Microsoft AB-731 Online Practice Questions and Exam Preparation

Microsoft AB-731 Online Questions & Answers

• Question 11:

  HOTSPOT

  You are evaluating the reliability of responses generated by a generative AI solution. The solution sometimes generates responses that appear fluent and realistic, but the responses include facts that are inaccurate or unsupported.

  Select the answer that correctly completes the sentence.

  

  

  ---

  The answer image selects model inaccuracy. Fluent but unsupported or incorrect generated content is an accuracy problem often discussed as hallucination risk; it is not data leakage, overreliance, or prompt injection by itself.

• Question 12:

  Your company plans to use generative AI to help project managers and engineers work with construction blueprints stored as PDF files.

  You need to recommend a generative AI solution that meets the following business requirements:

  1. Processes both images and text

  2. Summarizes the design of a building

  3. Answers user questions about a building's design

  4. Extracts information from blueprints, such as the location of electrical, heating, and plumbing systems.

  What should you recommend?

  A. a multi-modal solution
  B. an optical character recognition (OCR) solution
  C. a text completion solution
  D. a document summarization solution
  A. a multi-modal solution

  ---

  Explanation

  The requirement includes images, text, summarization, question answering, and extraction from blueprints, so the solution must be multimodal. A document summarization capability alone would not cover visual reasoning and targeted extraction from building-design images.

• Question 13:

  HOTSPOT

  You are evaluating the appropriate use cases for generative AI in an organization. The organization wants to understand which tasks are suitable for generative AI and which tasks require other types of systems or specialized models.

  For each of the following statements, select Yes if the statement is true. Otherwise, select No.

  NOTE: Each correct selection is worth one point.

  

  

  ---

  The answer image should be Yes, Yes, and No. Generative AI can assist with complex workflow automation and new content creation, but specialized mathematical or deterministic systems are usually better for formal calculations and equation solving.

• Question 14:

  You have a historical dataset that contains 1,000 records.

  You need an AI solution that can analyze the data to identify patterns and predict future outcomes.

  What should you include in the solution?

  A. Microsoft Foundry
  B. Azure Document Intelligence in Foundry Tools
  C. Azure Machine Learning
  D. Azure Content Understanding in Foundry Tools
  C. Azure Machine Learning

  ---

  Explanation

  Azure Machine Learning is appropriate when historical data must be analyzed to learn patterns and predict future outcomes. Document Intelligence, Foundry content tools, and general platform selection do not replace a predictive machine learning model for this requirement.

• Question 15:

  HOTSPOT

  A company wants to use a generative AI solution to answer employee questions by using the company's current internal documents. The company does not want to retrain the base model.

  For each of the following statements, select Yes if the statement is true. Otherwise, select No.

  NOTE: Each correct selection is worth one point.

  

  

  ---

  The answer image should be No, Yes, and Yes. RAG does not require retraining the large language model; it retrieves relevant external or enterprise content at runtime and uses that content to ground more contextual responses.

• Question 16:

  Your company is developing an AI-powered customer support agent.

  You need to ensure that the solution follows Microsoft responsible AI principles.

  Which two actions should you perform?

  A. Retain all customer conversations.
  B. Test the agent to ensure that responses are inclusive and culturally sensitive.
  C. Enable the agent to operate independently.
  D. Ensure that the agent can be used for multiple purposes.
  E. Provide a clear disclaimer that users are interacting with an AI solution.
  B. Test the agent to ensure that responses are inclusive and culturally sensitive.
  E. Provide a clear disclaimer that users are interacting with an AI solution.

  ---

  Explanation

  Inclusive testing helps ensure the support agent works respectfully for different users, and a clear AI disclosure supports transparency. Retaining all conversations, removing oversight, or broadening the agent without controls does not by itself satisfy responsible AI.

• Question 17:

  HOTSPOT

  A company is estimating the cost of building and running generative AI solutions in Azure. The company wants to understand how data volume, token usage, and model capabilities affect cost.

  For each of the following statements, select Yes if the statement is true. Otherwise, select No.

  NOTE: Each correct selection is worth one point.

  

  

  ---

  The answer image should be Yes, Yes, and No. Larger data can raise Azure Machine Learning cost, token processing drives many Azure OpenAI costs, and generative AI cost varies by model capability and usage rather than staying the same.

• Question 18:

  Your company has a Microsoft 365 subscription and uses Microsoft 365 Copilot Chat.

  Some users need to build and use declarative agents that can access work data.

  Which type of license should you recommend for the users?

  A. a Microsoft 365 Copilot add-on license
  B. Microsoft Copilot Studio user license
  C. a Copilot Chat pay-as-you-go plan
  A. a Microsoft 365 Copilot add-on license

  ---

  Explanation

  Using declarative agents that reason over work data requires Microsoft 365 Copilot licensing for those users. A basic Microsoft 365 entitlement or pay-as-you-go chat plan does not provide the same work-data grounded agent capability.

• Question 19:

  HOTSPOT

  A data science team is preparing training data for generative AI models. The team wants to understand how data quality and representativeness can affect model performance, reliability, and cost.

  For each of the following statements, select Yes if the statement is true. Otherwise, select No.

  NOTE: Each correct selection is worth one point.

  

  

  ---

  The answer image should be Yes, Yes, and Yes. Poor data quality can increase training cost, overly large models can increase cost and complexity, and non-representative data can produce inaccurate or unbalanced outputs.

• Question 20:

  You need to create a custom Azure Machine Learning model. The data used to train the model is consistent and uniform.

  What should you do first?

  A. Prepare the training data.
  B. Evaluate the model.
  C. Train the model.
  D. Tune hyperparameters.
  E. Deploy the model.
  A. Prepare the training data.

  ---

  Explanation

  A custom machine learning project starts with preparing suitable training data. Evaluation, training, hyperparameter tuning, and deployment come later because the model first needs consistent data that represents the problem it must learn.