Microsoft AB-100 Online Practice Questions and Exam Preparation

Microsoft AB-100 Online Questions & Answers

• Question 21:

  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
  C. Total Cost of Ownership (TCO) Calculator

  ---

  Explanation

  The Total Cost of Ownership Calculator is appropriate when the task is to evaluate costs for an Azure-based AI solution across business units. TCO analysis focuses on the cost side of the business case, including resources, operations, and migration or modernization impact. It is not a model quality or agent analytics tool. Use TCO when the scenario asks for cost evaluation rather than adoption or response quality.

• Question 22:

  A company plans to deploy a Microsoft Foundry agent.

  You need to recommend an application lifecycle management (ALM) process to ensure that the agent evaluates against baseline accuracy metrics before being deployed.

  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.
  C. Use Observability in Foundry Control Plane with evaluation and drift monitoring.

  ---

  Explanation

  Foundry observability with evaluation and drift monitoring supports an ALM gate that checks baseline accuracy before deployment. The requirement is not just to deploy the agent, but to verify quality against expected metrics. Direct production deployment would skip the control, and generic monitoring after release would be too late. AI ALM should include pre-release evaluation plus ongoing drift detection.

• Question 23:

  HOTSPOT

  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.

  

  

  ---

  The completed hotspot selects Microsoft Copilot Studio for AI agent creation and a Fallback topic for unexpected agent actions. The answer image uses Copilot Studio because the requirement is to build and configure a business agent in the low-code authoring environment. The Fallback topic handles user inputs or paths that do not match expected topics, which helps control unexpected conversation behavior. Microsoft Foundry and Dynamics 365 Sales are not the right authoring and conversation-control combination for this requirement.

• Question 24:

  A finance organization uses generative help and guidance in Dynamics 365 Finance. Users need the help experience to answer questions from company-specific procedure files in addition to the default product documentation.

  What should you recommend?

  A. Add the procedure files as knowledge in Copilot Studio for the finance and operations apps agent.
  B. Disable the out-of-box conversational boosting topic.
  C. Export all finance transactions to a public website.
  D. Replace the Finance app with a standalone chatbot.
  A. Add the procedure files as knowledge in Copilot Studio for the finance and operations apps agent.

  ---

  Explanation

  Adding company procedure files as knowledge for the finance and operations apps agent lets generative help answer using the organization's own process material. The default product documentation is not enough when users ask about internal procedures. Uploading or configuring the approved files grounds responses without changing application code. For in-app help, add company-specific documents as knowledge.

• Question 25:

  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:

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

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

  3. 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
  C. Microsoft Copilot Studio and AI Builder

  ---

  Explanation

  Microsoft Copilot Studio and AI Builder meet the invoice-processing requirements with low-code tools. AI Builder can extract structured invoice data from documents, while Copilot Studio and Power Platform automation can guide Teams-based review and approval workflows. A fully custom Azure Document Intelligence and Logic Apps design could work, but it would require more development effort. The exam clue is the combination of document AI, approvals, Teams interaction, and minimized development.

• Question 26:

  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
  A. Computer Use in Copilot Studio

  ---

  Explanation

  Computer Use is appropriate because the agent is automating a web application through the interface, and the failures are caused by UI changes. Computer Use is designed for screen-based interaction when an API or connector is not available or practical. Conversation topics and agent flows do not make screen automation resilient by themselves. The key is that the target system is being operated through its UI rather than a stable service endpoint.

• Question 27:

  DRAG DROP

  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:

  

  

  ---

  The completed drag-and-drop maps unanswered questions to generated answer rate and quality and abandoned sessions to conversation outcomes. The answer image is using Copilot Studio analytics categories: generated-answer quality explains how often knowledge answers work, while outcomes classify sessions as resolved, escalated, abandoned, or similar results. Reactions and satisfaction help with feedback, but they do not directly count unanswered questions or abandoned sessions. Choose the metric that matches the session or answer behavior being measured.

• Question 28:

  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.
  B. Modify the system instructions of Solution1.

  ---

  Explanation

  Modifying the system instructions of Solution1 is the right remediation when the generated marketing emails violate internal policy. System instructions define the agent or solution's behavioral constraints and can direct the model away from prohibited claims, tone, or content. Changing only monitoring or deployment settings would not change the generated text behavior. Policy-related generation issues should be corrected in the controlling instructions and then retested.

• Question 29:

  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.
  B. Create a connection to Microsoft Foundry in the agent.

  ---

  Explanation

  Creating a connection to Microsoft Foundry in the Copilot Studio agent gives the agent a secure way to invoke the custom Foundry model during conversations. The requirement is runtime connectivity to a model, not a change from generative to classic orchestration. A skill or custom engine agent would be a different architecture and does not directly describe the secure connection from the existing Copilot Studio agent. Use the supported connection/tool integration path when Copilot Studio must call a model-hosted service.

• Question 30:

  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)
  D. AI quality (AI assisted)

  ---

  Explanation

  AI quality with AI-assisted evaluation uses a judge model such as GPT-4o to score outputs for quality dimensions like coherence. That matches the requirement for a numeric score produced by a model evaluator. NLP metrics and groundedness cover narrower or different checks, and risk and safety metrics focus on harmful content. Use AI-assisted quality metrics when a judge model evaluates generated text quality.