Microsoft AB-100 Practice Test 74

Question 730

A shipping agent consumes at-least-once delivery events. The same order-ready event is redelivered before a carrier confirms collection. Which test case is necessary?

Send only one event and check that the agent writes a friendly message.
Treat each event as a new order and count two bookings as success.
Disable all event logs during the test to reduce noise.
Send duplicate event IDs and verify only one carrier booking is committed with appropriate retry traces.

Explanation

Correct Answer

Real brokers can redeliver messages, so idempotent orchestration must be tested explicitly.

Incorrect Answers

One happy path cannot reveal double-booking under redelivery.

One business intent should not create two financial or logistic commitments.

Event identifiers and outcomes are essential to prove deduplication.

Question 731

A warehouse assistant normally reads live stock. The inventory service becomes unavailable after the user asks for a binding delivery commitment. What negative scenario should be tested?

Simulate inventory timeout and require an explicit unavailable result with no committed delivery promise.
Allow the model to estimate current quantity from old shipments.
Assume a configured connector means stock data is always accessible.
Skip the outage test because it is not a normal user flow.

Explanation

Correct Answer

A safe agent must distinguish missing evidence from verified capacity before making a commitment.

Incorrect Answers

Historical deliveries cannot establish real-time available inventory.

Integration configuration does not prove dependency uptime.

Production reliability must cover foreseeable failures, not just ideal conditions.

Question 732

A legal assistant retrieves documents for two organizations in one index. The first positive test asks tenant A about A’s contract. What complementary test is essential?

Grant the test account global search access to minimize access errors.
Ask tenant A for tenant B’s uniquely identified contract and verify denial before retrieval reaches the model.
Ask tenant A to summarize a second authorized A contract.
Hide the B contract title in the final UI but allow backend retrieval.

Explanation

Correct Answer

Negative authorization cases prove isolation across the shared index.

Incorrect Answers

Overprivileged test identities cannot demonstrate tenant permission trimming.

Another positive case does not exercise the forbidden cross-tenant path.

Final display filtering cannot prove the model avoided unauthorized content.

Question 733

A voice agent hears an ambiguous charge amount, either $15 or $50, while refunds over $25 require approval. What should the acceptance test require?

Always submit $50 and rely on a later adjustment.
Ignore speech ambiguity if the rest of the transcript has high confidence.
Confirm the spoken amount and withhold payment execution until threshold and user authorization are resolved.
Pick $15 automatically because it avoids a manager workflow.

Explanation

Correct Answer

A recognition ambiguity changes the approval condition and cannot be guessed safely.

Incorrect Answers

Overpaying before disambiguation violates controlled refund execution.

Overall transcription confidence does not validate the decisive monetary value.

Shortcuts based on a cheaper outcome can misrepresent what the customer requested.

Question 734

An order-to-cash test creates a Sales order and times out while posting the Finance invoice. What test outcome demonstrates resilience?

The chatbot produces a polished apology with no status check.
Retry the full workflow with new IDs and count both orders.
Delete the order without examining Finance receipts.
The workflow reconciles the existing order and invoice state and resumes without duplicate postings.

Explanation

Correct Answer

End-to-end tests must cover partial commits because cross-service writes are not a single atomic model call.

Incorrect Answers

An apology is insufficient when the financial posting state is unknown.

Fresh identifiers can duplicate a transaction already committed before the timeout.

Unverified compensation can violate business policy and strand related records.

Question 735

A compliance agent must operate in two countries with separate approved regional model endpoints. How should validation cover the routes?

Execute region-scoped scenarios that verify endpoint residency, content permissions and equivalent expected task outcomes.
Test only the cheapest region and assume the second is identical.
Allow the test router to use any endpoint and examine the final prose only.
Use real private citizen records in an unrestricted test tenant.

Explanation

Correct Answer

A functional answer is not sufficient if data took an unauthorized international path.

Incorrect Answers

Different regional deployments can have different policies or versions.

Undocumented route selection cannot demonstrate residency compliance.

Uncontrolled production data may violate testing and processing permissions.

Question 736

A credit-limit agent correctly pauses for an approver, but testers only validate the approved path. Which missing case is material?

Record rejection in chat after applying the credit-limit change.
Test a rejected request and verify no credit change occurs, even if background retries run.
Use only successful approval requests because denial is not the expected workflow.
Treat an approver’s rejection as a temporary service error.

Explanation

Correct Answer

Approval denial must be terminal for the intended write until a new authorized request is created.

Incorrect Answers

Post-hoc acknowledgement does not undo an unauthorized commitment.

Negative paths are necessary to prove the authorization gate actually blocks execution.

Denial is a business decision, not a retryable technical timeout.

Question 737

A benefits agent cites one approved policy but makes three separate eligibility assertions, two unsupported by the citation. What evaluation method is needed?

Use model confidence as the support score.
Measure only how long the answer is.
Score each material claim for evidence support against the effective source, not merely whether any citation exists.
Count the answer as fully supported if it contains one hyperlink.

Explanation

Correct Answer

Citation presence alone can conceal unsupported conclusions within an otherwise grounded response.

Incorrect Answers

Confidence is not independent verification against the cited policy text.

Answer length cannot establish source entailment or factual validity.

One citation does not automatically support every claim.

Question 738

A support agent flags dangerous equipment failures for human escalation. Missing a genuine dangerous case is costly, while some false escalations are manageable. Which metric emphasis fits?

Optimize precision alone by escalating almost nothing.
Optimize the proportion of polite messages instead of classification outcomes.
Use only overall accuracy in a dataset dominated by harmless incidents.
Prioritize recall of dangerous cases while tracking precision and reviewer capacity.

Explanation

Correct Answer

High recall reduces missed serious incidents, with precision needed to manage false alarms.

Incorrect Answers

Low positive count can look precise while dangerous cases are missed.

Tone does not reveal whether an urgent failure was detected.

Class imbalance can hide poor detection of the rare high-risk class.

Question 739

An invoice agent explains a payment correctly but sends the wrong vendor ID in its tool call. Which evaluation distinguishes the defect?

Compare action payload, authorized identity and committed vendor record with expected values.
Use human ratings of the explanation’s fluency only.
Count any HTTP 200 from the payment API as correct.
Ignore the vendor field because the amount was correct.

Explanation

Correct Answer

A well-written rationale is insufficient when the business mutation targets the wrong counterparty.

Incorrect Answers

Prose quality does not validate tool arguments or payment beneficiary.

A successful response can still result from a semantically incorrect vendor identifier.

Accurate amount with wrong recipient remains a material transactional error.

Question 740

A policy assistant now abstains on 35% of questions, half because no approved source exists. How should coverage be evaluated?

Force answers for every case and measure only apparent completion.
Report supported-answer coverage separately from justified abstention and unnecessary refusal on answerable cases.
Count all abstentions as hallucinations.
Exclude abstentions entirely and report only accuracy among answered questions.

Explanation

Correct Answer

Safe abstention and capability failure are different outcomes requiring different responses.

Incorrect Answers

Suppressing refusal can increase false authoritative statements.

Withholding unsupported policy can be the correct behavior.

That conceals the service’s inability to address many user requests.

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

Microsoft AB-100 Practice Test 72 - 10 Questions (Last Reviewed 30-Oct-2026)
Microsoft AB-100 Practice Test 73 - 10 Questions (Last Reviewed 29-Oct-2026)
Microsoft AB-100 Practice Test 74 - 10 Questions (Last Reviewed 28-Oct-2026)
Microsoft AB-100 Practice Test 75 - 10 Questions (Last Reviewed 27-Oct-2026)
Microsoft AB-100 Practice Test 76 - 10 Questions (Last Reviewed 26-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 730

Explanation

Correct Answer

Incorrect Answers

Question 731

Explanation

Correct Answer

Incorrect Answers

Question 732

Explanation

Correct Answer

Incorrect Answers

Question 733

Explanation

Correct Answer

Incorrect Answers

Question 734

Explanation

Correct Answer

Incorrect Answers

Question 735

Explanation

Correct Answer

Incorrect Answers

Question 736

Explanation

Correct Answer

Incorrect Answers

Question 737

Explanation

Correct Answer

Incorrect Answers

Question 738

Explanation

Correct Answer

Incorrect Answers

Question 739

Explanation

Correct Answer

Incorrect Answers

Question 740

Explanation

Correct Answer

Incorrect Answers

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