AB-731 Microsoft AI Transformation Leader Free Practice Exam Questions (2026 Updated)

The problem is harmful output language (inappropriate or exclusionary/ableist content). The requirement says you must prevent those responses while minimizing costs . The most cost-effective and direct control is to add a content-moderation filter (B) to screen and block (or rewrite/escalate) responses that violate your safety or inclusion standards. Moderation can be applied at the output stage (and often also at input) without retraining the model, which keeps costs and delivery time low. It also provides an immediate safety layer even if the underlying model occasionally produces biased or exclusionary phrasing.

Option A is not reliable: a newer model version might reduce issues but does not guarantee elimination of ableist language, and you still need policy enforcement. Option C (retraining on only inclusive content) can help, but it is typically expensive (data curation, re-training, re-evaluation, regression testing, re-deployment) and not the “minimize costs” path—also it can reduce coverage/utility if overly restrictive. Option D is clearly wrong because it would amplify the harmful behavior.

In practice, the lowest-cost, high-impact approach is to implement moderation thresholds and handling actions (block, warn, regenerate with constraints, human review) and then, if needed, follow up later with deeper mitigations like prompt constraints, targeted fine-tuning, red-teaming, and continuous evaluation.

A pretrained generative AI model is trained initially on a large, broad, and diverse dataset so it learns general language (or multimodal) patterns and capabilities. Fine-tuning then takes that pretrained base and performs additional training on a smaller, task- or domain-specific dataset to specialize behavior—improving performance for a particular use case, tone, style, or domain knowledge representation. That is exactly what option C states, making it the correct answer.

Option A is incorrect because both pretraining and fine-tuning may use labeled or unlabeled data depending on the technique; the distinction is not “labeled vs. unlabeled.” Option B is incorrect because a pretrained model is not “faster to train” due to fewer parameters; pretraining is typically the most compute-intensive phase precisely because it’s done at large scale, while fine-tuning is smaller but still trains the same model architecture. Option D is reversed: the pretrained model is the general-purpose foundation, while the fine-tuned model is the specialized variant for a specific task or dataset.

The correct answer is data security. The principle of least privilege means users, apps, and services receive only the minimum access required to perform their work. Microsoft 365 Copilot respects existing Microsoft 365 permissions and does not grant users new access to content they are not already authorized to view. That behavior is a security control because it helps prevent unauthorized access to company information. Data classification is about labeling or categorizing information by sensitivity. Data compliance focuses on meeting policy, legal, or regulatory requirements. Data loss prevention helps prevent sensitive information from being shared inappropriately. Least privilege is primarily an access-control practice, so the best category is data security.

The correct answer is D. predicting customer retention. Azure Machine Learning is best suited for predictive analytics scenarios where historical data is used to train a model that predicts future outcomes. Customer retention prediction typically uses customer behavior, purchase history, engagement data, service interactions, churn signals, and account attributes to estimate whether a customer is likely to stay or leave. That is a classic machine learning use case. Generating marketing campaigns and summarizing customer service transcripts are generative AI or natural language processing tasks. Creating product descriptions from images is a multimodal or generative AI scenario. The wording “predicting” is the key indicator: when the business requirement is to forecast outcomes from patterns in data, Azure Machine Learning is the correct fit.

To align an AI customer support agent with Microsoft’s Responsible AI principles, two high-impact actions are fairness/inclusiveness validation and transparency to users . B is correct because testing for inclusive and culturally sensitive responses directly supports fairness and helps reduce harm. In practice, you evaluate responses across diverse user personas, languages/dialects, accessibility scenarios, and sensitive contexts. You look for biased assumptions, stereotyping, exclusionary language, and disparate quality of service. This also implies ongoing monitoring because model behavior can drift as prompts, knowledge sources, and user inputs evolve.

E is correct because a clear disclaimer supports transparency: customers should know they are interacting with an AI system, understand the type of assistance it can provide, and know what to do if the response is incorrect or they need a human. A disclosure is also a practical risk-control that reduces overreliance and sets expectations about limitations.

The other options are not best for Responsible AI alignment: A (retain all conversations) can conflict with privacy/data minimization; retention must be justified and governed, not automatic. C (operate independently) undermines accountability and human oversight. D (multiple purposes) increases scope and risk rather than improving responsible use.

Answer Area

The marketing department at your company wants AI to summarize emails and create presentations. Answer: Use Microsoft 365 Copilot

The HR department at your company wants a conversational agent for policy questions and leave requests. Answer: Build with Microsoft Copilot Studio

The manufacturing department at your company wants AI to predict maintenance schedules. Answer: Build with Azure Machine Learning

The finance department at your company wants AI-powered access to enterprise resource planning ERP data by using familiar productivity tools. Answer: Extend with Microsoft 365 Copilot connectors

These scenarios map to four distinct solution patterns: out-of-the-box productivity assistance, low-code conversational agents, predictive ML, and enterprise data integration.

Marketing’s need to summarize emails and create presentations is a core “productivity copilot” use case. Microsoft 365 Copilot is embedded in Outlook, Word, PowerPoint, and Teams, so it directly supports summarization, drafting, and presentation generation without building a custom solution—making Use Microsoft 365 Copilot the best fit.

HR’s requirement is a conversational agent tailored to internal policies and workflows such as leave requests. That typically needs custom dialog, grounded knowledge sources, and possibly actions/workflows. Microsoft Copilot Studio is designed to build and manage such agents with organizational knowledge and business process integration, so Build with Microsoft Copilot Studio fits best.

Manufacturing’s predictive maintenance scheduling is classic predictive analytics: learning patterns from historical telemetry/maintenance data to forecast failures or optimal service windows. This is best addressed with Azure Machine Learning , which supports training, evaluating, and deploying custom predictive models.

Finance wants AI-powered access to ERP data “using familiar productivity tools,” which implies bringing external line-of-business data into the Microsoft 365 Copilot experience. That is precisely where Microsoft 365 Copilot connectors help—indexing and exposing enterprise data sources so Copilot can reference them in a governed way—so Extend with Microsoft 365 Copilot connectors is the best approach.

The correct answer is B. Testing the AI solution to identify and mitigate unfair or inconsistent outcomes directly supports Microsoft responsible AI principles, especially fairness, reliability and safety, accountability, and inclusiveness. Before deployment, organizations should evaluate whether the system behaves consistently across different user groups, avoids discriminatory outcomes, and produces safe and reliable outputs. Automatically approving or rejecting loan applications increases risk unless strong governance and human oversight are in place. Collecting personal data can increase privacy exposure if it is not necessary and controlled. Prioritizing model performance alone is also insufficient because a technically accurate model can still produce biased, unsafe, or non-transparent outcomes. Responsible AI requires evaluation, mitigation, oversight, and monitoring.

Answer Area

Content filtering controls can prevent AI-generated responses from exposing confidential and sensitive information. Answer: Yes

AI-generated content can unintentionally reveal sensitive information if the generative AI model has access to unsecured data sources. Answer: Yes

To prevent data exposure, only the prompts used by users must be protected by using policies. Answer: No

Yes — Content filtering (and related safety controls) can help reduce the chance that responses contain policy-violating or sensitive outputs by detecting and blocking certain categories of content. While filtering is not a perfect guarantee, it is a recognized control to prevent or reduce exposure risk in outputs (for example, blocking regulated data patterns, disallowed content categories, or unsafe disclosures).

Yes — If a model (or the solution’s retrieval layer) can access poorly governed repositories—such as broadly shared folders, misconfigured SharePoint sites, or unsecured databases—then the system can surface sensitive information in responses even without malicious intent. This is why access control, data classification, and permission hygiene are critical prerequisites for deploying AI assistants grounded in organizational content.

No — Protecting only user prompts is insufficient. Data exposure can occur through multiple paths: retrieved documents, generated outputs, logs/telemetry, training/fine-tuning data, and connector/index configuration. Preventing exposure requires layered controls: data governance (labels, DLP, least privilege), secure connectors, output filtering, auditing, and user training—not prompt policy alone.

Provisioned (PTUs)

For continuous, large-scale workloads, the key needs are predictable throughput, consistent latency, and cost/performance stability . Azure OpenAI Provisioned Throughput Units (PTUs) are designed for this scenario because you reserve model capacity to meet sustained demand. This reduces the risk of variability that can occur with purely on-demand usage during peak periods and provides a more predictable operating model when the workload is always “on” and high volume.

Standard (On-Demand) is best when usage is variable or you’re starting small (PoCs, pilots, spiky workloads), because it is flexible pay-as-you-go but can be less predictable at sustained scale. Batch API is optimized for asynchronous, non-interactive processing where you can tolerate delayed results (for example, large offline summarization jobs), not for always-on, real-time interactions at scale.

Therefore, for continuous high-volume production workloads, Provisioned (PTUs) is the best pricing model choice.

uses reasoning capabilities to generate deep insights based on organizational data and the web.

The sentence is best completed by the option describing Researcher as a research-oriented reasoning agent that combines information from the web and your work data to produce deeper insights. Microsoft describes Researcher as an agent built into Microsoft 365 Copilot to tackle complex, multistep research and to help users gather, analyze, and summarize information from “the web, your work documents, or both,” producing a structured output that supports decision-making. That is exactly what the completion “uses reasoning capabilities to generate deep insights based on organizational data and the web” captures.

The other dropdown options are better matches for different tools/agents: “creates visual dashboards from structured data in Excel and Power BI” is more aligned to BI/reporting workflows; “generates a pivot table and performs time series forecasting” is spreadsheet/analytics functionality; and “performs complex, multi-step, data analysis and code execution tasks over arbitrary datasets” is the hallmark positioning of the Analyst agent (“virtual data scientist”) rather than Researcher. Researcher’s differentiator is deep research across both organizational context and the open web, while Analyst’s differentiator is data analysis and computation .

Most generative AI language model pricing is based on token consumption , which measures the amount of text processed by the model. Tokens are sub-word units used internally by language models (for example, parts of words, whole words, or punctuation). When you send a prompt, the model consumes input tokens (your prompt + any system instructions + retrieved grounding context). When it generates a response, it consumes output tokens (the generated completion). Costs typically scale with the total input + output tokens processed, which is why long prompts, large grounding passages, and lengthy responses increase spend. This also explains why prompt optimization, response length limits, caching, and careful grounding are common cost-control techniques in enterprise solutions.

By contrast, “documents” is too coarse (a document can be 1 page or 500 pages). “Requests” is not the primary unit for most LLM pricing models because request sizes vary dramatically. “Words” is not used because the model’s actual compute unit is tokens, and tokenization differs across languages and text patterns. Therefore, the most accurate completion is tokens .

The requirement describes a predictive analytics / machine learning scenario: using historical data to learn patterns and then predict future outcomes . The Microsoft service that directly supports the end-to-end machine learning lifecycle—data preparation, model training, evaluation, deployment, and MLOps—is Azure Machine Learning , which is why C is the best choice. Azure Machine Learning is explicitly designed to help data scientists and engineers train and deploy models and manage the ML project lifecycle, making it the right fit for building a predictive model from your dataset.

The other options focus on different problem classes: Azure Document Intelligence is for extracting structured data from documents (OCR, key-value pairs, tables), not for general predictive modeling. Azure Content Understanding is for deriving structured insights from multimodal content (documents, images, audio, video) into a user-defined schema; it’s not the primary service for training predictive models from a tabular historical dataset. Microsoft Foundry is a broader platform for building AI apps/agents and orchestrating models/tools, but the specific need here is classical ML training and prediction—handled most directly by Azure Machine Learning.

A common reason models degrade after being successful in production is data drift (also called concept drift). Over time, the distribution of input data changes —for example, customer behavior shifts, product catalog changes, seasonality changes, new categories appear, sensors get recalibrated, or business processes evolve. When the model sees data that differs from what it was trained on, its predictions can become less accurate. This is exactly what option A describes and is the most likely “six months later” cause.

Option B is not a primary explanation for reduced predictive accuracy. More compute can improve throughput/latency, but it does not inherently improve correctness of predictions. If anything, compute constraints typically cause timeouts or slower responses, not a systematic accuracy drop.

Option C (trained incorrectly) would usually manifest earlier—poor performance from the start—unless the “incorrectness” is that the model was trained on a snapshot that later became stale (which again maps back to drift). The correct operational response is to monitor for drift, validate performance regularly, and retrain/refresh the model using newer representative data and updated features/labels.

Answer Area

Retrieval Augmented Generation RAG requires model fine-tuning. Answer: No

Retrieval Augmented Generation RAG is helpful when you need a generative AI solution that can access current, verifiable information. Answer: Yes

Retrieval Augmented Generation RAG enables you to get more relevant responses based on your organization’s documents without retraining the base model. Answer: Yes

RAG is an architecture pattern that improves generative AI responses by retrieving relevant information from external knowledge sources (for example, a document index, database, or knowledge base) and injecting that information into the model’s prompt/context at runtime.

No — RAG does not inherently require fine-tuning. Fine-tuning changes the model weights. RAG, instead, keeps the base model as-is and augments it with retrieved context. Fine-tuning can be complementary (for style, domain tone, or specialized tasks), but it is not required for RAG to work.

Yes — RAG is especially valuable when you need current and verifiable information because the retrieval layer can pull the latest approved content (updated policies, product specs, incident runbooks) and provide it to the model. This reduces hallucinations and makes answers traceable to known sources.

Yes — A major benefit of RAG is improved relevance to organizational documents without retraining. Instead of rebuilding the model whenever documents change, you update the underlying content store/index; the model then generates responses grounded in the retrieved passages, producing answers that align with your organization’s latest information and terminology.

uses structured data and provides insights by using text, charts, tables, and other visuals.

The Analyst agent in Microsoft 365 Copilot is positioned as a “data analysis” reasoning agent that helps users work through structured information (for example, tables, spreadsheets, and other dataset-like inputs) and then produces analytical outputs . The best completion is the option stating it “uses structured data and provides insights by using text, charts, tables, and other visuals,” because that describes the hallmark outcome of analyst-style work: summarizing patterns, highlighting key drivers, and presenting results in formats that business users can act on. Analyst-style assistance typically includes exploring the data, identifying trends and anomalies, comparing segments, and explaining findings clearly—often accompanied by tables and visual representations that make the insights easier to consume.

The other dropdown options align to different use cases: “compiles background research for a new market or initiative” describes a research-oriented agent, “generate audio summaries” is a media summarization function, and “answer employee FAQs” describes a conversational knowledge assistant. Analyst is the one most directly associated with structured-data interpretation and producing a mix of narrative plus analytical artifacts (tables/charts) to communicate conclusions.

Answer Area

Retrieval Augmented Generation (RAG) architectures require that a large language model (LLM) be retrained on proprietary data. Answer: No

Retrieval-Augmented Generation (RAG) grounds a language model to produce more factual and context-relevant responses. Answer: Yes

Retrieval Augmented Generation (RAG) retrieves information from external knowledge sources at runtime instead of relying solely on the knowledge of a generative AI model. Answer: Yes

1) No — RAG does not require retraining or fine-tuning the base LLM on proprietary data. The defining idea of RAG is to keep the model as-is and instead supply it with relevant context retrieved from trusted sources at inference time. Fine-tuning can be optional for style or specialized behavior, but it is not a requirement for RAG.

2) Yes — RAG is a grounding approach. By retrieving authoritative passages (policies, manuals, product specs, internal knowledge bases) and injecting them into the prompt context, the model’s answer is constrained by evidence that is relevant to the user’s question. This improves factuality and domain relevance and helps reduce hallucinations.

3) Yes — RAG explicitly depends on runtime retrieval from external knowledge sources, such as indexed documents, databases, or enterprise repositories. The retrieval layer finds the best matching content for the query, and the generation layer uses that retrieved content to craft the response. This is why RAG is valuable when information changes frequently: you update the source documents/index rather than retraining the LLM.

Overall, RAG is best understood as an architecture pattern that combines search/retrieval + generation , improving accuracy and freshness without the cost and risk of retraining the underlying model each time the knowledge base changes.

The requirement is to analyze text and identify “people, locations, and companies” mentioned in news articles. This is a classic Named Entity Recognition (NER) / entity extraction scenario, which falls under natural language processing. Azure Language in Foundry Tools is the correct choice because it provides text analytics capabilities that detect and categorize entities in unstructured text—commonly including Person , Location , and Organization . This enables downstream experiences such as topic tagging, search filters (e.g., “all articles mentioning Company X”), trend dashboards (top people/places mentioned this week), and improved content discovery.

The other options do not match the requirement. Content Safety focuses on moderating harmful or policy-violating content (for example, hate, violence, self-harm, sexual content) and is not the primary tool for extracting named entities. Azure Vision is for analyzing images and performing OCR; it would only be relevant if the articles were images or scans, but the task here is entity extraction from text articles. Azure Speech is for speech-to-text, text-to-speech, and audio analysis; it would be used if your input were audio recordings rather than written articles.

Therefore, to identify key entities (people, locations, companies) from daily news article text, the best recommendation is Azure Language in Foundry Tools .

improves the accuracy and reliability of the predictions and outputs of AI.

High-quality grounding data improves a generative AI solution by anchoring responses to trusted, relevant, and up-to-date information , which increases the likelihood that outputs are accurate, consistent, and aligned with the organization’s expectations. This is why the best completion is “ improves the accuracy and reliability of the predictions and outputs of AI .” When the model is given authoritative context (for example, approved policy text, product specifications, knowledge base articles, or controlled enterprise content), it has less need to “guess” based on general patterns in its training data. That reduces hallucinations and improves response relevance to the user’s question and the business domain.

It does not “ensure that all responses are factually accurate” because grounding reduces errors but cannot eliminate them completely—retrieval can return incomplete or irrelevant passages, user prompts can be ambiguous, and the model can still misinterpret context. It also does not inherently “increase performance of an AI model” in the sense of speed/throughput or model capability; grounding is an architecture and data strategy that improves output quality, not compute efficiency. Finally, grounding is not about “increasing storage required to host an AI model.” While you may store documents in an index or repository, the core benefit is improved response quality through better context, not larger model hosting requirements.

 Extract information from invoices and credit notes: Azure Document Intelligence in Foundry Tools

 Identify discussion topics in customer service call recordings: Azure AI Search

 Detect customer sentiment and opinions: Azure AI Search

To minimize development time, you should select services that provide prebuilt extraction and enrichment rather than building custom models.

For invoices and credit notes , the best fit is Azure Document Intelligence because it is purpose-built for intelligent document processing and includes prebuilt models (such as invoice-style extraction) that return structured fields (invoice number, vendor, totals, dates, line items). This directly satisfies “extract information” from financial documents without custom ML training.

For customer service call recordings , the raw input is audio, but the task is to identify “discussion topics,” which is an information discovery and text-enrichment problem once the calls are transcribed. Azure AI Search is the correct recommendation because it can index large collections of content and apply built-in AI enrichment (for example, key phrase extraction and entity extraction) to help derive and query “topics” across many records.

For sentiment and opinions , Azure AI Search is again the best choice in this constrained list because it can run built-in enrichment skills over ingested text (such as transcripts or notes) to annotate content with sentiment-related metadata and then make that enriched data searchable and filterable at scale. This approach delivers fast time-to-value: ingest → enrich → index → query, without standing up a custom training pipeline.