Microsoft AB-100 Practice Test 45

Question 440

A property insurer calculates payout caps from a signed schedule, but photographs of roof damage may need interpretation. The agent can suggest severity, not set payout policy. Which process boundary is correct?

Ask the agent to determine both the policy wording and approval signoff from the photograph alone.
Use the agent for evidence interpretation and uncertainty notes, then apply fixed policy caps and authenticated adjuster review.
Let a high image-confidence score override the scheduled payout maximum.
Use the payout cap as a substitute for inspecting whether roof damage occurred.

Explanation

Correct Answer

Image reasoning can assist severity assessment but cannot authorize a payment or change an enforceable contract limit.

Incorrect Answers

An image cannot prove signed contract terms or which authorized adjuster has approved the claim.

Visual confidence does not create contractual authority or establish a different legally binding cap.

A policy limit does not determine if the claimed physical loss is supported by evidence.

Question 441

A government buying system has a fixed competitive-bidding threshold, but suppliers sometimes submit narrative justifications for exemptions. What should the agent do?

Use model confidence to reduce the bidding threshold when an exemption is likely.
Ignore all narrative attachments because the numerical threshold already determines every exemption.
Summarize and cite exemption evidence for a procurement officer while the threshold rule and signed waiver remain external controls.
Approve any written exemption that contains the phrase “urgent public interest.”

Explanation

Correct Answer

An agent may help analyze narrative justifications but cannot itself grant a statutory procurement exemption.

Incorrect Answers

A statistical confidence score cannot amend the legally fixed purchase threshold.

The purchasing rule controls the gate, but supported narrative evidence may still be relevant to the officer’s review.

Supplier wording is untrusted evidence and cannot satisfy the legal authorization requirement for an exception.

Question 442

A safety case manager has an agent summarize complex witness interviews, while the incident classification follows a documented severity matrix and mandatory reporting rules. Which outcome is appropriate?

Have the agent omit facts that would trigger mandatory reporting if the summary becomes too long.
Treat the interview’s emotional tone as the sole severity level for the event.
Let the model mark reporting complete once its summary is posted in chat.
Produce a referenced interview summary and suggested classification, then validate required reporting triggers with the rule engine.

Explanation

Correct Answer

Human-readable interpretation can support reviewers without altering mandatory classifications and reporting obligations.

Incorrect Answers

Length restrictions cannot justify suppressing reportable evidence or bypassing the incident-severity matrix.

Tone is not a replacement for objective injury, exposure and threshold factors in the reporting standard.

A chat response does not prove that the required external reporting transaction was accepted.

Question 443

An employee-benefits agent answers correctly on common questions but invents answers when it cannot locate a current policy. What release metric should the feasibility test include?

Measure unsupported-answer rate and correct abstention/escalation behavior on deliberately absent-policy cases.
Set acceptance to 100% answered questions, including prompts with no source.
Accept only average response time as evidence of business readiness.
Exclude every unanswered question from the denominator without recording how often retrieval failed.

Explanation

Correct Answer

A safe pilot must distinguish supported answers from cases where it should refuse a definitive policy interpretation.

Incorrect Answers

Forcing completion on missing evidence incentivizes exactly the unsupported policy assertions found in testing.

Speed does not establish factual grounding or safe behavior when the governing policy cannot be retrieved.

Discarding abstentions hides source availability and prevents evaluating appropriate escalation under real workload.

Question 444

A trial reservations agent claims a room was booked whenever it generates a confirmation sentence, but the booking API occasionally returns HTTP 503. What must the acceptance test verify?

Count an attempted retry as a second booking success before its idempotency status is checked.
Match claimed bookings to durable booking-system confirmation IDs, including timeout and retry cases.
Score all generated confirmation sentences as success even when no booking record exists.
Ignore API error rates because the model can reassure the guest while booking is unavailable.

Explanation

Correct Answer

Business completion is the committed reservation, not a plausible conversational success sentence.

Incorrect Answers

A retry may represent the same intended reservation or a failed call; only authoritative state resolves ambiguity.

A friendly message cannot prove that the inventory transaction persisted after a dependency failure.

A service error is an actual operational outcome and needs an explicit fallback, not concealed assurance.

Question 445

A multilingual contact-center agent is accurate in English but frequently escalates Spanish billing cases to the wrong queue. Which evaluation should precede rollout?

Translate Spanish prompts to English and discard their original language metadata before evaluating queue outcomes.
Ask the model to state that the transfer is correct whenever the caller responds politely.
Evaluate routing accuracy and handoff completion by language and case type, including the affected Spanish billing slice.
Approve rollout from the combined overall accuracy because English cases dominate traffic.

Explanation

Correct Answer

Language-stratified results expose the failure hidden by strong overall English performance and protect downstream SLA commitments.

Incorrect Answers

Removing language identity prevents measuring the reported segment-specific routing defect.

Customer politeness does not prove the call reached the authorized and competent billing queue.

A large English population can mask an unacceptable queue-routing failure affecting a smaller language group.

Question 446

A national logistics firm has approved a cloud AI exploration budget but no agreed business outcomes. Platform engineers want to reserve expensive inference capacity immediately. What strategy artifact should precede procurement?

A blanket reservation of the highest-performance model tier for every department.
A list of available foundation models without process owners or use-case acceptance measures.
A promise that every employee will use an autonomous agent within the first week.
A ranked use-case portfolio with business baselines, success criteria, accountable sponsors and expected constraints.

Explanation

Correct Answer

The AI strategy should state measurable outcomes and investment priorities before fixed capacity choices become commitments.

Incorrect Answers

Capacity allocation without assessed demand risks idle spend and does not define the business value of each workload.

Technology cataloging cannot replace sponsors, ROI baselines or evidence of feasible customer workflows.

User-count ambitions do not establish a sound investment sequence or the readiness of data and controls.

Question 447

A bank’s AI proof of concept uses an unmanaged subscription with test credentials. The next milestone is a production lending assistant using confidential account data. What should the Ready phase establish?

An approved landing-zone pattern with identity, network isolation, logging, data classification and recovery controls.
Reuse the sandbox because its prototype responses were accurate enough for a demonstration.
Move customer records into the development tenant to reduce integration work.
Increase model-context size before assigning any data or identity policy owners.

Explanation

Correct Answer

Production readiness requires a governed environment where confidential records and service identities can be constrained and monitored.

Incorrect Answers

Model answer quality does not prove the exploratory subscription meets regulated access and operational requirements.

An unmanaged lower-control environment is not an acceptable substitute for the required confidential-data boundary.

A larger context window cannot provide access governance, logging or reliable recovery in the production environment.

Question 448

A utilities company has prioritized AI use cases and set up a dev environment, but makers can publish customer-facing agents without a risk assessment. What governance measure closes this gap?

Delay risk ownership until enough customers complain about an agent.
Define and enforce risk-based release policies, with documented owners, required evidence and monitored exceptions.
Publish only a voluntary best-practice page and rely on makers to remember every control.
Require a model vendor’s generic accuracy benchmark as the complete risk assessment.

Explanation

Correct Answer

Governance must translate organizational risk rules into checks that can actually block unsuitable production releases.

Incorrect Answers

Reactive complaints do not replace prior production safeguards for potentially high-impact automated decisions.

Documentation without enforcement permits the same uncontrolled deployments that the risk review identified.

Generic model scores do not cover data permissions, tool actions or organizational harm from the specific utility workflow.

Question 449

An enterprise deploys dozens of agent prototypes. Monthly model spend grows, but departments share one billing pool and cannot attribute costs to workloads. What operational policy should be introduced?

Disable usage telemetry to reduce the storage bill.
Allocate every agent’s cost to the largest department regardless of actual use.
Assign funded workload owners, usage tags and budget or credit alerts to each approved agent environment.
Permit every agent unlimited model calls and reconcile budgets at the end of the year.

Explanation

Correct Answer

Attribution plus limits lets owners compare model consumption with the business case and investigate unexpected increases.

Incorrect Answers

Removing usage evidence makes token, license and escalation cost management less reliable, not more.

Arbitrary charging conceals which workload needs cost control and distorts investment decisions.

Unbounded consumption can create immediate overspend without enough information to assign costs to responsible workflows.

Question 450

A healthcare provider finished a chatbot proof of concept. Security assessment, incident ownership and business continuity tests are unfinished, but leaders request immediate rollout to all clinics. What should the adoption roadmap require?

Roll out to all clinics and defer safety controls until model usage patterns are known.
Treat user satisfaction in the pilot as equivalent to tested business continuity.
Move the same exploratory credentials to production to avoid creating separate access policies.
Complete defined governance, operational recovery and safety acceptance gates before phased clinic rollout.

Explanation

Correct Answer

A successful pilot establishes feasibility but does not replace production security, monitoring and recovery readiness.

Incorrect Answers

Widespread release would multiply the impact of unresolved privacy, reliability and support defects.

Positive feedback cannot prove operation during connectivity loss, downstream service errors or credential revocation.

Prototype permissions may be excessive or unmanaged and should not be carried into patient-facing production service.

Agentic AI Business Solutions Architect (AB-100) Practice Tests

Microsoft AB-100 Practice Test 43 - 10 Questions (Last Reviewed 02-Oct-2026)
Microsoft AB-100 Practice Test 44 - 10 Questions (Last Reviewed 01-Oct-2026)
Microsoft AB-100 Practice Test 45 - 10 Questions (Last Reviewed 30-Sep-2026)
Microsoft AB-100 Practice Test 46 - 10 Questions (Last Reviewed 29-Sep-2026)
Microsoft AB-100 Practice Test 47 - 10 Questions (Last Reviewed 28-Sep-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 440

Explanation

Correct Answer

Incorrect Answers

Question 441

Explanation

Correct Answer

Incorrect Answers

Question 442

Explanation

Correct Answer

Incorrect Answers

Question 443

Explanation

Correct Answer

Incorrect Answers

Question 444

Explanation

Correct Answer

Incorrect Answers

Question 445

Explanation

Correct Answer

Incorrect Answers

Question 446

Explanation

Correct Answer

Incorrect Answers

Question 447

Explanation

Correct Answer

Incorrect Answers

Question 448

Explanation

Correct Answer

Incorrect Answers

Question 449

Explanation

Correct Answer

Incorrect Answers

Question 450

Explanation

Correct Answer

Incorrect Answers

Agentic AI Business Solutions Architect (AB-100) Practice Tests