Microsoft AB-100 Practice Test 26

Question 250

A knowledge agent receives positive feedback on article summaries but repeated reports of missing citations. How should product management interpret that?

Close all citation issues because users like the writing.
Record citation availability as a separate quality defect rather than allowing satisfaction to mask traceability gaps.
Ask the model to fabricate a plausible URL for every answer.
Display article titles as citations without checking that each title resolves to the paragraph supporting the answer.

Explanation

Correct Answer

A useful summary still needs independently checkable evidence for factual and compliance-sensitive claims.

Incorrect Answers

Good sentiment does not ensure trustworthy attribution.

Invented URLs mislead reviewers.

Decorative or unresolved titles are not traceable evidence, even when the answer is otherwise well-liked.

Question 251

During a high-volume sales campaign an agent logs many 429 responses from its CRM connector. Which monitoring report should be created?

Report aggregate CRM requests per hour without separating throttled 429 responses from successful requests.
A summary that excludes throttled calls from availability metrics.
Report successful CRM authentication counts without distinguishing successful calls from rate-limited requests.
Requests per minute, rate-limit counts, retry delay and affected business actions by tenant.

Explanation

Correct Answer

Rate-limit behavior is a measurable dependency bottleneck needing bounded retry and capacity planning.

Incorrect Answers

Traffic volume does not establish the rate-limit impact without status distribution, backoff time and failed business actions.

Omitting 429s misstates reliability.

Authentication success does not measure connector throttling. The report must include 429 incidence, request rate and bounded retry impact.

Question 252

An agent’s tool says a purchase was accepted, but 10% of orders lack confirmed ERP record IDs. Which success KPI is correct?

Percentage of tool responses containing the word “accepted”.
Percentage of orders with verified durable ERP identifiers and correct financial values.
The model’s confidence that its order explanation is complete.
Count approved-looking order confirmations in chat without reconciling them to a durable ERP order identifier.

Explanation

Correct Answer

Durable business state and amount reconciliation are stronger evidence than API or text success claims.

Incorrect Answers

Transport messages can be provisional.

Generated confidence cannot replace ERP evidence.

Generated confirmations can exist even when ERP posting fails; business success needs a persisted receipt and value reconciliation.

Question 253

A service agent’s quality declined after policy documents were reindexed. What evidence should trigger rolling back the new index?

A single lower-than-normal token count without incorrect outcomes.
Rebuild the index because its filename changed without comparing source revision coverage or answer grounding outcomes.
The fact that indexing took longer this week than last week.
A controlled before/after increase in source-attribution errors tied to the index version.

Explanation

Correct Answer

A verified quality regression linked to the index change warrants rollback or correction.

Incorrect Answers

Token counts alone may reflect benign content differences.

A filename change is not evidence of a retrieval defect; investigate the actual published corpus and unsupported claims.

Elapsed indexing time need not imply answer corruption.

Question 254

A model router promotes low-risk translation requests to a cheaper model, while legal summaries remain on a stronger deployment. What should monitoring retain?

One success metric for all language and legal workloads with no segmentation.
Publish only an aggregate model success score and remove the route-class labels needed to reproduce class-specific errors.
Per-class routing decisions, model version, quality outcomes and token expense.
Only aggregate spend without tracking which model served which work.

Explanation

Correct Answer

The router must show that cost savings preserve risk-specific accuracy and traceability.

Incorrect Answers

Different workloads need different acceptance measures.

Without class attribution, a route responsible for a bad business decision can be masked by stronger unrelated routes.

Unattributed spend cannot diagnose regressions.

Question 255

An agent uses three tools but traces never show which one returned a conflicting customer ID. Which data should be added?

A generic success flag for the whole conversation.
Use a generic conversation screenshot without recording which tool returned the conflicting values or their underlying record IDs.
Tool name/version, call inputs, validated output identifiers and a correlated request trace.
Only the final chatbot reply, without tool metadata.

Explanation

Correct Answer

Tool-level structured evidence identifies where identity mismatches enter the workflow.

Incorrect Answers

One flag cannot isolate an individual dependency.

An image of the UI cannot identify the failing dependency or distinguish tool output from generated model claims.

Replies can hide intermediate data errors.

Question 256

A retail customer agent receives repeated “wrong language” feedback after a channel migration. Which feedback analysis is appropriate?

Cross-tabulate the complaint by language preference, channel and deployment version.
Change every user’s language setting to English without consent.
Ignore complaints if the overall active-user count increased.
Use aggregate language-detection accuracy without grouping wrong-language cases by affected channel and migration version.

Explanation

Correct Answer

Segmenting feedback by deployment context reveals language negotiation or migration defects.

Incorrect Answers

Forced language changes can reduce usability and trust.

Volume growth does not invalidate customer issues.

Overall detection can remain strong while one migrated route selects the wrong locale; the analysis must retain channel and release dimensions.

Question 257

A call summarization model has high average accuracy but frequently omits the customer’s cancellation request. Which telemetry should be elevated?

Score summaries only by embedding similarity to a reference answer without testing whether specific contractual obligations were omitted.
Use overall agent satisfaction to certify that cancellation requests are captured, without labeled cancellation-intent recall tests.
Recall of cancellation intent and downstream action completion on labeled call samples.
The number of audio files processed without human validation.

Explanation

Correct Answer

Missing critical intents requires task-specific recall and workflow outcome checks.

Incorrect Answers

A semantically similar summary can still omit a critical obligation; evaluate evidence-backed required facts and omissions.

A pleasant interaction can still omit a high-impact instruction; the desired outcome must be measured directly against call evidence.

Throughput does not establish faithful summaries.

Question 258

A claim approval agent is accurate for established products but poor on newly introduced coverage tiers. What monitoring action is justified?

Increase retrieval top-K for new products without verifying that the published product descriptions are indexed and current.
Alert on product-tier segmented error rates and verify retrieval or training coverage before tuning.
Wait until every product category becomes inaccurate.
Remove new-product examples from QA to preserve the aggregate pass rate.

Explanation

Correct Answer

Segmented drift monitoring detects targeted distribution changes early.

Incorrect Answers

More search results cannot correct stale or absent new-product information. Confirm source freshness and the version used by retrieval.

Waiting for universal decline leaves high-risk errors active.

Excluding failures misrepresents production quality.

Question 259

After a contract-summary agent is moved to a lower-cost model, its error rate increases for long documents and newly issued contract templates. Which drift/regression signal should trigger investigation?

Track the cheaper model’s unit cost only and consider the change successful whenever average spend falls.
Segment grounded-summary failures by contract length and template/version, comparing them against the validated prior deployment and review threshold.
Disable legal-source attribution tests for long contracts so short responses dominate the quality score.
Use only overall chat satisfaction, without examining unsupported clauses or document-length effects.

Explanation

Correct Answer

Versioned, segment-specific failure rates identify whether deployment and input-distribution changes caused unsupported or missing contract conclusions.

Incorrect Answers

A cheaper route can create hidden quality regressions; unit price does not indicate whether new templates or truncated context harm legal accuracy.

Removing hard cases hides a regression rather than measuring its source or consequence.

Aggregate satisfaction can conceal failures on long or recently updated legal documents that require grounded, clause-level evaluation.

Question 260

A model router serves FAQ, invoice, and regulated legal requests using different deployments. The monthly invoice rises even though total request count is unchanged. Which logging design allows an architect to diagnose workload-specific cost and quality?

Log only the blended monthly invoice total, omitting request class and selected model.
Log request class, model deployment/version, route reason, input/output token charges and task outcome for each correlated request.
Record only which user opened the chat page, not the actual routed model call or its token usage.
Track raw token volume without model rates, routing decision or business class, then attribute all cost change to user growth.

Explanation

Correct Answer

Per-request class, route, model version and charged usage show whether expensive paths were overused and whether quality thresholds were preserved.

Incorrect Answers

Aggregate spend cannot distinguish a routing change from a model-rate increase or a higher-cost request mix.

Page views do not establish actual inference cost or whether the router escalated requests to an expensive deployment.

Token count alone lacks per-model price and routing context; unchanged traffic can still cost more when request-class allocation shifts.

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

Microsoft AB-100 Practice Test 24 - 10 Questions (Last Reviewed 21-Oct-2026)
Microsoft AB-100 Practice Test 25 - 10 Questions (Last Reviewed 20-Oct-2026)
Microsoft AB-100 Practice Test 26 - 10 Questions (Last Reviewed 19-Oct-2026)
Microsoft AB-100 Practice Test 27 - 10 Questions (Last Reviewed 18-Oct-2026)
Microsoft AB-100 Practice Test 28 - 10 Questions (Last Reviewed 17-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 250

Explanation

Correct Answer

Incorrect Answers

Question 251

Explanation

Correct Answer

Incorrect Answers

Question 252

Explanation

Correct Answer

Incorrect Answers

Question 253

Explanation

Correct Answer

Incorrect Answers

Question 254

Explanation

Correct Answer

Incorrect Answers

Question 255

Explanation

Correct Answer

Incorrect Answers

Question 256

Explanation

Correct Answer

Incorrect Answers

Question 257

Explanation

Correct Answer

Incorrect Answers

Question 258

Explanation

Correct Answer

Incorrect Answers

Question 259

Explanation

Correct Answer

Incorrect Answers

Question 260

Explanation

Correct Answer

Incorrect Answers

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