics by outcome（按结果分类的主题）

解析与考点分析

第一部分：提供有效且相关的回答（Provides effective and relevant responses）

考点： Copilot Studio 生成式 AI (Generative Answers) 的分析指标。
解析：
- 在 Copilot Studio 中，当 Agent 使用提示词（Prompts）和生成式回答时，评估其回答是否有效且相关（effective and relevant）的核心指标是 Generated answer rate and quality。
- 该指标监控生成式回答触发的成功率（Generated answer rate）以及模型输出内容的相关度与质量（Quality），用于衡量 RAG / Prompt 回答的效果。

第二部分：提供对话业务结果（Provides conversational outcomes）

考点： Copilot Studio 主题与会话结果（Topic Outcomes）分析。
解析：
- 在 Copilot Studio 的 Analytics -> Topic Analytics 仪表板中，评估对话是否达成了明确的业务结果（Conversational outcomes）的标准指标是 Topics by outcome。
- 该指标会按照具体的对话主题（Topics）对对话结果进行分类统计，展示哪些 Topic 达到了已解决（Resolved）、转人工（Escalated）或已放弃（Abandoned）的结果，从而帮助分析对话流的最终转化率。

QUESTION 9

A company extends Copilot in Microsoft Dynamics 365 Customer Service.

You need to recommend an automated application lifecycle management (ALM) process so that the Copilot components can be safely developed, tested, and promoted to production.

Which two actions should you include in the ALM process? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. Use an unmanaged solution in production.

B. Rebuild the agents in each environment.

C. Use Microsoft Power Platform pipelines.

D. Include the components in a solution.

E. Store the agent transcripts in source control.

正确答案：✅ C、D

C. Use Microsoft Power Platform pipelines. → 自动化 Dev → Test → Production 的部署和 Promotion。
D. Include the components in a solution. → 将 Copilot 相关组件统一打包、版本化并管理依赖关系。

为什么其他选项不对？

❌ A → Production 应使用 managed solution，不是 unmanaged。
❌ B → 不应在每个环境手工重建 Agent。
❌ E → Agent transcripts 是运行数据，不是 ALM 组件，不应作为源代码管理。

记忆：

Copilot ALM = Solution + Power Platform Pipelines

答案：C + D

Correct Answer: CD Section: (none)

Explanation/Reference:

Explanation:

To implement an automated Application Lifecycle Management (ALM) process for extending Microsoft Dynamics 365 Customer Service with Copilot, you should leverage Power Platform solutions and pipelines. This approach ensures that custom agents, knowledge sources, and connector actions are developed and promoted safely across environments.

Tools used include:

Power Platform Pipelines: Automates the deployment process across environments.

To automate the application lifecycle management (ALM) for Copilot components within Microsoft Dynamics 365 Customer Service, follow this structured process using Power Platform pipelines and solutions:

Solution-Centric Development

All Copilot components (agents, topics, custom actions, and connector plugins) must be developed within a Power Platform solution.

Create or Select a Solution: In Copilot Studio or the Power Apps maker portal, ensure you are working within an unmanaged solution in your Development environment.

Add Components: When you create new agents or actions in Microsoft Copilot Studio, they are automatically associated with the preferred solution you have set, ensuring they are portable.

Pipeline Configuration

Power Platform pipelines democratize ALM by providing a built-in CI/CD experience directly within the maker portal.

Incorrect:

[Not A] Use Managed Solutions for Production: Export and deploy solutions as managed when promoting to test or production environments to prevent unauthorized direct changes in those environments and ensure a consistent state.

QUESTION 10 💩

You are designing a testing solution for a Microsoft Copilot Studio agent that integrates with Microsoft Dynamics 365 Customer Service and Dynamics 365 Sales.

You need to design end-to-end scenarios to test the agent's ability to perform the following actions:

Coordinate tasks and data interactions across both Dynamics 365 apps.
Interpret user input and provide contextually relevant outputs.

Which test scenario and metric should you include in the design? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Test scenario:

In each app, test isolated tasks without using workflows.
Run task-based scenarios that involve both apps.
Test visual consistency across both apps.

Metric:

Measure the initial prompt response time for each app.
Track the average click rate across both apps.
Track the successful completion of cross-app tasks.

✅ 答案

Test scenario: → Run task-based scenarios that involve both apps.
Metric: → Track the successful completion of cross-app tasks.

🧠 记忆：

Cross-app coordination → Test both apps together

Task success → Successful completion rate

Test scenario（测试场景）:
- 正确选项： Run task-based scenarios that involve both apps.（运行涉及这两个应用的基于任务的场景。）
Metric（测试指标）:
- 正确选项： Track the successful completion of cross-app tasks.（追踪跨应用任务的成功完成率。）

解析与考点分析

第一部分分析：Test scenario（测试场景）

需求关键点：测试 Agent 跨 Dynamics 365 Customer Service 和 Dynamics 365 Sales 协调任务与数据交互的能力（Coordinate tasks and data interactions across both Dynamics 365 apps）。
解析：
- 要验证 Agent 是否能跨不同系统进行数据同步和任务协同（例如：在客服模块中接收问题后，自动在销售模块中创建潜在客户记录），必须设计并执行端到端（End-to-End）的涉及两个应用的多步骤任务场景（Run task-based scenarios that involve both apps）。
- 隔离单应用测试（In each app, test isolated tasks）无法验证跨应用交互能力；视觉一致性测试（Visual consistency）测试的是 UI 界面样式，与 Agent 的业务自动化逻辑无关。

第二部分分析：Metric（测试指标）

需求关键点：评估跨应用协同与意图理解的实际效果。
解析：
- Track the successful completion of cross-app tasks（追踪跨应用任务的成功完成率）：是衡量端到端集成测试成功与否最关键的业务指标。它直接验证了 Agent 是否准确理解了用户的指令，并在两个 Dynamics 365应用中成功完成了全流程的数据更新和任务结案。
- 响应时间（Response time）仅衡量网络延时，无法衡量任务是否成功；点击率（Click rate）属于前端网页分析指标，不适用于 Agent 业务自动化测试。

Box 1: Run task-based scenarios that involve both apps

Box 2: Track the successful completion of cross-app tasks

QUESTION 11

A company has multiple AI models that support generation of sales transactions.

Each release of the models must be reviewed by a security and compliance team before being deployed to the production environment.

The security and compliance team must have access to prior versions to properly determine potential exposures introduced*.

You need to recommend a solution to evaluate the impact of each deployment to production. The solution must enhance business continuity.

What should you recommend?

A. Create a central model registry that uses version history.

B. Establish a promotion process by using a quality gate.

C. Implement version control for all the AI system components.

D. Track model retirement schedules to prevent service disruptions.

✅ 答案：A. Create a central model registry that uses version history.

原因：

题目重点是：

每个模型 release 都要经过 security/compliance review
审核团队需要访问 prior versions
评估新版本部署到 production 的影响
提高 business continuity

Central model registry + version history 可以保存不同模型版本，方便：

比较新旧模型
审查潜在安全风险
必要时回滚到之前版本
支持生产环境的连续性

🧠 记忆：

Model releases + prior versions + impact review + rollback → Model registry + version history

答案：A ✅

A. Create a central model registry that uses version history.（创建一个使用版本历史记录的中央模型注册表。）

核心考点速记（备考复习）

中央模型注册表（Central Model Registry）的核心作用：在 MLOps 和 AI 安全合规治理中，当安全团队需要在模型部署前审查新版本并访问调阅以前的模型历史版本（Access to prior versions）时，中央模型注册表（Model Registry）是提供模型版本管理（VersionHistory）、血缘追溯、安全审计以支持生产快速回滚（保障业务连续性）的标准组件。

QUESTION 12

A company has an AI solution that uses a Microsoft Copilot Studio agent.

You need to monitor the agent's performance. The solution must meet the following requirements:

Monitor the agent's telemetry in near-real-time (NRT).
Download transcripts of full conversations.
Monitor the agent's usage and performance.

What should you use for each requirement? To answer, drag the appropriate options to the correct requirements. Each option may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

NOTE: Each correct selection is worth one point.

Options

Application Insights
Copilot Studio
Log Analytics
Microsoft Power Apps

Answer Area

Monitor the agent's telemetry in NRT:
Download transcripts of full conversations:
Monitor the agent's usage and performance:

Options

✅ 答案

Monitor the agent's telemetry in NRT: → Application Insights
Download transcripts of full conversations: → Copilot Studio
Monitor the agent's usage and performance: → Copilot Studio

🧠 记忆：

NRT telemetry → Application Insights

Transcripts → Copilot Studio

Usage & performance → Copilot Studio

Monitor the agent's telemetry in NRT (Near-Real-Time): 👉 Application Insights
Download transcripts of full conversations: 👉 Copilot Studio
Monitor the agent's usage and performance: 👉 Copilot Studio

(注：题目明确提示“Each option may be used once, more than once, or not at all”，因此 Copilot Studio 被使用了两次。)

解析与考点分析

实时/近实时（NRT）监控 Agent 的遥测数据
- 正确选择： Application Insights
- 解析：
- Application Insights 与 Copilot Studio 原生集成。当连接后，Copilot Studio 会将底层系统事件、API 调用耗时、异常及实时 Trace 轨迹以近实时（Near-Real-Time, NRT）流的形式推送到 Application Insights 中，用于实时性能监控与故障排查。
下载完整对话记录（Transcripts of full conversations）
- 正确选择： Copilot Studio
- 解析：
- 在 Copilot Studio 门户的 Analytics（分析）-> Sessions（会话）选项卡中，管理员和开发者可以直接查看所有已发生的对话列表，并直接下载（Download）完整的对话轨迹（Transcripts）文件进行内容审查与质量分析。
监控 Agent 的使用率与性能（Usage and performance）
- 正确选择： Copilot Studio
- 解析：Copilot Studio 内置了开箱即用的仪表板（Analytics Tab），可直接提供 Agent 的使用率（Usage，如活跃用户数、会话总数、Credits 消耗趋势）和性能指标（Performance，如解决率 Resolution Rate、转人工率 Escalation Rate、CSAT 满意度评分）。成本最低且无需额外开通云服务。

QUESTION 13

A company deploys a Microsoft Copilot Studio agent that integrates with a Microsoft Power Automate desktop flow.

You need to recommend a testing solution that meets the following requirements:

Test cases must validate the most recent changes to the agent before the agent is released.
The flow must be validated as part of the agent's orchestration.

What should you recommend for each requirement? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Answer Area

Validate the most recent changes to the agent before release:

Publish the agent to a channel and test the agent on live users.
Run tests against the latest unpublished version of the agent.
Run tests against the production version of the agent.

Validate the flow as part of the agent's orchestration:

Add the flow to the agent as a tool.
Add the flow to a canvas app.
Use the Power Automate for desktop console.

正确答案

Requirement	答案
Validate the most recent changes to the agent before release	✅ Run tests against the latest unpublished version of the agent.
Validate the flow as part of the agent's orchestration	✅ Add the flow to the agent as a tool.

原因

1. Latest unpublished version → Run tests

Copilot Studio 可以在 Agent 发布之前测试最新修改，这样可以验证尚未发布的 changes，而不影响 production。

2. Flow orchestration → Add flow as a tool

题目要求：

flow must be validated as part of the agent's orchestration

因此需要把 Power Automate flow 作为 Agent 的 Tool，让 Agent 在测试过程中实际调用该 flow。

🎯 考试记忆

Before release → Test latest unpublished version Agent + Power Automate flow → Add flow as a Tool

**答案：① Run tests against latest unpublished version

② Add the flow to the agent as a tool**

Validate the most recent changes to the agent before release:

正确选项： Run tests against the latest unpublished version of the agent.（针对最新的未发布版本 Agent 运行测试。）

Validate the flow as part of the agent's orchestration:

正确选项： Add the flow to the agent as a tool.（将 Flow 作为工具添加给 Agent。）

解析与考点分析

第一部分分析：在发布前验证 Agent 的最新更改

考点： Copilot Studio 的版本与测试机制。
解析：
- 未发布版本（Unpublished Version/Draft）：在 Microsoft Copilot Studio 中，做出的任何修改都是保存在未发布的草稿状态（Draft）中。Copilot Studio 内置的测试面板（Test Canvas）或自动化测试套件（如 Power CAT Copilot Studio Kit）可以让你直接针对未发布（Unpublished）的版本运行测试用例。这能确保在正式Publish（发布）给终端用户之前验证最新更改。
- 错误选项排除：
  - Publish the agent to a channel and test the agent on live users：直接发布给线上真实用户测试属于违规操作，无法控制风险。
  - Run tests against the production version of the agent：生产环境运行的是旧的已发布版本，无法测试到“最新未发布的修改”。

第二部分分析：验证 Flow 作为 Agent 编排的一部分

考点： Copilot Studio 的 Generative Orchestration（生成式编排）与 Action/Tool（工具）集成。
解析：
- 将 Flow 作为 Tool 添加：在 Copilot Studio 中，要让 AI Agent 在编排逻辑（Orchestration）中调用 Power Automate（无论是 Cloud Flow 还是 Desktop Flow），必须将其注册为 Agent 的 Tool（工具）或 Action（动作）。这样，Agent 的生成式编排器（Orchestrator）才能在运行时识别该 Flow 的功能描述（Description），并自动决定何时触发该 Desktop Flow。添加为 Tool 后，才能在 Agent 的整体编排中对其进行端到端测试。
- 错误选项排除：
  - Add the flow to a canvas app：这是在 Canvas App 中测试 Flow，无法测试 Copilot Studio Agent 的 AI 编排逻辑。
  - Use the Power Automate for desktop console：控制台只能单独测试 Desktop Flow 本身是否能正常运行，无法测试 Agent 编排（Orchestration）调用 Flow 的过程。

QUESTION 14

A company has a Microsoft Copilot Studio agent that provides answers based on a knowledge base for customer support.

Users report that, occasionally, the agent provides inaccurate answers.

You need to use metrics from the Analytics tab in Copilot Studio to identify the cause of the inaccuracies.

Which two options should you use? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. survey results

B. session information and session outcomes

C. topic usage and topics with low resolution

D. engagement, resolution, and escalation rates

E. quality of generated answers

正确答案与详细解析

正确答案：

C. topic usage and topics with low resolution（主题使用情况及低解决率的主题）
E. quality of generated answers（生成式回答的质量）

解析与考点分析

为什么选 C 和 E？

题目背景是：Agent 基于知识库回答问题，但用户反映偶发性提供不准确的回答（inaccurate answers）。我们需要通过 Copilot Studio 中的 Analytics（分析）选项卡来定位导致回答不准确的具体原因。

C. topic usage and topics with low resolution（选 C）：
- 作用：在 Copilot Studio 的 Analytics -> Topic Analytics（主题分析）中，系统会列出所有 Topic 的触发频率和解决率。
- 分析原因： 低解决率（Low resolution）或高转人工率（High escalation）的主题通常意味着该主题包含不准确、过期或有误导性的回答。作者可以通过筛选低解决率主题，直接调出具体的对话轨迹（Transcripts），定位是哪些答案导致了问题。
E. quality of generated answers（选 E）：
- 作用：由于 Agent 是基于知识库（Knowledge Base）提供回答的，这使用了 Copilot Studio 的生成式回答（Generative Answers）功能。
- 分析原因：在 Analytics -> Generative AI（生成式 AI 分析）选项卡中，系统专门记录了生成式回答的质量（Quality of generated answers），包括用户点赞/点踩（Thumbs up / Thumbs down）的反馈、知识源检索失败率等。这能直接定位是否是知识库文档本身内容不准确或 AI 生成回答时产生了幻觉。

为什么不选其他选项？

A. survey results（调查结果）：仅显示用户满意的宏观分数（如 CSAT评分），能告诉你用户“不满意”，但无法告诉你具体是“哪一条知识或哪个回答不准确”。
B. session information and session outcomes（会话信息与会话结果）：提供整体会话的状态分类（Resolved/Escalated/Abandoned），属于宏观运营数据，无法精细化定位内容准确度问题。
D. engagement, resolution, and escalation rates（参与率、解决率和升级率）：属于 Analytics 首页（Summary）上的顶层宏观 KPI 指标，用于衡量整体运行健康度，无法精细定位到具体的错误回答内容。

Correct Answer: CD

Explanation/Reference:

Explanation:

[C] You can use the Analytics dashboard to identify and fix inaccurate responses by focusing on specific performance signals:

*-> Topics with Low Resolution: This metric identifies individual topics where sessions frequently end without a successful outcome. A low resolution rate often indicates that the topic's logic is flawed, the content is outdated, or it fails to address the user's specific intent.

*-> Topic Usage: High usage of certain topics combined with low resolution or high escalation rates points to critical knowledge gaps. Use this to prioritize which parts of your knowledge base need immediate refinement.

Unrecognized Utterances: Review these to find user phrases that didn't trigger any custom topic, indicating missing content or the need for new trigger phrases.

Analyze User Questions by Theme: Copilot Studio uses AI to group generative responses into themes. You can review these clusters to see which themes have poor response quality and need better grounding or data hygiene.

User Feedback (Thumbs Up/Down): Review specific messages with negative feedback in the Analytics tab to understand exactly where the AI is hallucinating or providing incomplete data.

[D] In Microsoft Copilot Studio, you can use analytics to pinpoint why your agent is providing inaccurate answers by following a structured improvement checklist.

Using Metrics to Locate Problems *-> Engagement Rate: A low engagement rate often indicates that your triggers are misconfigured or too broad, causing the agent to initiate the wrong topic or fail to recognize user intent entirely.

*-> Resolution Rate: Identify specific topics with low resolution. If a topic has a high volume of sessions but fails to reach a "Resolved" state, it usually means the knowledge source is outdated, mismatched, or the generative answers are not grounded properly.

*-> Escalation Rate: High escalation rates for particular topics are "red flag" drivers. Use the Escalation Rate Drivers chart to see which topics most frequently force a hand-off to a human agent, indicating where the AI's knowledge or logic is insufficient.

QUESTION 15

A company uses a fine-tuned Microsoft Foundry model that requires frequent updates as new customer feedback becomes available.

You need to design an application lifecycle management (ALM) process that meets the following requirements:

Data changes must be tracked and versioned.
The model must be retrained consistently by using approved training data.

Which two actions should you include in the design? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. Associate the storage location to the fine-tuning job.

B. Create a content filter.

C. Store the training data in Azure Files.

D. Upload the training data to Microsoft Foundry data files

E. Store the training data in Azure Blob Storage that has version control enabled.

D. Upload the training data to Microsoft Foundry data files

E. Store the training data in Azure Blob Storage that has version control enabled.

为什么选择 D 而不是 A？

Microsoft Foundry 模型微调的具体机制（关键点）：
- 在 Microsoft Foundry / Azure OpenAI 中发起模型微调（Fine-tuning job）时，系统并不支持直接给微调任务关联一个“存储位置/文件夹（Storage location）”。
- 实际上，微调任务接收的输入是一个明确的、经过验证的数据文件（Data File / File ID，例如 file-abc123...）。
- 因此，标准的工作流程是：在启用了版本控制的 Blob Storage (E) 中准备好数据后，将该特定的受信任数据上传/注册到 Microsoft Foundry data files (D) 中生成 Data Asset，再由微调任务调用该文件。
完整的工作流逻辑（E + D）：
- 满足需求 1（版本控制与追踪）： E (Store in Azure Blob Storage with version control) — Blob 存储负责底层原始数据文件的版本快照与历史追溯。
- 满足需求 2（受信任数据的可重复微调）： D (Upload to Microsoft Foundry data files) — 将每次版本控制后经批准的数据集导入到 Foundry 数据文件中，作为模型微调任务的确定性输入（Training File）。

总结

正确答案确定为：D 和 E
之前的分析中 A 属于概念性的误判，D 才是符合微软 Azure AI Foundry 接口规范的标准操作。

Correct Answer: DE Section: (none)

Explanation/Reference:

Explanation:

To implement a new Application Lifecycle Management (ALM) process for your fine-tuned Microsoft Foundry model, you can leverage Azure Blob Storage for raw data versioning and Azure AI Foundry (formerly Azure AI Studio) for managed data assets and model retraining.

1.Data Tracking and Versioning in Azure Blob Storage Enable blob versioning to automatically maintain a history of your training data.

Enable Versioning: In the Azure portal, navigate to your storage account under Data management > Data protection and select Enable versioning for blobs.

Immutable Tracking: This ensures that every update to a training file creates a new version, allowing you to restore or reference specific historical states of your dataset for auditability.

2.Uploading and Managing Data in Microsoft Foundry Integrate your versioned blob storage with Microsoft Foundry to create "Data assets" that the model can use for retraining.

QUESTION 16

A company deploys agents that generate responses by using Azure OpenAI resources. The agents are deployed to both the United States and Europe.

You need to recommend a governance solution that meets the following requirements:

Enforces the deployment of the resources to only approved Azure regions
Provides continuous compliance verification of the resources

What should you include in the recommendation for each requirement? To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

Hot Area:

Answer Area

Enforces the deployment of the resources to only approved regions:

Azure Monitor
Azure Policy
Microsoft Defender for Cloud
Microsoft Purview
Microsoft Sentinel

Provides continuous compliance verification of the resources:

Azure Monitor
Azure Policy
Microsoft Defender for Cloud
Microsoft Purview

正确答案是：

Requirement	Correct option	Why
Enforces the deployment of the resources to only approved regions	✅ Azure Policy	Azure Policy 可以通过 `Allowed locations` policy 限制 Azure 资源只能部署到指定区域，例如只允许 US 和 Europe 的批准区域。
Provides continuous compliance verification of the resources	✅ Azure Policy	Azure Policy 会持续评估资源是否符合 policy，并提供 compliance state；不符合要求的资源会被标记为 Non-compliant。

最终选择

Enforces the deployment of the resources to only approved regions:

➡️ Azure Policy

Provides continuous compliance verification of the resources:

➡️ Azure Policy

这道题的核心是区分 Governance / Monitoring / Security：

Azure Policy → Govern / Enforce / Compliance
Azure Monitor → Metrics, Logs, Alerts, Performance
Microsoft Defender for Cloud → Security posture, recommendations, workload protection
Microsoft Purview → Data governance, data catalog, classification, lineage
Microsoft Sentinel → SIEM / SOAR / security analytics

所以看到：

approved regions + continuous compliance

几乎直接联想到：

Azure Policy → Allowed locations + Compliance evaluation

而且这里是 Azure OpenAI resources 部署到 US 和 Europe，并不改变答案；关键在于题目要求的是限制资源部署区域并持续检查合规性。

为什么 Box 2 不是 Microsoft Purview？

这里最容易被题目里的 "compliance" 误导。

Microsoft Purview 的核心定位是：

Data governance
Data catalog
Data classification
Data lineage
Data discovery
Compliance / risk management for data and regulatory requirements

而题目说的是：

continuous compliance verification of the resources

这里的 resources 指 Azure resources，本质上是要求持续检查 Azure 资源是否符合组织定义的配置/治理规则。

这正是 Azure Policy 的核心能力：

Define → Enforce → Audit → Continuously evaluate compliance

例如：

Azure Policy
     │
     ├── Allowed locations
     │      ├── East US
     │      ├── West Europe
     │      └── ...
     │
     └── Compliance evaluation
             │
             ├── Compliant
             └── Non-compliant

一个很重要的考试陷阱

你看到的答案把：

compliance

自动联想到：

Microsoft Purview Compliance Manager

这是一个很常见的误区。

考试中要看 compliance 的对象是什么：

题目关键词	首选
Azure resources 的配置是否符合规则	Azure Policy
Allowed locations	Azure Policy
Enforce / Deny deployment	Azure Policy
Continuous compliance state	Azure Policy
Regulatory compliance assessment	Microsoft Purview / Compliance Manager
Data classification	Microsoft Purview
Data governance / lineage	Microsoft Purview
Security posture / vulnerabilities	Defender for Cloud

所以这道题实际上是一个非常典型的：

Azure Policy + Azure Policy

题目甚至故意把 Microsoft Purview 放进去干扰你。

关于原回答中的一个细节

它说：

"Microsoft Purview Compliance Manager ... continuous monitoring against global regulations..."

这个描述本身并非完全错误。Purview Compliance Manager 确实可以做 regulatory compliance assessment。

但它不符合这道题的 requirement。

题目不是问：

“如何持续验证公司是否符合 GDPR / EU AI Act？”

而是：

“如何持续验证这些 Azure resources 是否符合治理要求？”

因此应该选择 Azure Policy。

🧠 AB-100 / Azure Governance 考试记忆

你可以直接记：

Policy = Resource governance & compliance

看到以下关键词，优先想到 Azure Policy：

Allowed locations / enforce / deny / audit / compliance / resource configuration / organizational standards

所以最终答案：

Box 1 → Azure Policy ✅

Box 2 → Azure Policy ✅

QUESTION 17

A company has an AI solution that uses Azure OpenAI models.

You need to recommend a governance solution that monitors and audits changes to model configurations and data usage.

The solution must minimize administrative effort.

What should you include in the recommendation?

A. Azure Monitor

B. Azure Stream Analytics

C. Azure API Management

D. Azure Policy

E. Microsoft Purview

E. Microsoft Purview

核心需求分析：
- 需要对 Azure OpenAI 模型及系统的数据使用（Data usage）和模型配置与合规变化进行监控与审计（Monitors and audits）。
- 限制条件：必须最小化管理工作量（Minimizes administrative effort）。
产品功能与治理定位：
- Microsoft Purview（Purview AI Hub & Audit）：是微软统一的企业级数据治理与合规审计服务。
- 开箱即用的 AI 治理能力： Purview 提供了开箱即用的 AI 治理平台，能够自动监控和审计 AI 模型访问了哪些敏感数据（Data Usage Tracking）、生成式回答的数据使用痕迹以及整体合规策略审计。由于具备内置的自动化扫描和集中式合规仪表板，它能够以最小的管理工作量满足数据使用和模型合规审计要求。

这里考的是治理（governance）+ 审计（audit）+ 数据使用（data usage），而不是单纯的资源配置监控。

为什么不选其他选项？

A. Azure Monitor：虽然 Azure Monitor 可以通过 Activity Logs 记录配置更改，通过 Diagnostic Logs 记录 API 调用，但要实现对“数据使用（Data usage）”的合规审计和治理分析，需要手动编写大量的 KQL 查询、日志路由规则和自定义仪表板，不符合“最小化管理工作量”的要求。
B. Azure Stream Analytics：属于实时流数据分析服务，用于编写 SQL 实时处理流数据，需要大量手动开发工作量。
C. Azure API Management：侧重于 API 网关路由、认证和限流，不属于全面的数据使用治理与合规审计平台。
D. Azure Policy：主要用于控制 Azure 基础设施资源层面（ARM Control Plane）的配置合规性，无法追踪和审计数据平面（Data Plane）的具体数据使用（Data Usage）情况。

QUESTION 18

A company uses Azure OpenAI models that use grounding data from Microsoft Fabric for agents. The models are fine-tuned by using proprietary datasets.

You need to design a governance solution that meets the following requirements:

Restricts access to the grounding data to only assigned roles
Restricts model fine-tuning to only the AI engineering team

What should you include in the design? To answer, select the appropriate options in the answer area. NOTE: Each correct selection is worth one point.

Answer Area

Restricts access to the grounding data:

Azure AI Content Safety
Azure Monitor alerts
Azure Policy compliance rules
Microsoft Purview access policies

Restricts model fine-tuning:

Azure Policy compliance rules
Azure Resource Manager (ARM) resource locks
Microsoft Entra Conditional Access
Role-based access control (RBAC) in Microsoft Foundry

1. Restricts access to the grounding data:

正确选项： Microsoft Purview access policies（Microsoft Purview 访问策略）

2. Restricts model fine-tuning:

正确选项： Role-based access control (RBAC) in Microsoft Foundry（Microsoft Foundry 中的基于角色的访问控制）

第一部分分析：限制对接地数据（Grounding Data）的访问

考点： 企业数据资产的数据平面访问控制（Data-plane Access Control）。
解析：
- 题目说明 grounding data 来自 Microsoft Fabric。
- Microsoft Purview access policies（Microsoft Purview 访问策略）：允许企业集中管理和定义针对 Microsoft Fabric、Azure Data Lake 等数据资产的细粒度访问控制策略。通过 Purview 访问策略，可以确保只有获得授权角色（Assigned roles）的人员或 Agent 才能读取和访问这些接地数据。

第二部分分析：限制模型微调（Model Fine-tuning）权限

考点： AI 平台管理操作权限控制（Control-plane & Data-plane RBAC）。
解析：
- 模型微调（Fine-tuning）是 Microsoft Foundry（Azure AI Foundry / Azure OpenAI）内部的一种高算力消耗且涉及核心资产的操作。
- Role-based access control (RBAC) in Microsoft Foundry：微软在 Foundry 平台中提供了专门的 RBAC 角色（如 Azure AI Developer、Cognitive Services OpenAI Contributor 等）。通过为 AI 工程团队（AI engineering team）分配特定的 Foundry RBAC 角色，可以精准限制只有该团队拥有发起和运行模型微调任务的权限，对其他普通用户进行隔离。

干扰项排除说明

Azure AI Content Safety：用于实时检测和过滤违规内容（如暴恐、仇恨言论），不用于数据访问控制或用户权限管理。
Azure Resource Manager (ARM) resource locks：用于防止 Azure 资源被意外删除（CanNotDelete）或修改（ReadOnly），无法精细化控制“谁可以进行模型微调”这一业务操作。
Microsoft Entra Conditional Access：侧重于用户登录时的身份认证条件限制（如 MFA 强制、IP 限制、设备合规要求），不直接管理 Foundry 内部的具体 AI 模型微调权限。

Explanation/Reference:

Explanation:

Box 1: Microsoft Purview Access Policies Restricts access to the grounding data to only assigned roles

To secure and manage grounding data from Microsoft Fabric for Azure OpenAI agents and fine-tuned models, you can use Microsoft Purview to enforce role-based access and data protection policies.

Note:

Securing Data with Microsoft Purview & RBAC

Access Control Policies: Microsoft Purview enables role-based access controls (RBAC) over Fabric items, ensuring that when an AI agent retrieves data, it only accesses information the user is permitted to see.

Sensitivity Labels: Data in Fabric can be labeled (e.g., "Confidential"). Purview policies can restrict AI agents from accessing or acting upon content that violates these security labels.

OneLake Security: Fine-grained security in Fabric (Row-Level Security and Column-Level Security) is automatically honored by agents, guaranteeing that even with access to a dataset, sensitive PII (Personally Identifiable Information) can be restricted.

QUESTION 19

A company has a Microsoft Copilot Studio agent that has been in production for three months.

The agent has received positive feedback from users.

You need to identify the number of questions unanswered by the agent and the number of abandoned sessions between the users and the agent.

Which Copilot Studio insights should you use? To answer, drag the appropriate insights to the correct requirements. Each insight may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

NOTE: Each correct selection is worth one point.

Select and Place:

Insights

Conversation outcomes
Generated answer rate and quality
Reactions
Survey results

Answer Area

The number of unanswered questions:
The number of abandoned sessions:

答案：

Requirement	Correct Insight	原因
The number of unanswered questions	✅ Generated answer rate and quality	显示 AI 回答率、未回答问题（Unanswered Questions）及回答质量。
The number of abandoned sessions	✅ Conversation outcomes	显示对话结果，包括 Resolved、Escalated、Abandoned 等会话统计。

为什么其他选项不对？

❌ Reactions：统计用户 👍 / 👎 反馈，不统计未回答或放弃会话。
❌ Survey results：统计问卷调查结果，不统计会话数据。

考试记忆

Unanswered questions → ✅ Generated answer rate and quality
Abandoned sessions → ✅ Conversation outcomes

Explanation/Reference:

Explanation:

Box 1: Generated answer, rate and quality

The number of questions unanswered

In Microsoft Copilot Studio, you can find the number of unanswered questions by navigating to the Analytics page and viewing the Generated answer rate and quality section.

This specific analytics area provides insights into how often the agent successfully generates a response versus when it fails to do so.

Box 2: Conversation outcomes

The number of abandoned sessions between the users and the agent

To find the number of abandoned sessions in a Microsoft Copilot Studio agent's analytics, you can use the Conversation outcomes chart within the Analytics dashboard.

This chart specifically tracks and visualizes sessions with an "Abandoned" outcome.

QUESTION 20

You are evaluating a Microsoft Copilot Studio agent that supports Microsoft Dynamics 365 Customer Service representatives.

You need to recommend a testing solution that meets the following requirements:

Evaluates agent effectiveness during active sessions
Validates whether the agent delivers accurate and helpful responses
Provides measurable, actionable insights for continuous improvement

What should you recommend?

A. Track resolution, deflection, and accuracy by using dashboards and use scripts to ensure consistent responses.

B. Perform load testing to validate how the agent scales under a high chat volume.

C. Review historical tickets to find agents that have the shortest resolution times.

D. Measure uptime and page load times.

正确答案：

A. Track resolution, deflection, and accuracy by using dashboards and use scripts to ensure consistent responses.（使用仪表板跟踪解决率、偏转率和准确性，并使用脚本确保一致的响应。）

解析与考点分析

为什么选 A？

对照题目提出的三项核心需求：

评估活动会话期间的 Agent 有效性（Evaluates agent effectiveness during active sessions）：
Copilot Studio 内置的分析仪表板（Analytics Dashboards）可以实时追踪会话，分析解决率（Resolution rate）和偏转率（Deflection rate，即无需转人工即可自主解决问题的比例）。
验证 Agent 是否提供准确且有帮助的回答（Validates whether the agent delivers accurate and helpful responses）：
通过使用测试脚本（Scripts）和内容准确度评估，可以验证生成式回答和主题回复的一致性与正确性。
提供可衡量、可操作的持续改进洞察（Provides measurable, actionable insights for continuous improvement）：
仪表板提供的核心 KPI（解决率、放弃率、CSAT 满意度评分、未解决问题主题分类）能够为后续的 Prompt 优化和知识库扩充提供直接的数据支撑。

为什么不选其他选项？

B. Perform load testing to validate how the agent scales under a high chat volume（进行压力测试以验证高并发下的扩容能力）： 这是性能与并发测试（Performance/Load Testing），关注的是系统吞吐量和并发响应，无法评估回答的“准确度”和“内容有用性”。
C. Review historical tickets to find agents that have the shortest resolution times（审查历史工单以寻找解决时间最短的客服人员）： 评估的是人类客服（Human Representatives）的历史工单处理速度，而不是 AI Copilot Agent 的性能。
D. Measure uptime and page load time（测量在线时间与页面加载时间）： 这是基础架构的可用性与网络性能指标（Infrastructure Metrics），与 AI Agent 的对话质量和解答准确性无关。

Correct Answer: A

Explanation/Reference:

Explanation:

To establish a testing and evaluation setup for your Microsoft Copilot Studio agent within Dynamics 365 Customer Service, you should leverage specialized AI-driven evaluation agents and integrated analytics dashboards.

Evaluate Effectiveness and Accuracy Use the Quality Evaluation Agent in Dynamics 365 to automate the assessment of agent performance during and after active sessions.
Track Measurable Insights with Dashboards Utilize the built-in and customizable dashboards to monitor key performance indicators (KPIs) like resolution and deflection.
Ensure Consistency with Agent Scripts To maintain uniform and company-endorsed communication, implement Agent Scripts within the Customer Service Admin center.

QUESTION 21

A company uses multiple Microsoft Copilot Studio agents across different channels.

You need to recommend a monitoring solution that provides comprehensive telemetry data and performance insights for the agents.

What should you include in the recommendation?

A. Application Insights

B. Azure Advisor

C. Azure DevOps

D. Microsoft Dynamics 365 Customer Voice

正确答案与详细解析

正确答案：

A. Application Insights

解析与考点分析

为什么选 A？

业务场景：公司在不同渠道上部署了多个 Microsoft Copilot Studio Agent，需要一个能够提供全面遥测数据（Comprehensive telemetry data）和性能洞察（Performance insights）的监控解决方案。
官方原生集成：
- Azure Application Insights（Azure 应用程序洞察）是 Microsoft 推荐且与 Copilot Studio 原生集成的遥测与监控服务。
- 通过配置 Application Insights，系统会自动捕获并记录 Agent 的详细遥测日志，包括：
  - 对话事件（Events）与会话轨迹（Traces）
  - 各 Topic（主题）的执行情况与触发频率
  - 系统响应延迟、API 调用耗时与性能瓶颈
  - 运行时的错误日志与异常信息（Exceptions）

为什么不选其他选项？

B. Azure Advisor（Azure 顾问）：针对 Azure 基础设施架构提供最佳实践建议（如成本优化、安全性、合规性），不能收集和分析 Copilot Studio Agent 的对话遥测数据。
C. Azure DevOps：用于 DevOps 流程（代码托管、CI/CD 自动化构建与部署 pipelines、项目工作项追踪），不具备实时监控和运行时遥测分析功能。
D. Microsoft Dynamics 365 Customer Voice：用于创建和分发用户满意度（CSAT/NPS）调查问卷，只能收集用户的主观反馈，无法提供技术维度的底层遥测数据和系统性能洞察。

Correct Answer: A Section: (none)

Explanation/Reference:

Explanation:

To achieve comprehensive monitoring, telemetry, and performance insights for Microsoft Copilot Studio agents across different channels, you should integrate them with Azure Application Insights.

Integrating Application Insights provides a centralized view of agent health, user interactions, topic performance, and latency, which is crucial for monitoring multi-channel deployments.

QUESTION 22

A company has an AI solution built by using Microsoft Copilot Studio and Power Platform. The solution is used by the company's sales, marketing, and customer service teams.

You are performing a return on AI investment (ROAI) analysis to evaluate the impact of the solution.

You need to identify which measurable business drivers to include in the analysis.

Which two business drivers should you identify? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. the reduced average case resolution time

B. market capitalization

C. economic market predictability

D. increased employee productivity

E. brand awareness

正确答案：

A. the reduced average case resolution time（平均案件解决时间的缩短）
D. increased employee productivity（员工生产力的提高）

解析与考点分析

为什么选 A 和 D？

在评估 AI 投资回报率（ROAI, Return on AI Investment）时，需要引入可量化（Measurable）且能直接归因于 AI 解决方案落地的业务驱动因素（Business Drivers）：

A. the reduced average case resolution time（平均案件解决时间的缩短）：
- 适用团队：客户服务团队（Customer Service Team）。
- 可量化分析：属于标准的客服核心指标（MTTR, Mean Time to Resolution）。使用 Copilot Studio 智能体协助客服人员快速检索知识库并生成回复后，每个工单的处理时长（如从 15 分钟缩短至 5 分钟）可以被系统精准记录，从而直接换算为节省的人工工时成本与财务收益。
D. increased employee productivity（员工生产力的提高）：
- 适用团队：销售、营销和客服团队（Sales, Marketing & Customer Service Teams）。
- 可量化分析：属于 AI 投资回报率模型中最核心的驱动指标。例如：销售人员准备会议的时间减少、营销人员生成邮件草稿的速度加快、客服人员每日处理工单量增加。通过“每周节省的工时 × 员工时薪成本”，可以非常直观地计算出 ROAI 中的经济回报。

为什么不选其他选项？

B. market capitalization（总市值）：受宏观经济、股市波动、利率等多重复杂外部因素影响，无法归因并量化为单个 Copilot Studio AI 项目的回报。
C. economic market predictability（经济市场可预测性）：属于宏观经济学指标，非企业内部 AI 项目的业务驱动力。
E. brand awareness（品牌知名度）：属于难以精准量化且难以直接与内部 AI 效率提升挂钩的软性指标，不适合作为量化 ROAI 分析的核心财务/运营驱动因子。

Correct Answer: AD Section: (none)

Explanation/Reference:

Explanation:

[A] Reducing the average case resolution time is a high-impact, measurable business driver for an AI investment (ROAI) analysis, particularly when implementing Microsoft Copilot Studio and Power Platform for local sales, marketing, and customer service teams.

According to Microsoft and industry research, this metric directly translates to reduced operational costs, increased agent productivity, and higher customer satisfaction.

[D] Increased employee productivity is a primary and highly measurable business driver for an AI investment (ROAI) analysis involving Microsoft Copilot Studio and Power Platform, particularly for sales, marketing, and customer service teams. These tools allow organizations to automate manual tasks, streamline workflows, and enhance content generation, directly resulting in time savings and efficiency gains.

Incorrect:

[Not B] While market capitalization is a macro-level indicator that reflects investor confidence in a company's AI maturity, it is generally not considered a direct, measurable business driver for a specific internal project's Return on AI Investment (ROAI). Market cap is influenced by many external factors beyond a single implementation of Microsoft Copilot Studio or Power Platform.

QUESTION 23

A company has a Microsoft Foundry agent that summarizes customer feedback and recommends products to customers. The agent references data from multiple knowledge sources.

Users report that the agent response time is slow.

Telemetry data shows that the agent frequently reaches its token usage limit.

You need to recommend a solution to reduce token usage without degrading the quality of the generated responses.

What should you recommend?

A. Chunk documents during indexing.

B. Reduce the number of knowledge sources used by the agent.

C. Reconfigure the prompts to limit the amount of retrieved content from the knowledge sources.

D. Lower the maximum token usage limit for the responses.

正确答案：✅ C. Reconfigure the prompts to limit the amount of retrieved content from the knowledge sources.

原因

问题是：

Agent 经常达到 token usage limit
导致响应变慢
但要求不能降低回答质量

最合适的是控制 retrieved content 的数量，让模型只接收与问题最相关的信息。

其他选项

❌ A. Chunk documents during indexing → 可以改善检索，但不一定直接减少最终传给模型的 token。
❌ B. Reduce knowledge sources → 可能丢失有用信息，影响回答质量。
❌ D. Lower maximum token usage → 直接限制模型，可能降低回答质量。

记忆

Token 太多 + 保持质量 → 减少 Retrieval Context，而不是减少 Knowledge Sources。

答案：C

Correct Answer: C Section: (none)

Explanation/Reference:

Explanation:

To reduce token usage and improve performance in a Microsoft Foundry agent with slow response times and high token usage, reconfigure the prompts to limit the amount of retrieved content from the knowledge sources. This directly reduces the input token count in the prompt, addressing the bottleneck without removing valuable sources.

To reduce token usage while maintaining quality, you must transition from a "wide-net" retrieval strategy to a highly filtered, multi-stage retrieval process. The goal is to ensure that every token entering the prompt adds unique, high-value information.

The Root Cause Redundancy: Multiple knowledge sources often contain overlapping feedback.

Noise: Raw data often includes metadata or conversational filler.

Context Bloat: Large retrieval windows push the agent toward its limit and slow down processing.

QUESTION 24

A company has Microsoft Power Platform development, staging, and production environments. Each environment has its own Microsoft Dataverse tables and Azure AI Search index.

You are designing an application lifecycle management (AIM) process to deploy a Microsoft Copilot Studio agent between the environments.

The company has a Copilot Studio agent named Agent1 in development. Agent1 uses the following grounding data sources:

A Dataverse table named CustomerOrders
An Azure AI Search index named customer-knowledge

You need to deploy Agent1 to production. The solution must ensure that the agent uses the production grounding data sources, minimizes downtime, and handles credentials and endpoints securely.

What should you include in the deployment package solution, and what should you reconfigure after the deployment?

To answer, select the appropriate options in the answer area. NOTE: Each correct selection is worth one point.

Include in the deployment package solution:

Agent1 only
The data sources only
Agent1 and the data source connections
Agent1 and references to the data sources
Agent1, the data sources, and the data source connections

Reconfigure after the deployment:

The Dataverse connection only
The Azure AI Search connection only
The Dataverse and Azure AI Search connections
The Agent1 configuration
The environment variables

正确答案

Include in the deployment package solution : ✅ Agent1 and references to the data sources

Reconfigure after the deployment : ✅ The environment variables

为什么其他答案不对？

选项	为什么错
Agent1 only	❌ 丢失数据源引用
Data sources only	❌ Agent 不会部署
Agent1 + data source connections	❌ Connection 不应跨环境迁移，涉及凭据
Agent1 + data sources + connections	❌ 不会迁移 Dataverse 数据和 AI Search Index，也不建议迁移连接
Dataverse connection only	❌ 还需要 Azure AI Search 的环境配置
Azure AI Search connection only	❌ 同理
Dataverse + Azure AI Search connections	❌ 推荐通过 Environment Variables 管理，而不是修改连接
Agent1 configuration	❌ Agent 配置无需修改，引用会通过环境变量解析

📌 考试记忆口诀

Copilot Studio ALM：

Solution → Agent + References（引用）
不要打包 → 数据、Search Index 内容、凭据
跨环境切换 → Environment Variables
Secrets / Endpoints → Environment Variables + Connection References

最终答案：

Include in the deployment package solution: ✅ Agent1 and references to the data sources
Reconfigure after the deployment: ✅ The environment variables

Box 1: Agent1 and references to the data sources

Box 2: The environment variables

Box 1: Agent1 and references to the data sources

Include in the deployment package solution

To deploy a Microsoft Copilot Studio agent with Dataverse and Azure AI Search dependencies across environments, you must use a Managed Solution as your primary deployment package. This ensures environment-specific configurations are handled without hardcoding, maintaining security and minimizing downtime.

Key Components of the Deployment Package Your solution package must include the following components to ensure a seamless transition between development, staging, and production:

*-> Copilot Studio Agent: The core bot configuration and its topics.

*-> Connection References: Placeholders that link the agent to specific connectors (e.g., Azure AI Search, Dataverse) in each environment.

Environment Variables: Used to store the Azure AI Search Endpoint URL and Index Name. Dataverse Table Metadata: If using custom tables, include the table definitions (schema) within the solution.

Power Automate Flows: If flows are used to bridge the agent and search index, they must be included in the same solution.

Box 2: The environment variables

Reconfigure after the deployment

To successfully deploy a Copilot Studio agent across environments while maintaining the integrity of its data grounding, you must reconfigure the Connection References and Environment Variables within the target environment.

In a professional ALM (Application Lifecycle Management) pipeline, you should not manually "edit" the agent after deployment. Instead, the solution should be designed to use placeholders that are updated during the import process.

Update Environment Variables Purpose: Swaps out API endpoints and resource names.

Action: Provide the production values for:

Azure AI Search Endpoint: (e.g., https://windows.net) Index Name: The specific name of the production index.

API Keys: Ideally stored in Azure Key Vault and referenced via an environment variable.

Benefit: Allows the same agent logic to run in any environment by simply changing the variable values

QUESTION 25

A company has a Microsoft Copilot Studio agent that uses generative AI to assist Microsoft Dynamics 365 Customer Service representatives.

The agent currently exhibits a low resolution rate and a high escalation rate.

You need to identify the issue.

What should you use?

A. the Agent dashboard of Dynamics 365 Customer Service historical analytics

B. the Insights tab from the Search & intelligence settings of the Microsoft 365 admin center

C. the Copilot hub in the Power Platform admin center

D. the Analytics tab in Copilot Studio

正确答案：✅ D. the Analytics tab in Copilot Studio

原因

题目要分析的是 Copilot Studio Agent 的：

Low resolution rate
High escalation rate
找出 Agent 为什么表现不好

Copilot Studio → Analytics tab 提供 Agent 的运行分析，包括：

Resolution rate
Escalation rate
Engagement
Conversation outcomes
Answer quality

因此最直接。

其他选项

❌ A. Dynamics 365 Customer Service historical analytics → 更偏 Customer Service 整体运营/客服历史数据。
❌ B. Microsoft 365 admin center → 不用于 Copilot Studio Agent 的详细运行分析。
❌ C. Power Platform admin center Copilot hub → 偏管理、治理和整体 Copilot 管理，不是 Agent performance troubleshooting 的主要工具。

🎯 最终答案

D. the Analytics tab in Copilot Studio ✅

Correct Answer: D Section: (none)

Explanation/Reference:

Explanation:

To locate the problem behind a low resolution and high escalation rate in your Copilot Studio agent, you should focus on a multi-layered diagnostic approach. High escalation typically signals that the agent is hitting a "Fallback" trigger or failing to find grounded answers in its knowledge base.

Core Diagnostic Steps Analyze Analytics Dashboards *-> Use the built-in Copilot Studio Analytics tab to identify high-level trends.

Escalation Rate Drivers: Look for the top 5–10 topics causing escalations. Improving these key topics by even 10% can significantly boost overall deflection

Outcome Reason: Check the outcomeReason in the ConversationTranscript table in Dataverse to see if sessions end due to abandonment, system errors, or explicit user requests for a human.

QUESTION 26

A company has a Microsoft Foundry generative AI model.

You need to evaluate the model's output to measure the overall quality and coherence of generated responses.

The evaluation must use GPT-4o as a judge and return a numeric score for each output.

Which type of metric should you use?

A. AI quality (NLP)

B. risk and safety

C. Groundedness

D. AI quality (AI assisted)

答案：✅ D. AI quality (AI assisted)

解析

题目关键词：

使用 GPT-4o as a judge
评估输出质量和连贯性（quality & coherence）
返回数值评分（numeric score）

这正是 AI quality (AI assisted) 的特点。

它使用 GPT-4o 作为评审（LLM-as-a-Judge），对生成内容进行自动评分，例如：

Relevance（相关性）
Coherence（连贯性）
Fluency（流畅性）
Overall quality（整体质量）

为什么其他选项不对？

❌ A. AI quality (NLP)
使用传统 NLP 指标（如 BLEU、ROUGE、F1），不会使用 GPT-4o 作为评审。
❌ B. Risk and safety
评估有害内容、暴力、仇恨、越狱等安全风险，不评估回答质量。
❌ C. Groundedness
评估回答是否基于提供的数据或知识来源，不是整体质量和连贯性评分。

考试记忆

GPT-4o as judge / LLM-as-a-Judge → ✅ AI quality (AI assisted)
BLEU、ROUGE 等传统指标 → AI quality (NLP)
是否基于数据回答 → Groundedness
安全、违规内容 → Risk and safety

最终答案：✅ D. AI quality (AI assisted)

Correct Answer: D Section: (none) Explanation/Reference:

Explanation:

To evaluate overall quality and coherence in Microsoft Foundry, use AI-assisted quality evaluators. These metrics use a "judge" model (like GPT-4o) to score responses on a numeric Likert scale (typically 1 to 5). Recommended Metric: Coherence The Coherence evaluator is the most direct fit for measuring how well-structured and logical a response is.

What it measures: The collective quality of all sentences and how they flow together.

Numeric Score: Returns an integer from 1 (Poor) to 5 (Excellent).

Output: Includes a coherence score, a coherence_reason (explanation from the judge), and a pass/fail result based on a threshold (default is 3).

QUESTION 27

You use Microsoft Copilot Studio analytics to analyze the performance of a deployed Copilot Studio agent.

You need to identify which performance metrics to use to measure the following:

The percentage of engaged sessions that are escalated to a live customer service representative
The number of agent queries that cause a knowledge source error

What should you identify for each requirement?

To answer, select the appropriate options in the answer area.

NOTE: Each correct selection is worth one point.

The percentage of engaged sessions that are escalated to a representative:

Answer quality
Customer Satisfaction (CSAT) score
Engagement rate
Escalation rate

The number of agent queries that cause a knowledge source error:

Answer quality
Engagement rate
Escalation rate

答案：

Percentage of engaged sessions escalated to a representative: ✅ Escalation rate
Number of agent queries that cause a knowledge source error: ✅ Answer quality

记忆

Escalation rate → 衡量有多少会话被转人工。
Answer quality → 包含回答失败、知识源错误等质量问题。
Engagement rate → 用户参与/互动程度。
CSAT → 用户满意度。

最终：

Escalated sessions → Escalation rate Knowledge source errors → Answer quality

Box 1: Escalation rate

The percentage of engaged sessions that are escalated to a live customer service representative To measure the percentage of engaged sessions that are escalated to a live customer service representative in Microsoft Copilot Studio analytics, you should use the Escalation rate metric.

Lowering this percentage is a key objective to improve the overall deflection rate and enhance agent performance.

Definition: An escalated session is an engaged session that ends when the Escalate topic is triggered, or when a Transfer to agent node is reached.

Context: This metric helps you understand the proportion of users who requested to speak to a human or were passed to a representative because the agent couldn't solve their issue.

Where to find it: It is available in the Analytics dashboard within Copilot Studio, specifically under "Escalation Rate Drivers" which shows which topics cause the most escalations.

Box 2: Answer quality

The number of agent queries that cause a knowledge source error When analyzing Microsoft Copilot Studio agent performance, the answer quality performance metric should be used to measure queries that cause a knowledge source error. This AI-driven metric assesses if responses are grounded in provided knowledge, helping to identify, evaluate, and reduce errors in generative answers.

Key Aspects of Copilot Studio Analytics:

Answer Quality: Utilizes Large Language Models (LLMs) to categorize chat messages, focusing on grounding, relevance, and accuracy.

QUESTION 28

You need to recommend a security solution for agents in a Microsoft Power Platform environment.

The agents must use only approved connectors and services.

The solution must prevent the agents from accessing sensitive data.

What should you recommend?

A. Configure Azure Monitor to capture connector activity logs.

B. Enable a Microsoft Dataverse audit.

C. Deploy data loss prevention (DLP) policies in Power Platform.

D. Enable customer-managed keys in Microsoft Dataverse.

正确答案：✅ C. Deploy data loss prevention (DLP) policies in Power Platform.

原因

题目两个关键要求：

只能使用 approved connectors and services
Prevent agents from accessing sensitive data

Power Platform DLP policies 可以：

控制哪些 Connectors 可以使用
将 Connectors 分为 Business / Non-business / Blocked
阻止不允许的 Connector 组合和数据流
防止 Power Platform Agent/Flow 将业务数据流向未经批准的服务

其他选项

❌ A. Azure Monitor → 主要用于监控和日志，不负责阻止访问。
❌ B. Dataverse audit → 记录数据访问/变更，但不能实施 Connector 使用限制。
❌ D. Customer-managed keys → 主要解决数据加密密钥控制，不负责 Connector/DLP 治理。

🎯 考试记忆

Approved connectors + prevent data leakage → Power Platform DLP

答案：C. Deploy data loss prevention (DLP) policies in Power Platform.

Correct Answer: C Section: (none)

Explanation/Reference:

Explanation:

To secure Microsoft Power Platform agents, deploy DLP policies via the Power Platform Admin Center to categorize connectors as Business, Non-business, or Blocked. This prevents unauthorized data exfiltration by separating sensitive connectors and restricting approved services, ensuring agents only use compliant actions.

Note Key Actions for Agent DLP Policies:

Create Policy: Create a new policy in the Power Platform Admin Center and define its scope.

Categorize Connectors: Classify connectors used by agents (e.g., HTTP, SQL) into Business/Nonbusiness to prevent data sharing between them.

Block High-Risk Connectors: Block unauthorized or risky connectors (e.g., public social media, unapproved AI tools).

Configure Endpoint Filtering: Block specific URLs for HTTP connectors to ensure data only travels to trusted domains.

Apply to Agents: According to Platforms of Power, specific connectors for Copilot Studio can be blocked to limit agent capabilities, such as preventing autonomous triggers.

Quarantine Agents: If agents violate Microsoft Learn policies, 365 Copilot Studio allows quarantining the agents to prevent further data exposure.

QUESTION 30

A company has an AI business solution that uses Microsoft Copilot Studio agents.

You need to recommend prompt best practices to improve the effectiveness of agent interactions.

Which two actions should you include in the recommendation? Each correct answer presents part of the solution.

NOTE: Each correct selection is worth one point.

A. Track the duration of the average user session.

B. Analyze the prompt length distribution.

C. Regularly test and refine the prompts based on user input.

D. Use clear and specific instructions in the prompts.

E. Measure system resource usage during prompt processing.

正确答案：✅ C、D

C. Regularly test and refine the prompts based on user input. → 持续根据实际用户反馈优化 Prompt。
D. Use clear and specific instructions in the prompts. → 明确、具体的指令可以提高 Agent 回答的准确性和一致性。

记忆

Prompt Best Practices = Clear & Specific + Test & Refine

答案：C + D

Correct Answer: CD

Explanation/Reference:

Explanation:

To maximize the effectiveness of Microsoft Copilot Studio agents through prompt engineering, the two best practices you mentioned are critical, along with several key refinements:

Use Clear and Specific Instructions in Prompts

Define Goal, Context, and Persona: Provide a clear objective (what to do), necessary context (background information), and a defined persona (e.g., "You are an HR assistant...") to set the tone.

Be Specific and Actionable: Use precise verbs like "search," "summarize," "draft," or "list." Avoid vague instructions. For example, instead of "tell me about sales," use "Summarize Q3 sales performance in the North region".

Use Structured Formats: Utilize Markdown (headings, bullet points, numbered lists) to make instructions easy to scan and parse. Organize instructions into sections like # PURPOSE, ## RESTRICTIONS, and ## ACTIONS.

Provide Examples (Few-Shot Prompting): Include examples of inputs and desired outputs to help the agent understand complex tasks or required formatting.

Define Constraints: Explicitly state what the agent should not do, what information to exclude, or when to decline a request.

Regularly Test and Refine Prompts Based on User Input

Test within Copilot Studio: Utilize the testing pane to simulate user queries and review agent responses. Experiment with different prompt variations to see which yields the best result.

Use Prompt Advisor: Leverage the Prompt Advisor feature to get a confidence score and feedback on your prompts, which helps identify areas for improvement.

Review Analytics: Regularly check conversation logs, customer feedback (thumbs up/down), and user queries in the analytics dashboard to find where the agent struggles.

Iterate and Improve: Treat prompting as a conversation with the agent. If the first output isn't perfect, refine the instructions, add more context, and try again.

Monitor Performance (Red Teaming): Test edge cases and potential failure modes to ensure the agent remains reliable and secure.

By following these practices, you can make agent behavior more predictable, accurate, and useful for end-users

QUESTION 31

A company has a cloud-based AI solution that uses Azure OpenAI models.

You need to design a monitoring solution that meets the following requirements:

Monitors performance metrics and operational health for the models
Monitors AI apps and agents for compliance
Uses Azure-native capabilities

Minimizes development effort What should you use for each requirement?

To answer, drag the appropriate options to the correct requirements. Each option may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content.

NOTE: Each correct selection is worth one point.

Select and Place:

Options

Azure API Management
Azure Policy
Microsoft Defender
Azure Monitor
Azure Stream Analytics
Microsoft Purview

Answer Area

Monitors AI app and agents for compliance: [______]

Monitors performance metrics and operational health: [______]

这是该道拖拽匹配题（Select and Place）的正确答案与详细解析：

QUESTION 答案 (Answer Area)

Monitors AI app and agents for compliance: 👉 Microsoft Purview ✅
Monitors performance metrics and operational health: 👉 Azure Monitor ✅

解析与考点分析

监控 AI 应用和 Agent 的合规性 (Monitors AI app and agents for compliance)

正确选择： Microsoft Purview

解析：在微软的原生云架构中，Microsoft Purview（Purview AI Hub）是专门用于针对 AI 应用、Azure OpenAI模型及 Copilot Agent 进行合规性监控（Compliance Monitoring）、敏感数据审计、策略违规检测以及交互日志保留的官方原生服务。

监控模型的性能指标和运行健康状况 (Monitors performance metrics and operational health)

正确选择： Azure Monitor

解析： Azure Monitor 是 Azure 原生的运维监控服务。

它原生集成了 Azure OpenAI 模型与 AI 服务，能够自动收集、分析并可视化呈现性能指标与运行健康度（Performance metrics & operational health），例如：API 响应延迟（Latency）、每分钟 Token 消耗量（TPM）、HTTP 错误率（429 Rate Limit 超限告警）以及服务可用性（Uptime）。

干扰项排除说明

Azure API Management：用于 API 网关控制、限流与路由，非合规监控或原生运维分析服务。
Azure Policy：用于定义资源部署规则和预防性合规约束，不用于持续监控 AI 应用运行时的合规内容
Microsoft Defender： 侧重于安全威胁防御与漏洞检测（Security/Threat Protection），而非合规审计（Compliance）。
Azure Stream Analytics：用于实时流数据分析处理，需要大量开发工作（违背了 Minimize development effort 的要求）。

Explanation/Reference:

Explanation:

Box 1: Microsoft Purview Monitors

AI apps and agents for compliance

To monitor Azure OpenAI models and agents for compliance with minimal effort, you can leverage Microsoft Purview through its native integration with Azure AI Foundry. This "turn-key" approach allows security administrators to capture and audit AI interactions without requiring developers to write extensive custom code.

Native Integration: The "Zero-Code" Approach The most efficient way to achieve your goal is by enabling the Microsoft Purview integration directly within your Azure subscription or AI Foundry project.

Turn-on-and-go: Azure Admins can enable a single setting at the subscription level.

Automatic Data Flow: Once enabled, prompts and responses from Azure AI-based apps flow into Purview immediately.

Centralized Hub: Use the Microsoft Purview AI Hub to visualize AI activity and detect sensitive data leaks across your entire tenant.

Box 2: Azure Monitor

Monitors performance metrics and operational health for the models

For monitoring Azure OpenAI models with an Azure-native approach and minimal effort, the most adequate solution is Azure Monitor. Specifically, utilizing the Azure OpenAI Insights workbook provides a pre-built, comprehensive dashboard for tracking operational health and performance metrics without custom code.

Azure Monitor (Platform Metrics) Azure Monitor automatically collects "Platform Metrics" for every Azure OpenAI resource at no additional cost or configuration. These provide near real-time visibility into basic operational health:

Total Requests: Track volume and status codes (e.g., 200 Success vs. 429 Rate Limited).

Latency: Monitor "Time to Response" to ensure performance meets user expectations.

Token Consumption: View "Processed Inference Tokens" and "Generated Completion Tokens" to manage costs and quota limits.

QUESTION 32

A company has Microsoft Copilot Studio agents.

The company plans to deploy custom connectors across development, test and production environments.

You need to design an application lifecycle management (ALM) process to ensure consistency and prevent direct editing in production.

Which two actions should you include in the design? Each correct answer presents part of the solution. NOTE: Each correct selection is worth one point.

A. Deploy managed solutions to production.

B. Deploy unmanaged solutions to production.

C. Manually rebuild the agents in each environment.

D. Move the agents between the environments by using data export and import.

E. Include agents and connectors in a solution.

A. Deploy managed solutions to production.
E. Include agents and connectors in a solution.

Managed solutions are intended for production deployment and help prevent direct changes to deployed components. Power Platform solutions package components for transport between environments and can include custom connectors; Copilot Studio agents are managed and deployed across environments by using solutions.

答案：✅ A、E

解析

题目要求：

在 Development、Test、Production 之间一致部署
防止 Production 被直接修改
符合 ALM 最佳实践

✅ A. Deploy managed solutions to production

正确。

Production 环境应部署 Managed Solution：

防止直接修改组件
支持版本管理
符合微软 ALM 最佳实践

✅ E. Include agents and connectors in a solution

正确。

Copilot Studio Agent 和 Custom Connector 都应放入 Solution 中统一管理，便于：

ALM Pipeline 部署
版本控制
跨环境迁移

为什么其他选项不对？

❌ B. Deploy unmanaged solutions to production
Unmanaged Solution 适用于开发环境，不适合生产环境。
❌ C. Manually rebuild the agents in each environment
无法保证一致性，也不符合 ALM。
❌ D. Move the agents by using data export/import
官方推荐使用 Solutions，不是手工导入导出数据。

考试记忆

Development → Unmanaged Solution
Production → Managed Solution
Agent / Connector → 放入 Solution 管理
跨环境部署 → ALM Pipelines + Solution

最终答案：✅ A、E

QUESTION 33

A company uses multiple Microsoft Copilot Studio agents across different channels.

You need to recommend a monitoring solution that provides comprehensive telemetry data and performance insights for the agents.

What should you include in the recommendation?

A. Application Insights

B. Microsoft Dynamics 365 Customer Voice

C. Log Analytics

D. Microsoft Purview

正确答案：✅ A. Application Insights

原因

题目关键词是：

multiple Copilot Studio agents + different channels + comprehensive telemetry + performance insights

Application Insights 用于收集详细的应用 telemetry，例如：

请求和响应
性能指标
错误和异常
Agent 运行 telemetry
跨不同渠道的运行情况

所以适合做详细监控、故障排查和性能分析。

其他选项

❌ B. Dynamics 365 Customer Voice → 主要用于调查和客户反馈。
❌ C. Log Analytics → 是日志分析平台，但本题强调的是 Agent 的comprehensive telemetry 和 performance insights，Application Insights 更直接。
❌ D. Microsoft Purview → 数据治理、分类、合规，不是 Agent 性能监控。

🎯 最终答案

A. Application Insights

Correct Answer: A

Explanation/Reference:

Explanation:

Azure Application Insights (part of Azure Monitor) is the best Microsoft solution for monitoring comprehensive telemetry and performance for Copilot Studio agents across multiple channels.

While Copilot Studio has built-in analytics, Application Insights allows you to centralize data from many different agents and channels into one unified view.

Why Use Azure Application Insights?

Connecting your agents to Application Insights provides deeper technical and operational visibility that the standard dashboard lacks:

Centralized Monitoring: View telemetry for all your agents in a single Azure resource rather than checking each agent individually in Copilot Studio.

Custom Kusto Queries: Use KQL (Kusto Query Language) to run complex analyses, such as identifying common errors, measuring specific topic latency, or tracking custom events.

Real-Time Alerts: Set up proactive alerts to notify you if an agent has a high error rate, spike in latency, or fails to respond.

Enhanced Transcripts: Capture detailed message-level logs, including user prompts and agent responses, to troubleshoot specific conversation failures.

QUESTION 34

A company has an AI solution named Solution1 that is deployed to the production environment. Solution1 uses an Azure OpenAI model to generate marketing emails for existing customers.

During an internal review, you identify that Solution1 creates different emails depending on the customers’ traits.

You need to recommend a strategy to mitigate the bias. The strategy must adhere to Microsoft responsible AI principles.

What should you recommend?

A. Modify Solution1 to randomly generate emails for different traits.

B. Modify the system instructions of Solution1.

C. Retrain the model by using a larger dataset.

D. Modify the contents of the training dataset.

正确答案：

B. Modify the system instructions of Solution1.（修改 Solution1 的系统指令。）

解析与考点分析

为什么选 B？

业务场景与问题：
- 系统在使用 Azure OpenAI 模型为现有的客户自动生成营销电子邮件。
- 审阅发现生成的内容存在偏见/不公平现象（Bias）：模型根据客户的特定特征（Traits）生成了不公或带有偏见偏向的邮件内容。
- 目标：根据微软负责任 AI 原则（Microsoft Responsible AI Principles - 公平性 Fairness & 包容性 Inclusiveness）减缓偏见。
生成式 AI（LLM）缓解偏见的标准做法：
- 系统指令（System Instructions / Metaprompt）：在 Azure OpenAI中，系统指令定义了模型的角色、行为准则、语气约束以及安全/公平性边界。
- 通过调整系统指令（系统 Prompt），明确要求模型在生成营销内容时“保持公平、包容，不得基于客户的个人特征（如性别、年龄、地域等特征）产生刻板印象或区别对待”，这是在生产环境中缓解生成式 AI 偏见最直接、成本最低且即时生效的官方推荐策略。

为什么不选其他选项？

A. Modify Solution1 to randomly generate emails for different traits（随机生成邮件）：随机生成无法解决偏见问题，反而会导致输出内容不可控且失去营销相关性。
C. Retrain the model by using a larger dataset（使用更大的数据集重新训练模型）：
D. Modify the contents of the training dataset（修改训练数据集的内容）：
- Azure OpenAI 使用的是预训练好的大语言基础模型（Foundation Models）。用户在生产环境中通常直接调用 API 或进行 Prompt 调整，而不是从零开始去重训（Retrain）基础大模型。使用系统指令（System Message）进行安全防护是优先且最标准的方法。

Correct Answer: B Section: (none)

Explanation/Reference:

Explanation:

To reduce bias in your Azure OpenAI email generation system while adhering to Microsoft’s Responsible AI (RAI) Principles, you should implement a multi-layered approach within your System Instructions.

The goal is to move from generic prompts to specific, constraint-based instructions that enforce Fairness and Inclusiveness.

QUESTION 35

A company has a canvas app named App1 in a Microsoft Power Platform environment named Env1. Env1 uses a customer-managed key for data encryption.

App1 connects to multiple data sources to retrieve and update customer and order information.

You need to recommend a solution to add Microsoft Copilot components to App1.

The solution must NOT modify the current security or encryption configurations of Env1.

What should you include in the recommendation?

A. Modify the data sources of App1 to make them compatible with Copilot.

B. Duplicate App1 and republish the app in Env1.

C. Enable Copilot features for Env1.

D. Move App1 to a new environment that uses Microsoft-managed keys.

正确答案与详细解析

正确答案：

C. Enable Copilot features for Env1.（为 Env1 启用 Copilot 功能。）

解析与考点分析

为什么选 C？

业务需求与限制条件：
- 需要在 Canvas App（App1）中添加 Microsoft Copilot 组件（控件）。
- 限制条件：绝对不能修改 Env1 当前的安全或加密配置（NOT modify the current security or encryption configurations）。当前 Env1 使用的是客户托管密钥（CMK, Customer-Managed Keys）。
微软 Power Platform 配置规范：
- 要在特定环境中的 Canvas App 中加入 Copilot 控件或让用户使用 Copilot 功能，前提条件是在 Power Platform 管理中心（PPAC）中对该环境（Env1）开启 Copilot 特性开关（Enable Copilot features for the environment）。
- Power Platform 完全支持在开启了 CMK（客户托管密钥）加密的环境中使用 Copilot 功能。
- 直接为 Env1 开启 Copilot 开关，既能在不破坏现有 CMK 加密配置和安全架构的前提下满足需求，又是实现该功能的最简单、最符合规范的做法。

为什么不选其他选项？

A. Modify the data sources of App1 to make them compatible with Copilot（修改 App1 的数据源）： Copilot 控件能够基于 Canvas App 现有的多种标准数据源工作，无需修改或替换底层数据源。
B. Duplicate App1 and republish the app in Env1（复制 App1 并重新发布）：如果环境层面的 Copilot 功能开关未开启，复制应用没有任何作用。
D. Move App1 to a new environment that uses Microsoft-managed keys（将 App1 迁移到使用微软托管密钥的新环境）：这直接违背了“不修改当前加密配置”的要求，属于不必要且破坏现有安全合规性的错误操作。

Correct Answer: D Section: (none)

Explanation/Reference:

Explanation:

To add Microsoft Copilot components to your canvas app without changing the current environment's security or encryption (Customer-Managed Key), you should move the app to a new environment that uses Microsoft-managed keys.

Core Technical Limitation Microsoft Power Platform environments configured with a Customer-Managed Key (CMK) currently do not support certain Copilot and generative AI features. These features continue to rely on Microsoft-managed encryption keys for their specific internal data processing.

QUESTION 37

A company has a team that analyzes its customers by using a manual process.

You are designing an AI-based agent to automate and improve the process.

You need to recommend on which platform to build the agent. The solution must meet the following requirements:

Use generative AI to answer common questions.
Provide analytics to review AI performance.
Identify customer demographics.
Minimize custom development.

Solution: You recommend Microsoft Security Copilot.

Does this meet the goal?

A. Yes

B. No

✅ 答案：B. No

Microsoft Security Copilot 是面向网络安全场景的 AI 平台，不适合用于客户分析、客户 demographics 等业务场景。

题目要求：

Generative AI 回答常见问题
AI performance analytics
Customer demographics
Minimize custom development

更适合使用 Microsoft Copilot Studio 来构建业务 Agent。

🧠 记忆：

Security Copilot → Cybersecurity Business AI Agent → Copilot Studio

Correct Answer: B Section: (none)

Correct:

You recommend Microsoft Copilot Studio.

Copilot Studio is specifically designed to create customizable AI agents that automate business workflows with minimal coding.

Generative AI Answers: It uses "Generative Answers" to scan your specific data (websites, files, or SharePoint) to answer customer-related questions instantly.

Analytics Dashboard: It includes built-in analytics to track resolution rates, customer satisfaction (CSAT), and overall AI performance.

Customer Demographics: You can configure the agent to extract specific entities (like age, location, or industry) from conversations to build demographic profiles.'

Low-Code/No-Code: It minimizes custom development by providing a visual interface to build logic, rather than writing raw code.

Incorrect:

You recommend GitHub Copilot.
You recommend Microsoft Security Copilot.

Low-Code/No-Code: It minimizes custom development by providing a visual interface to build logic, rather than writing raw code.

Incorrect:

You recommend GitHub Copilot.
You recommend Microsoft Security Copilot.

QUESTION 38

A company has a team that analyzes its customers by using a manual process.

You are designing an AI-based agent to automate and improve the process.

You need to recommend on which platform to build the agent. The solution must meet the following requirements:

Use generative AI to answer common questions. Provide analytics to review AI performance. Identify customer demographics.

Minimize custom development.

Solution: You recommend Microsoft Copilot Studio.

Does this meet the goal?

A. Yes

B. No

✅ 答案：A. Yes

Microsoft Copilot Studio 符合要求：

Generative AI → 回答常见问题
Analytics → 查看 Agent 性能
Customer demographics → 可连接 Dataverse / 业务数据进行分析
Minimize custom development → Low-code / no-code

答案：A. Yes ✅

Correct Answer: A Section: (none)

Explanation/Reference:

Explanation:

Correct:

You recommend Microsoft Copilot Studio.

Copilot Studio is specifically designed to create customizable AI agents that automate business workflows with minimal coding.

Generative AI Answers: It uses "Generative Answers" to scan your specific data (websites, files, or SharePoint) to answer customer-related questions instantly.

Analytics Dashboard: It includes built-in analytics to track resolution rates, customer satisfaction (CSAT), and overall AI performance.

Customer Demographics: You can configure the agent to extract specific entities (like age, location, or industry) from conversations to build demographic profiles.' Low-Code/No-Code: It minimizes custom development by providing a visual interface to build logic, rather than writing raw code.

Incorrect:

You recommend GitHub Copilot.
You recommend Microsoft Security Copilot.

QUESTION 39

A company uses multiple Microsoft Copilot Studio agents across different channels.

You need to recommend a monitoring solution that provides comprehensive telemetry data and performance insights for the agents.

What should you include in the recommendation?

A. Microsoft Power BI

B. the Analytics tab in Copilot Studio

C. Log Analytics

D. Azure Advisor

正确答案与详细解析

正确答案：

B. the Analytics tab in Copilot Studio（Copilot Studio 中的“分析”选项卡）

解析与考点分析

为什么选 B？

业务场景：公司在不同渠道上部署了多个 Microsoft Copilot Studio Agent，需要一个监控解决方案来提供全面的遥测数据（Comprehensive telemetry data）和性能洞察（Performance insights）。
开箱即用的原生功能：
- Microsoft Copilot Studio 内置了强大的 Analytics（分析）选项卡。
- 无需编写代码或进行复杂的 Azure 资源配置，管理员和开发者即可直接在 Analytics 选项卡中查看到跨渠道部署的所有 Agent 的开箱即用遥测数据与性能洞察，包括：
  - Summary（概要）：参与率、解决率、转人工升级率、放弃率。
  - Topic Usage（主题使用情况）：各 Topic 的触发频率与解决效果。
  - Customer Satisfaction（客户满意度）：CSAT 评分与用户反馈分析。
  - Sessions（会话）：完整的会话轨迹（Transcripts）与结果分类。

为什么不选其他选项？

A. Microsoft Power BI： Power BI 是商业智能报表工具。虽然可以通过导出数据或调用 API 在 Power BI 中自定义报表，但它不是 Copilot Studio 原生内置的实时遥测与性能洞察监控入口。
C. Log Analytics（日志解析）： Log Analytics 是 Azure Monitor 的底层日志查询数据库。如果题目选项中包含 Application Insights，Application Insights 是捕捉遥测的标准服务；但在不引入额外 Azure 运维配置的前提下，Copilot Studio 内置的 Analytics tab 是直接获取 Agent 性能洞察的最直接答案。
D. Azure Advisor（Azure 顾问）：仅针对 Azure 基础设施架构提供成本、安全性和可用性优化建议，不提供 Copilot Studio Agent 的对话遥测数据。

Question 40

Which tool should be used for the prospect-communication requirements in Dynamics 365 Sales?

A Azure Al Search
B Copilot email assist
C the Voice template Microsoft Copilot Studio
D Deep Research in Microsoft Foundry Agent Service

B Copilot email assist helps Dynamics 365 Sales sellers compose customer-specific email content and is supported for lead and opportunity records through Dynamics 365 email, directly enabling faster prospect correspondence.

正确答案：✅ B. Copilot email assist

原因

题目关键词是 prospect communication（与潜在客户沟通），尤其是 Dynamics 365 Sales 中：

起草/回复客户邮件
根据销售上下文生成邮件
加快与 prospects 的沟通

这些正是 Copilot email assist 的用途。

其他选项：

❌ Azure AI Search → 企业知识检索
❌ Copilot Studio Voice template → 语音 Agent
❌ Deep Research → 深度研究/信息收集

考试记忆：Dynamics 365 Sales + Prospect Email → Copilot email assist。

Question 41

A company has an Azure environment that supports several business units.

The company intends to implement an Al solution that will perform sentiment analysis on customer product reviews.

You need to assess the solution's potential cost to support return on Al investment (ROAI) analysis.

What should you use?

A Total Cost of Ownership (TCO) Calculator
B Azure Reservations
C Azure pricing calculator
D Azure Monitor

正确答案：✅ C. Azure pricing calculator

原因

题目要求：

Assess the solution's potential cost 用于 ROAI（Return on AI Investment）分析

需要估算 AI 方案未来可能产生的 Azure 服务成本，例如：

Azure AI Language / sentiment analysis
Storage
Compute
API usage

Azure Pricing Calculator 可以根据预计的资源和使用量估算成本。

其他选项

❌ A. TCO Calculator → 比较迁移到 Azure 前后的总体拥有成本（TCO），不是针对新 AI 方案的服务成本估算。
❌ B. Azure Reservations → 降低长期资源成本，不用于成本预测。
❌ D. Azure Monitor → 监控实际运行情况，不是成本估算工具。

记忆

新方案成本预估 → Azure Pricing Calculator ✅ On-prem → Azure 成本对比 → TCO Calculator ✅

答案：C

Question 42

A company operates an ecommerce support portal that uses Microsoft Dataverse.

You are designing a Microsoft Copilot Studio agent for the portal. The agent must meet these requirements:

Respond with a default help message when user input is unclear.
Start external processes, such as retrieving order status, when users make specific requests.

Generative orchestration will be enabled for the solution.

Recommend a feature for each requirement. Each feature may be used once, more than once, or not at

Drag & Drop

A tool (connector)
A trigger phrase
The Fallback topic
A skill
The Escalate topic

Answer area

Respond with a default help message when the user input is unclear: [___]

Initiate external processes when requested: [___]

答案：

Respond with a default help message when the user input is unclear: ✅ The Fallback topic
Initiate external processes when requested: ✅ A tool (connector)

关键点

Fallback topic → 当 Agent 无法理解用户输入时提供默认响应。
Tool (connector) → 在 generative orchestration 下，Agent 可以根据用户意图选择并调用 Connector 来执行外部操作，例如查询订单状态。

最终：

Unclear input → Fallback topic External process → Tool (connector)

Question 43

A company has an Azure environment that supports multiple business units.

The company plans to implement an Al solution that will perform sentiment analysis on customer product reviews.

You need to assess the solution's potential cost to support return on Al investment (ROAI) analysis.

What should you use?

A Azure Reservations
B Microsoft Fabric SKU Estimator
C Anomaly Detection in Azure Cost Management
D Azure pricing calculator

正确答案：✅ D. Azure pricing calculator

原因

题目关键词是：

assess the solution's potential cost support ROAI analysis

这是在 AI 方案实施之前估算预计成本，因此使用 Azure Pricing Calculator。

D. Azure Pricing Calculator ✅ → 估算 Azure 服务的预计成本，适合做 ROI/ROAI 规划。
A. Azure Reservations ❌ → 用于购买预留容量、降低长期成本。
B. Microsoft Fabric SKU Estimator ❌ → 用于估算 Fabric 容量，不是一般 Azure AI 方案。
C. Anomaly Detection in Azure Cost Management ❌ → 用于发现异常成本，不是事前成本估算。

考试记忆

Potential / estimated Azure cost → Azure Pricing Calculator Actual spending / cost analysis → Cost Management + Billing

答案：D. Azure pricing calculator

Question 44

A company uses a Microsoft Copilot Studio prompt-and-response agent.

You need to ensure that the agent meets these requirements:

Provides effective and relevant responses
Provides conversational outcomes

Which metric should be used for each requirement?

Provides effective and relevant responses: []

Generated answer rate and quality
Reactions
Tool use

Provides conversational outcomes: []

Satisfaction
Tool use
Topics by outcome

答案：

Provides effective and relevant responses: ✅ Generated answer rate and quality
Provides conversational outcomes: ✅ Topics by outcome

记忆

Generated answer rate and quality → 看 AI 生成回答是否有效、相关、质量好。
Topics by outcome → 看不同 Topic 最终是 resolved / abandoned / escalated 等结果。

最终：

① Generated answer rate and quality ② Topics by outcome

Question 45

A company intends to deploy a Microsoft Foundry agent.

You need to recommend an application lifecycle management (ALM) process that ensures the agent is evaluated against baseline accuracy metrics before deployment.

What should you recommend?

A Configure GitHub Actions for new agent versions.
B Deploy each new agent version directly to production.
C Use Observability in Foundry Control Plane with evaluation and drift monitoring.
D Enable Application Insights and use Azure Monitor.

正确答案：✅ C. Use Observability in Foundry Control Plane with evaluation and drift monitoring.

原因

题目的关键是：

evaluated against baseline accuracy metrics before deployment

需要在 Agent 发布前进行评估（Evaluation），并持续监控模型/Agent 是否发生 drift。

C 正好覆盖：

Baseline evaluation metrics
Evaluation before/around deployment
Drift monitoring
Agent/AI 生命周期中的质量监控

其他选项

❌ A. GitHub Actions → 可以做 CI/CD，但本身不提供 baseline accuracy evaluation。
❌ B. Directly to production → 明显违反测试/评估后再发布的要求。
❌ D. Application Insights + Azure Monitor → 主要用于应用性能、日志和 telemetry monitoring，不是专门的 AI baseline evaluation。

🎯 考试记忆

Baseline accuracy + Evaluation + Drift → Foundry Control Plane Observability

答案：C

Question 46

Which tool should be recommended to address concerns about sensitive information in the sales process?

A the Analytics tab in Microsoft Copilot Studio
B Model Context Protocol (MCP)
C Application Insights
D Microsoft Foundry Tracing UI
E Monitoring in Microsoft Foundry

正确答案：✅ E. Monitoring in Microsoft Foundry

题目关键词是 sensitive information / sales process，需要监控 Agent 运行过程中涉及的数据和安全风险。

E. Monitoring in Microsoft Foundry ✅ → 用于监控 AI Agent 的运行情况，包括安全、数据访问等方面。
A. Copilot Studio Analytics ❌ → 主要分析 Agent 使用情况、会话、满意度等。
B. MCP ❌ → 用于连接 Agent 与外部数据/工具，不是监控工具。
C. Application Insights ❌ → 主要是应用 telemetry/performance。
D. Foundry Tracing UI ❌ → 主要查看 Agent 执行 trace，不是主要的治理/敏感信息监控工具。

答案：E. Monitoring in Microsoft Foundry

Question 47

A company has a team that analyzes its customers using a manual process.

You are designing an AI-based agent to automate and improve this process.

You need to recommend which platform to use to build the agent. The solution must meet the following requirements:

Use generative AI to answer common questions.
Provide analytics to review AI performance.
Identify customer demographics.
Minimize custom development.

Solution: Microsoft Copilot Studio is recommended.

Does this satisfy the goal?

A Yes
B No

✅ 答案：A. Yes

Microsoft Copilot Studio 符合要求：

Generative AI → 回答常见问题
Analytics → 分析 Agent 的性能
Customer demographics → 可连接业务数据进行客户分析
Minimize custom development → Low-code / no-code Agent 开发

🧠 记忆：

Business Agent + Generative AI + Analytics + Low-code → Copilot Studio

答案：A. Yes