View Full Microsoft AB-731 Exam Dumps and Practice Test Dumps.

 

Question 121

A company wants an AI solution to classify incoming customer messages into categories such as billing, technical support, and account access. Which capability is most directly relevant?

1. Image generation
2. Text classification
3. Document storage
4. Network monitoring

Correct Answer: 2

Explanation

Text classification can help an organization categorize incoming messages according to their content. In a customer-support environment, an AI system could identify whether a message concerns billing, technical support, account access, or another predefined category. This can help route requests to the appropriate team and reduce repetitive manual sorting. The organization should evaluate classification accuracy using representative customer messages and monitor performance after deployment. Human review may also be appropriate for ambiguous cases. The chosen AI capability should match the actual business requirement rather than using generative AI unnecessarily.

Question 122

Which situation best illustrates scalability as a business benefit of generative AI?

1. An employee manually writes one document per week
2. A team uses AI to generate initial drafts for thousands of similar documents
3. A company removes all document review
4. A department stops measuring productivity

Correct Answer: 2

Explanation

Scalability means that a solution can support increasing workloads without requiring a proportional increase in manual effort. Generative AI can provide this benefit when organizations need to create large volumes of similar content, such as product descriptions, summaries, or personalized communications. Employees can review and refine generated outputs while the AI handles repetitive drafting activities. The organization should still establish quality standards and appropriate review processes. Scalability is valuable when AI allows a business process to handle significantly greater volume while maintaining acceptable quality, cost, and operational control.

Question 123

A company is selecting between two models. Model A is more capable but significantly more expensive, while Model B meets the quality requirement at lower cost. Which principle should guide the decision?

1. Always select the most expensive model
2. Select the model with the longest name
3. Match model capability and cost to the actual business requirement
4. Select a model randomly

Correct Answer: 3

Explanation

Model selection should be based on the requirements of the workload rather than simply choosing the most powerful available option. If a lower-cost model meets the required quality, reliability, latency, and other business criteria, it may be more appropriate for that scenario. A more capable model can be justified when the additional performance provides meaningful business value. Organizations should compare candidate models using representative tasks and consider cost, scalability, security, context requirements, and expected usage. The objective is to achieve the required business outcome efficiently.

Question 124

An AI application generates a response based on several retrieved documents. What should the organization verify before trusting the response for an important business decision?

1. That the retrieved sources are authoritative and relevant
2. That the response contains many words
3. That the application uses a modern interface
4. That the model produces the answer quickly

Correct Answer: 1

Explanation

Grounded responses depend partly on the quality of the information supplied to the model. Organizations should verify that retrieved documents are authoritative, relevant, current, and appropriate for the user’s request. If outdated or incorrect information is retrieved, the generated response may also be unreliable. Important decisions may require additional human verification even when the response is grounded. Organizations should therefore maintain trusted knowledge sources, monitor retrieval quality, apply access controls, and establish review procedures appropriate to the risk of the business process.

Question 125

Which data problem can reduce the reliability of an AI solution if it is not addressed?

1. Duplicate or inconsistent records
2. A simple file name
3. A standard document format
4. A clearly defined business objective

Correct Answer: 1

Explanation

Duplicate, inconsistent, incomplete, or inaccurate data can reduce the reliability of AI systems. When conflicting information appears across records, the system may produce inconsistent results or learn misleading patterns. Organizations should therefore assess data quality before using information for AI applications. Data preparation can include identifying duplicates, resolving inconsistencies, validating important fields, and removing obsolete information where appropriate. Good data governance also helps establish ownership and accountability for important data sources. Improving data quality can contribute to more reliable AI outcomes and more trustworthy business decisions.

Question 126

A business wants to reduce the risk that users will accept an AI-generated answer without questioning it. Which practice can help?

1. Encouraging appropriate human review and communicating system limitations
2. Hiding all information about AI limitations
3. Removing verification procedures
4. Allowing AI outputs to override every business rule

Correct Answer: 1

Explanation

Human review can be important when AI-generated information could influence significant business decisions. Users should understand that generative AI can produce inaccurate or unsupported information and should know when verification is required. Organizations can establish review procedures based on the potential impact of errors. Clear communication about system capabilities and limitations can also help users develop appropriate expectations. Human oversight does not require manually checking every low-risk response, but higher-risk scenarios should have stronger controls. This approach helps balance AI productivity with responsible decision-making.

Question 127

Which stage of the machine learning lifecycle involves checking whether a trained model performs adequately before production use?

1. Data deletion
2. Evaluation
3. Retirement
4. Procurement

Correct Answer: 2

Explanation

Evaluation occurs after training and helps determine whether a model performs adequately for its intended purpose. Organizations can use representative validation or test data and appropriate performance measures to identify weaknesses before deployment. Evaluation may examine accuracy, reliability, fairness, or other characteristics relevant to the use case. Testing should reflect realistic conditions rather than only ideal examples. If the model does not meet the required criteria, the organization may need to improve the data, training process, model selection, or business design before moving toward production deployment.

Question 128

An organization wants to prevent an AI application from accessing data that is unnecessary for its task. Which security principle is most relevant?

1. Maximum access
2. Least privilege
3. Unlimited sharing
4. Public accessibility

Correct Answer: 2

Explanation

Least privilege means granting users and applications only the access required to perform their intended functions. Applying this principle to AI solutions can reduce the amount of information exposed if an application is misconfigured, compromised, or used incorrectly. Organizations should review permissions for data sources, applications, service identities, and users. Least privilege should be combined with authentication, authorization, monitoring, and other security controls. Restricting unnecessary access is particularly important for AI systems that can retrieve information from multiple organizational sources.

Question 129

A company wants to help employees understand practical ways to use AI in their daily work. Which adoption activity is most useful?

1. Practical training using relevant business scenarios
2. Providing only technical documentation
3. Preventing employees from experimenting in approved environments
4. Measuring adoption only through license purchases

Correct Answer: 1

Explanation

Practical training helps employees understand how AI capabilities can be applied to tasks they actually perform. Scenario-based training can demonstrate appropriate use, limitations, verification requirements, security considerations, and effective prompting. Employees are more likely to adopt technology when they understand how it connects to their responsibilities and business objectives. Training should be supported by clear policies and ongoing feedback. Organizations can also use AI champions to share successful practices across teams. Adoption measurement should consider actual usage and business outcomes rather than licensing alone.

Question 130

What is one purpose of an AI council in an organization with multiple AI initiatives?

1. Coordinating strategic priorities and governance across departments
2. Writing every prompt for employees
3. Replacing all technical teams
4. Eliminating business ownership

Correct Answer: 1

Explanation

An AI council can provide cross-functional coordination for an organization’s AI initiatives. It can bring together business, technology, security, privacy, legal, compliance, and other stakeholders to establish priorities and governance expectations. The council can review proposed use cases, identify risks, and help align AI investments with organizational objectives. It does not replace specialized teams or individual business owners. Instead, it creates a mechanism for consistent strategic oversight across departments. This becomes particularly useful when multiple AI projects have overlapping data, security, or governance requirements.

Question 131

A company wants an AI assistant to answer questions using information stored in a controlled corporate knowledge repository. What should be established first?

1. A trusted and appropriately governed knowledge source
2. Unlimited access for every user
3. A policy to ignore source information
4. A requirement to use only random documents

Correct Answer: 1

Explanation

A controlled and trusted knowledge source provides a stronger foundation for AI applications that need organization-specific information. The organization should establish ownership, data-quality standards, update procedures, access controls, and appropriate governance for the repository. When an AI system retrieves information from the source, permissions should ensure that users receive only content they are authorized to access. Establishing the source alone does not guarantee accurate answers, but it improves the quality of information available to the AI system and creates a foundation for responsible retrieval and grounding.

Question 132

Which feature distinguishes a fine-tuned model from a standard pretrained model?

1. It has been further trained using task- or domain-specific examples
2. It cannot process any new input
3. It does not require evaluation
4. It automatically has access to every company database

Correct Answer: 1

Explanation

A fine-tuned model is further trained using data selected for a particular task, behavior, or domain. This can help adapt a pretrained model to specialized requirements when general prompting or other approaches do not provide the desired behavior. Fine-tuning requires suitable training data and careful evaluation because poor-quality or unrepresentative examples can produce undesirable results. It also does not automatically provide access to organizational data. Data access, grounding, authentication, and authorization are separate concerns that must be addressed through the overall application architecture.

Question 133

A business wants to understand whether an AI project is financially worthwhile. Which calculation concept is most relevant?

1. Return on investment
2. Number of prompts alone
3. Number of available models
4. Screen resolution

Correct Answer: 1

Explanation

Return on investment helps organizations compare the value generated by an initiative with the costs required to implement and operate it. For an AI project, costs may include licensing, model usage, integration, data preparation, security, training, maintenance, and change management. Benefits might include reduced processing time, increased productivity, lower operating costs, improved customer experiences, or additional revenue. Organizations should establish measurable outcomes and baseline performance before deployment where possible. ROI analysis becomes more meaningful when the expected benefits and total costs are evaluated over an appropriate period.

Question 134

Which scenario is most likely to require stronger human oversight?

1. AI generating creative brainstorming ideas
2. AI supporting a high-impact decision involving sensitive customer information
3. AI suggesting informal meeting titles
4. AI producing alternative document headings

Correct Answer: 2

Explanation

AI systems used in higher-impact scenarios generally require stronger oversight because errors can have more serious consequences. A system supporting decisions involving sensitive customer information may need additional validation, access controls, monitoring, and human review. Lower-risk activities such as brainstorming or generating headings may require less intensive oversight. The appropriate control level should be determined by factors such as potential harm, sensitivity of the data, regulatory requirements, and business impact. Responsible AI adoption therefore involves matching governance and review requirements to the risk of each specific use case.

Question 135

An organization wants to create a custom agent but first needs to determine whether an existing Copilot capability already meets the requirement. What should the organization do?

1. Compare the business requirement with available standard Copilot capabilities
2. Immediately build a completely new application
3. Disable existing Copilot services
4. Purchase every available AI service

Correct Answer: 1

Explanation

Organizations should first determine whether existing capabilities already satisfy the business requirement before creating a custom solution. Standard Copilot functionality may provide the required experience without additional development, integration, or maintenance. If the requirement cannot be met adequately, the organization can then evaluate extensibility or custom-agent options. This build-versus-extend decision should consider business value, customization requirements, security, cost, time to deployment, and ongoing maintenance. Starting with the simplest suitable option can reduce unnecessary complexity while still allowing customization when there is a clear business need.

Question 136

Which Microsoft capability is most relevant when an AI application needs to search and retrieve information from indexed organizational content?

1. Azure AI Search
2. Azure Vision
3. Microsoft Graph alone
4. Microsoft Copilot Studio alone

Correct Answer: 1

Explanation

Azure AI Search provides capabilities for indexing and retrieving information, making it relevant to AI solutions that need to locate useful content from organizational data. It can support retrieval-augmented generation by finding relevant information that is then provided to a generative model as context. Organizations should consider indexing strategy, relevance, data quality, security, and access controls when implementing search. Azure AI Search serves a different role from Azure Vision, which focuses on visual information, and Copilot Studio, which focuses on creating and customizing conversational agents.

Question 137

A company is concerned that an AI system may treat one customer group differently because its training data does not adequately represent that group. Which issue should be investigated?

1. Fairness and representativeness
2. Token billing only
3. Network bandwidth only
4. Interface design only

Correct Answer: 1

Explanation

Underrepresentation in training or evaluation data can contribute to unfair or unreliable AI outcomes for certain groups. Organizations should examine whether the data appropriately represents the populations and scenarios relevant to the intended use. Evaluation should include suitable measures across relevant groups where appropriate. If significant differences are identified, the organization may need to improve data quality, adjust the model or process, or introduce additional controls. Fairness should be considered throughout the AI lifecycle because problems can originate in data, model development, deployment, or the way outputs are used.

Question 138

A company expects AI usage to vary significantly from month to month and does not yet know its long-term consumption level. Which pricing approach may provide greater flexibility?

1. Pay-as-you-go
2. Fixed employee salary
3. Hardware depreciation
4. Permanent unlimited usage

Correct Answer: 1

Explanation

Pay-as-you-go pricing can provide flexibility when consumption is uncertain or fluctuates over time. It allows an organization to pay based on actual service usage rather than making a larger long-term commitment before demand is understood. This can be useful during experimentation, pilots, or workloads with variable demand. However, organizations should monitor consumption because increased usage can increase costs. If a workload becomes predictable and substantial, a commitment-based option may warrant evaluation. Pricing decisions should therefore reflect expected workload patterns, financial objectives, and operational requirements.

Question 139

Which action can help an organization identify adoption problems after an AI tool has been deployed?

1. Reviewing usage data, employee feedback, and business outcomes
2. Ignoring users who do not adopt the tool
3. Measuring only the number of licenses purchased
4. Disabling support channels

Correct Answer: 1

Explanation

Adoption measurement should examine more than whether users have been assigned licenses. Usage data can reveal which capabilities are being used, while employee feedback can identify usability issues, training gaps, trust concerns, or workflow problems. Business outcome measures can show whether adoption is producing the intended value. Combining these sources provides a more complete understanding of adoption. Organizations can then adjust training, communication, workflows, or governance based on evidence. Continuous measurement is particularly important when AI capabilities are introduced across different departments with different needs.

Question 140

A company has completed several AI pilots and wants to expand successful solutions responsibly. What should guide the decision to scale?

1. Evidence of business value, acceptable risk, user readiness, and operational feasibility
2. The number of AI announcements made by competitors
3. The age of the AI model
4. The number of employees attending demonstrations

Correct Answer: 1

Explanation

Scaling should be based on evidence that the AI solution provides meaningful business value while remaining manageable from security, privacy, governance, cost, and operational perspectives. Organizations should review pilot results, user readiness, data requirements, performance, risks, and ongoing resource needs before expanding. A successful demonstration alone does not establish production readiness. The organization should also confirm that appropriate training, support, monitoring, and governance processes are available. Using evidence from pilots allows leaders to make informed scaling decisions and address weaknesses before the solution reaches a larger population.