Microsoft AB-100 Practice Test 70

Question 690

Users give a shipment assistant low satisfaction scores during a carrier outage, although it correctly reports unavailable status. What diagnostic distinction matters?

Force the agent to invent tracking updates so customer satisfaction rises.
Call all low ratings model hallucinations without examining dependency status.
Delete negative ratings that coincide with outages.
Separate user sentiment, factual correctness and upstream service availability in feedback classification.

Explanation

Correct Answer

Low satisfaction may reflect unavoidable service unavailability rather than an inaccurate agent answer.

Incorrect Answers

False delivery facts harm reliability even when they temporarily reassure users.

The error could be in a carrier outage or expectation mismatch, not generated facts.

Feedback remains useful for designing clearer fallback and recovery experiences.

Question 691

A knowledge agent receives repeated reports that its refund policy is wrong. User comments say only “bad answer,” with no document or run reference. What should feedback capture?

Keep an authorized run ID, policy citation/revision and expected correct rule with the reported failure.
Aggregate only the total number of negative emoji reactions.
Attach the reporter’s entire inbox to each feedback item.
Ask the model to rewrite policy text from memory based on complaint frequency.

Explanation

Correct Answer

Actionable feedback needs the exact evidence and context used by the agent for reproducible review.

Incorrect Answers

An aggregate count cannot isolate a policy revision or incorrect clause.

Unrelated personal communications do not help fix a particular policy-grounding defect.

Source-of-truth corrections require documented ownership and review, not inferred popular sentiment.

Question 692

Two user complaints say “order not found.” One stems from wrong tenant scope and the other from delayed data synchronization. How should the backlog triage them?

Silently close both tickets when one sample query starts working.
Group the surface symptom but retain separate verified root-cause labels and remediations.
Merge them into one prompt-rephrasing task without investigating tools.
Fix tenant permissions broadly for every user as the common solution.

Explanation

Correct Answer

Identical wording does not imply identical authorization or freshness defects.

Incorrect Answers

One resolved scenario does not demonstrate the other failure mode is fixed.

A wording change cannot fix backend tenant access or synchronization timing.

Expanding access could expose other orders and still leave stale data.

Question 693

A new routing prompt is released after complaints about misclassified cancellations. What should the team compare?

Assume every newer prompt is safer because it is longer.
Replace regression cases with one favorable operator demo.
Before-and-after intent confusion rates on the same representative cancellation cohorts and downstream action errors.
Compare only the number of positive comments in the release announcement.

Explanation

Correct Answer

Measuring actual route and mutation outcomes shows whether the fix works without new harm.

Incorrect Answers

Prompt length has no guaranteed relationship to intent accuracy or access boundaries.

A single demonstration cannot cover the previously confused intents.

Announcements do not test which transactional action the agent actually chooses.

Question 694

A service agent performs well overall but receives frequent negative feedback from users in one supported language. What is a responsible diagnostic strategy?

Remove that language from reports to keep the global score high.
Export all conversations publicly so any developer can inspect them.
Assume all negative feedback is caused by users misunderstanding the system.
Compare language-specific task success, escalation and error patterns with minimized, authorized example evidence.

Explanation

Correct Answer

Aggregate accuracy may hide cohort-specific failures that require targeted evaluation and remediation.

Incorrect Answers

Suppression hides potential unequal service quality rather than correcting it.

Unrestricted disclosure is unnecessary to evaluate the specific language cohort.

Unverified attribution ignores possible translation, retrieval or action-routing defects.

Question 695

User feedback reveals one invoice was paid twice after an agent retry, while ten reports concern awkward formatting. How should the release owner prioritize?

Escalate the confirmed duplicate-write defect, disable unsafe retries and add idempotent transaction validation.
Fix invoice-formatting complaints first because more users reported them.
Tell users to disregard the second payment without checking Finance records.
Increase retry frequency to minimize perceived payment delays.

Explanation

Correct Answer

Duplicate financial commitments are a material correctness risk and require immediate containment.

Incorrect Answers

Complaint quantity should not outweigh verified incorrect payments.

A chat statement cannot reverse an external committed transaction.

More uncontrolled retries intensify the existing duplicate-write mechanism.

Question 696

A help agent receives thousands of “still waiting” comments after routing cases to human support. Some cases were never enqueued; others are delayed by staffing. What should monitoring distinguish?

Close pending cases automatically after a fixed delay to improve completion metrics.
Separate agent handoff failures, queue acceptance and human-queue service delays with linked case IDs.
Count every handoff request as a successfully resolved customer case.
Blame model reasoning for all queue delays without service metrics.

Explanation

Correct Answer

Those stages have distinct owners and remediation actions even though the user sees one long wait.

Incorrect Answers

Artificial closure hides unresolved business needs and misrepresents outcomes.

Queue placement does not establish that a person accepted or resolved the issue.

Human staffing and dispatch limits may dominate response times.

Question 697

An agent’s answer latency increases after its knowledge index doubles. Model inference timing is unchanged, but vector search p95 triples. What should tuning focus on?

Upgrade the model without evaluating index performance.
Disable permission filters to make retrieval faster.
Optimize retrieval indexing, filter selectivity and search throughput, then measure p95 again.
Change the prompt’s tone to more concise wording as the primary fix.

Explanation

Correct Answer

The measured regression is confined to search rather than the generation stage.

Incorrect Answers

A faster model may not compensate for the actual search bottleneck.

Removing authorization might reduce computation but introduces unacceptable data leakage.

Shorter prose does not address the measured vector-search delay.

Question 698

A policy assistant continues producing fluent text, but citation validation drops from 98% to 70% after a connector migration. What should monitoring isolate?

Increase temperature so the agent cites a wider variety of documents.
Treat fluent prose as evidence that citation links are valid.
Disable citation checks so the dashboard returns to green.
Source-link resolution, permission filtering and revision metadata for migrated retrieval results.

Explanation

Correct Answer

Citations can fail independently of text quality when the source contract or passage IDs change.

Incorrect Answers

Randomness cannot restore broken source identifiers or approval metadata.

A plausible answer can still refer to missing or superseded passages.

Suppressing a failed metric conceals lost auditability rather than repairing it.

Question 699

An agent averages two seconds per reply, but one in twenty requests takes 25 seconds during a staffing shift. What metric best exposes the user-impacting tail?

Use p95/p99 response latency with concurrency and route segmentation.
Report only the fastest successful response.
Sum all tokens generated as a proxy for end-to-end latency.
Count successful logins instead of request duration.

Explanation

Correct Answer

Percentiles reveal slow cohorts hidden by a low average and help identify queue saturation.

Incorrect Answers

The minimum says little about the delayed requests most users complain about.

Token usage alone omits queue, retrieval and external action delays.

Authentication success is not a measure of user-perceived response time.

Question 700

An agent’s failed shipment tool calls rise after the carrier API starts returning HTTP 429. The model token volume remains stable. What should operators examine?

Mark each throttled attempt as completed because the HTTP service responded.
Rate-limit responses, retry backoff, concurrency and carrier quota utilization.
Increase retries to zero delay while retaining the same request burst.
Switch the agent’s display language without changing tool behavior.

Explanation

Correct Answer

429 signals a throttled downstream dependency; load shaping is more relevant than generated-answer verbosity.

Incorrect Answers

An error status is not a successful shipment operation.

Immediate loops amplify throttling and may worsen the carrier outage.

Translation does not create API capacity or reduce calls.

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

Microsoft AB-100 Practice Test 68 - 10 Questions (Last Reviewed 03-Nov-2026)
Microsoft AB-100 Practice Test 69 - 10 Questions (Last Reviewed 02-Nov-2026)
Microsoft AB-100 Practice Test 70 - 10 Questions (Last Reviewed 01-Nov-2026)
Microsoft AB-100 Practice Test 71 - 10 Questions (Last Reviewed 31-Oct-2026)
Microsoft AB-100 Practice Test 72 - 10 Questions (Last Reviewed 30-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 690

Explanation

Correct Answer

Incorrect Answers

Question 691

Explanation

Correct Answer

Incorrect Answers

Question 692

Explanation

Correct Answer

Incorrect Answers

Question 693

Explanation

Correct Answer

Incorrect Answers

Question 694

Explanation

Correct Answer

Incorrect Answers

Question 695

Explanation

Correct Answer

Incorrect Answers

Question 696

Explanation

Correct Answer

Incorrect Answers

Question 697

Explanation

Correct Answer

Incorrect Answers

Question 698

Explanation

Correct Answer

Incorrect Answers

Question 699

Explanation

Correct Answer

Incorrect Answers

Question 700

Explanation

Correct Answer

Incorrect Answers

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