Q1.The IT department of a retail company is building an AI-powered customer support agent.
You need to ensure that this solution reliably complies with Microsoft’s responsible AI principles.
Which two actions should you perform?
Select the two most appropriate options.
Each option forms part of the solution.
Show answer
Microsoft’s responsible AI principles emphasize fairness and transparency.
Option B is correct because verifying that responses are considerate of diverse users and cultures ensures fairness and inclusiveness and reduces the risk of bias or discriminatory expression.
In addition, Option E is correct because clearly indicating that users are interacting with AI ensures transparency and prevents misunderstanding or excessive trust.
Option A may violate the principle of data minimization and is inappropriate.
Option C weakens human oversight and undermines accountability.
Option D increases risk by expanding the scope of use and is therefore not appropriate.
Responsible AI in Microsoft Foundry
Q2.Which service should you use for each of the following tasks?
In the answer area, select the appropriate option for each item.
– Extract structured data from forms and invoices
– Summarize the body content of a business report
– Add captions to uploaded images
Show answer
This question tests whether you correctly understand the use cases of Azure AI services.
Azure Document Intelligence (formerly Form Recognizer) is a service that extracts structured data from documents such as forms and invoices, making it ideal for this use case.
In addition, Azure Language handles text analytics and can perform text summarization and natural language processing, making it suitable for summarizing business reports.
Furthermore, Azure Vision handles image analysis and provides descriptions of image content (caption generation).
Each service has a clearly distinct role, so selecting the right one for the use case is important.
Microsoft Foundry documentation
Q3.Select the option that correctly completes the following sentence.
Show answer
A Microsoft 365 Copilot connector is a mechanism that ingests external data into Microsoft Graph so it can be searched and used across Microsoft 365, so the connection target is Microsoft Graph.
In addition, the ingested data can be searched and summarized in natural language from Copilot and Microsoft Search, functioning as a foundation for leveraging LOB data.
Azure AI Search is for custom search such as RAG, Microsoft Purview is for governance, and Microsoft SharePoint is one storage location; none of these is the appropriate connection target for this question.
Overview of Microsoft 365 Copilot connectors
Q4.An organization is considering using Microsoft Copilot Studio to automate business processes and provide employee self-service support.
Which capabilities does Copilot Studio directly support?
Select the two that constitute a complete solution.
Show answer
Copilot Studio is a service for building custom agents that connect to business data and automate processing, and Option D—business data integration and automation—is a core capability, so it is correct.
In addition, it also provides customization capabilities to adjust an agent’s conversation flow, responses, guardrails, and more, so Option E is also correct.
Option A is the domain of Security Copilot, Option B is information protection functionality such as Microsoft Purview, and Option C is Microsoft 365 Copilot functionality (Word and PowerPoint); none of these are direct capabilities of Copilot Studio.
Overview of Copilot Studio
Q5.For each of the following statements, select Yes if it is correct, or No if it is incorrect.
| Statement | Yes | No | |
|---|---|---|---|
| A Microsoft 365 Copilot license is required for users to access organizational data from a mobile device. | |||
| A Microsoft 365 Copilot license is required to reason over organizational data using Microsoft Graph. | |||
| A Microsoft 365 Copilot license is required to use the Analyst agent. |
Show answer
Simply accessing organizational data from a mobile device is possible with a standard Microsoft 365 license, and a Copilot license is not required, so No is correct.
On the other hand, reasoning that leverages Microsoft Graph and advanced Copilot capabilities (including the Analyst agent) are provided by the Microsoft 365 Copilot license, so Yes is correct.
Copilot understands and leverages organizational data through Graph, and these advanced capabilities depend on the additional license.
Overview of Microsoft 365 Copilot
Q6.For each of the following statements, select Yes if it is correct, or No if it is incorrect.
| Statement | Yes | No | |
|---|---|---|---|
| Microsoft Copilot in Outlook can automatically send follow-up replies to emails. | |||
| Microsoft Copilot in Outlook can identify and display tasks mentioned within an email. | |||
| Microsoft Copilot in Outlook can summarize long email threads and highlight action items and deadlines. |
Show answer
Copilot in Outlook can help draft emails and assist users, but it has no capability to automatically send follow-up replies, and sending requires user action, so No is correct.
On the other hand, summarizing email content, extracting key items (tasks and deadlines), and surfacing actions are primary capabilities of Copilot, and they improve work efficiency based on contextual understanding, so Yes is correct.
Copilot is designed to assist users, and fully automated processing such as autonomously sending emails is not intended.
Draft an email message using Copilot in Outlook
Q7.A company is considering deploying a proof-of-concept (PoC) agent that uses Azure OpenAI.
This solution needs to start small and scale flexibly as demand grows.
Which pricing model should you choose?
Show answer
In the PoC phase, usage is uncertain, and a pay-as-you-go model that lets you start small and scale as needed is best, so Standard On-Demand is correct.
In addition, Standard On-Demand bills based on token usage, keeps initial costs low, and adapts flexibly to fluctuating demand, making it suitable for the validation phase.
Provisioned PTUs is a fixed-capacity model for production, Batch API is for asynchronous processing, and Microsoft 365 Copilot is a separate SaaS license; none of these are suitable for this case.
Deployment types for Microsoft Foundry models
Q8.Select the appropriate answer to make the sentence correct.
Show answer
High-quality grounding data generates responses based on trustworthy information sources, so the key point is that it improves the accuracy and reliability of the AI’s predictions and outputs.
In addition, providing appropriate context reduces the model’s guessing (hallucination) and enables more consistent, practical answers.
However, it does not guarantee the accuracy of all responses, nor does it directly increase the model’s processing performance or storage requirements.
Azure OpenAI On Your Data (classic)
Q9.For each of the following statements, select Yes if it is correct, or No if it is incorrect.
| Statement | Yes | No | |
|---|---|---|---|
| A text-to-image generation model can be used for the purpose of translating content into another language. | |||
| A predictive analytics model can create new marketing content for a company’s online advertising. | |||
| A generative AI chatbot can interact with customers in personalized conversations and recommend products. |
Show answer
A text-to-image generation model is intended for image generation, and translation belongs to the domain of natural language processing with a different purpose, so No is correct.
In addition, a predictive analytics model predicts future trends from historical data, and generating new content is the role of generative AI, so the answer is No.
On the other hand, a generative AI chatbot can hold natural language conversations and make recommendations, making it suitable for advanced customer support, so Yes is correct.
Foundry Models sold directly by Azure
Q10.Which option correctly explains the difference between a pretrained generative AI model and a fine-tuned generative AI model?
Show answer
The key point is that a pretrained model is a foundation model trained on large and diverse data, acquiring general patterns of language and knowledge.
In addition, fine-tuning is a technique that further trains that foundation model on data from a specific task or domain, enabling more specialized and higher-accuracy responses.
Option A is wrong about the data requirement, Option B misunderstands the training characteristics, and Option D has the roles reversed, so they are incorrect.
Foundry Models sold directly by Azure