10 free Microsoft AB-731 practice questions with the correct answer and a full explanation for each, taken from the CertStash pack of 100 questions. Work through them, then open each answer to check your reasoning.
Get all 100 questions (US$39) · Download these 10 as a PDF
Question 1
Your company plans to build a generative AI solution based on internal data.
You recommend using Microsoft Foundry as a starting point to develop and manage the solution.
What is a key benefit of using Microsoft Foundry for this project?
Show answer and explanation
Correct answer: A. Provides a scalable platform for developing and deploying generative AI solutions.
Microsoft Foundry provides a scalable platform specifically designed for developing and deploying generative AI solutions. While it may offer low-code capabilities and support for business users, the primary key benefit is its scalability and comprehensive platform for the full lifecycle of generative AI solution development. Option B is incorrect because foundational models still require selection and configuration. Option C overstates accessibility, while some low-code features exist, it's not primarily positioned for business users without technical background. Option D, while partially true, is less comprehensive than the scalability focus.
Why the other options are wrong
- B. Foundry still requires selection and configuration of underlying AI models to customize solutions for specific needs.
- C. While Foundry includes low-code elements, it primarily targets development teams rather than non-technical business users.
- D. Although low-code capabilities exist, scalability of the platform for enterprise deployment is the primary key benefit.
Question 2
Which business requirement most closely relates to grounding a generative AI model?
Show answer and explanation
Correct answer: D. ensuring that verified company data sources are used for response generation
Grounding a generative AI model means anchoring its responses to verified, authoritative data sources within an organization. This directly ensures that the model generates responses based on trusted company data rather than relying solely on its training data, which may contain outdated or inaccurate information. Option A relates to multilingual capabilities, not grounding. Option B concerns analytics metrics unrelated to response accuracy. Option C describes natural language interaction, a general feature rather than grounding specifically.
Why the other options are wrong
- A. Supporting multiple languages is a localization feature independent of grounding data sources.
- B. User interaction metrics measure engagement, not the accuracy or reliability of response sources.
- C. Natural language query capability is a general feature of generative AI, separate from the grounding requirement.
Question 3
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?
Show answer and explanation
Correct answer: A. Prepare the training data.
Data preparation is the foundational first step in building any machine learning model. Even when data is described as consistent and uniform, it must be validated, cleaned, formatted, and split into appropriate training and testing sets before any modeling work begins. Training (C) cannot occur without prepared data. Evaluation (B) requires a trained model first. Hyperparameter tuning (D) comes after initial training. Deployment (E) occurs only after successful model validation.
Why the other options are wrong
- B. Evaluation cannot be performed until the model has been trained on prepared data.
- C. Training cannot begin without first preparing the data into appropriate formats.
- D. Hyperparameter tuning occurs during or after training, not before data preparation.
- E. Deployment is the final step after the model is trained, evaluated, and validated.
Question 4
Your company uses a non-reasoning generative AI model to create textual content.
You discover that the model’s responses are inconsistent and do NOT meet expectations.
You need to improve the prompts.
What should you do? More than one answer choice may achieve the goal.
Select the BEST answer.
Show answer and explanation
Correct answer: B. Add the context, sources, and expectations to the prompts.
The most effective approach to improving inconsistent generative AI responses is to add context, sources, and clear expectations to prompts. This comprehensive approach provides the model with necessary background information and explicit guidance on desired output quality and format. Option A, while helpful, addresses only output examples. Option C can actually harm comprehension by introducing unnecessary jargon. Option D is counterproductive, single concise requirements lack sufficient guidance and context for consistent, high-quality responses.
Why the other options are wrong
- A. While examples help, they alone are insufficient without broader context and expectations guidance.
- C. Technical jargon may confuse the model; clear, direct language works better for prompt optimization.
- D. Single concise requirements provide too little guidance and context for achieving consistent improvements.
Question 5
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:
Processes both images and text -Summarizes the design of a building Answers user questions about a building’s design Extracts information from blueprints, such as the location of electrical, heating, and plumbing systems
What should you recommend?
Show answer and explanation
Correct answer: A. a multi-modal solution
The requirements explicitly demand processing both images and text from PDF blueprints, summarizing designs, answering design questions, and extracting specific information about systems. This combination of capabilities, handling multiple content types and performing diverse analytical tasks, defines a multi-modal solution. Option B, OCR, only addresses text extraction from images. Option C, text completion, doesn't fit the analytical and question-answering requirements. Option D, document summarization, is too narrow for the comprehensive needs including image processing and detailed information extraction.
Why the other options are wrong
- B. OCR only converts images to text; it doesn't address summarization, questio-nswering, or multi-modal processing.
- C. Text completion generates text sequentially; it cannot summarize, extract specific information, or answer questions about blueprints.
- D. Document summarization alone cannot process images, answer user questions, or extract specific system locations from blueprints.
Question 6
In which scenario is Azure Machine Learning most likely to deliver strategic value for an organization?
Show answer and explanation
Correct answer: A. Using historical sales data to forecast demand across product categories.
Using historical sales data to forecast demand across product categories represents a classic machine learning use case that delivers measurable strategic business value through predictive analytics and decision support. This enables data-driven inventory, pricing, and resource allocation decisions. Option B describes process digitization, which is important but doesn't require machine learning specifically. Option C is manual data entry without analytical depth. Option D is basic personalization that doesn't require sophisticated machine learning models.
Why the other options are wrong
- B. Digitizing paper processes improves efficiency but doesn't necessarily require machine learning capabilities.
- C. Manual spreadsheet entry of feedback represents data collection, not machine learning analysis or strategic insight generation.
- D. Rule-based email personalization by location can be achieved with simple logic, not advanced machine learning.
Question 7
Your company uses a generative AI solution.
You need to improve the quality of responses by using grounding.
Which statement accurately describes how grounding improves accuracy and relevancy?
Show answer and explanation
Correct answer: C. anchors the responses in specific data sources
Grounding improves generative AI accuracy and relevancy by anchoring responses in specific, verified data sources rather than relying entirely on model-generated content based on training data. This ensures responses reference authoritative organizational data, reducing hallucinations and increasing factual accuracy. Option A describes diverse perspectives, which is a general principle but not the mechanism of grounding. Option B concerns model interpretability, a separate concern. Option D focuses on model limitations rather than accuracy improvement mechanisms.
Why the other options are wrong
- A. Referencing diverse perspectives relates to balanced viewpoints, not the grounding mechanism for accuracy improvement.
- B. Explaining model generation processes addresses interpretability, not the accuracy improvements from grounding.
- D. Specifying model strengths and weaknesses identifies limitations but doesn't explain how grounding improves accuracy.
Question 8
You plan to meet with a group of stakeholders to discuss how generative AI can benefit your company.
You need to provide the stakeholders with a relevant description of generative AI during the meeting.
Which description should you use?
Show answer and explanation
Correct answer: A. Generative AI is designed to generate responses based on a user’s natural language prompts.
Generative AI is fundamentally defined as technology designed to generate new responses, content, and outputs based on user prompts in natural language. This is the core defining characteristic that distinguishes it from other AI types and is the most relevant description for stakeholder awareness. Option B describes translation, a specific application but not generative AI's core purpose. Option C describes predictive analytics, a different AI category. Option D describes recommendation systems, another distinct AI application type.
Why the other options are wrong
- B. Translation is a specific application; it doesn't define the broader purpose of generative AI.
- C. Predictive analytics based on historical data is traditional machine learning, not generative AI.
- D. Recommendation systems use collaborative filtering or similar techniques, not generative AI specifically.
Question 9
Which statement accurately describes the difference between a pretrained generative AI model and a fine-tuned generative AI model?
Show answer and explanation
Correct answer: C. A pretrained model is trained on broad datasets, while a fine-tuned model is adapted to perform well on a narrower, domain-specific dataset.
A pretrained model is trained on large, broad datasets to learn general patterns and features across diverse data. A fine-tuned model takes this pretrained foundation and adapts it through additional training on a smaller, domain-specific dataset to optimize performance for particular tasks or industries. This approach leverages the general knowledge already learned while specializing the model for narrower use cases.
Why the other options are wrong
- A. Both pretrained and fine-tuned models can work with labeled data; the distinction is not about labeling requirements but dataset scope and specialization.
- B. Pretrained models typically have more parameters and take longer to train initially; fine-tuning is faster because it builds on existing learned representations.
- D. This reverses the actual relationship; pretrained models are designed for genera-urpose use, while fine-tuned models are optimized for specific tasks.
Question 10
Your company receives thousands of scanned invoices each month.
You need to recommend an AI solution that can automatically extract key details, such as invoice numbers, vendor names, and total amounts.
What is the best solution to recommend? More than one answer choice may achieve the goal.
Select the BEST answer.
Show answer and explanation
Correct answer: A. Azure Document Intelligence in Foundry Tools
Azure Document Intelligence is specifically designed for extracting structured data from documents such as invoices, receipts, forms, and contracts. It uses optical character recognition and machine learning to automatically identify and extract key fields like invoice numbers, vendor names, and amounts with high accuracy. This is its primary purpose, making it the best choice for automated invoice processing at scale.
Why the other options are wrong
- B. Azure Vision is optimized for image analysis and object detection rather than structured data extraction from documents.
- C. Azure AI Search is designed for indexing and searching large volumes of data, not for extracting specific fields from documents.
- D. Azure Machine Learning is a general-purpose platform requiring custom model development, whereas Document Intelligence provides pre-built, domain-specific capabilities for document processing.
That was 10 of 100.
The full Microsoft AB-731 pack has all 100 questions, each with the answer, the explanation and why the other options are wrong, plus a questions-only copy for timed runs. US$39, paid once, with free monthly updates and a pass-or-your-money-back guarantee.