Microsoft AB-100 Practice Test 24

Question 230

An enterprise model router moves routine chats to a cheaper deployment. Which monitoring set confirms that savings did not degrade quality?

Total API spend only, ignoring failed answers.
A single aggregate chat satisfaction score without workload segments.
Track request counts per model but omit correctness, routing decisions and failure rates by request class.
Per-model cost, routing reason, error rates and task-level success broken down by request class.

Explanation

Correct Answer

A routing policy needs both economic and outcome evidence for every class it redirects.

Incorrect Answers

Lower spend with higher errors may be a net loss.

Aggregate satisfaction can conceal high-risk class failures.

Counts can establish traffic and expense allocation but not whether moving a high-risk class to a cheap model degraded outcomes.

Question 231

An agent invokes a CRM tool after reading a customer name. The trace shows HTTP 200, but the wrong customer was updated. What should the trace capture to diagnose this?

Log only the final customer display name and a successful tool-return flag, omitting the resolved CRM record ID.
Only the successful HTTP status code.
Validated customer ID, caller authorization context and the business record returned by the update.
A model-generated statement that the update looked correct.

Explanation

Correct Answer

Transport success does not prove that the authorized business entity was changed.

Incorrect Answers

Customer names are not unique; without the resolved record ID, caller identity and before/after target, the trace cannot isolate the incorrect write.

HTTP status cannot distinguish a wrong-target update.

Narrative confidence is not a record audit.

Question 232

A customer contact agent receives complaints of unsatisfactory handoffs to staff. Which backlog dimensions should be tracked?

Aggregate customer handoff complaints only by the agent identifier, with no recorded reason, case stage or actual downstream outcome.
Discard every issue that required human assistance.
Score the overall sentiment of handoff conversations without checking final assignment, transferred context or downstream resolution.
Escalation reason, task stage, customer impact and whether the handoff preserved useful case context.

Explanation

Correct Answer

Root-cause dimensions reveal whether the agent fails to resolve, route or preserve context.

Incorrect Answers

Grouping by agent alone misses which workflow step lost context and whether escalated customers received a correct response.

Escalations are important evidence, not discardable noise.

Sentiment does not identify whether the right specialist received enough authorized case information to complete the task.

Question 233

A support agent has high model token latency but fast database queries. What should performance tuning target first?

Tune CRM query indexing before checking model prompt size or generation time even though traces show the queries complete quickly.
Increase the database index size without a related bottleneck.
Switch to a higher-parameter model without measuring token throughput or quality on the support workload.
Model request size, deployment capacity and generated-token requirements, with quality regression tests.

Explanation

Correct Answer

Tracing separates model generation from external tool duration, enabling targeted optimization.

Incorrect Answers

The observed database stage is already fast. Optimize the measured model inference bottleneck and then retest overall latency.

The measured database stage is not the bottleneck.

Greater capacity or parameter count does not guarantee faster inference; compare validated candidate latency under equivalent context and load.

Question 234

A purchase-recommendation agent often completes its chat but fails to post orders into ERP. Which KPI represents the relevant business outcome?

Percentage of authorized orders durably posted with correct amounts, not percentage of chats answered.
Count cases in which the agent receives an initial order API acknowledgment without checking the resulting ERP posting.
Number of proposed orders regardless of posting result.
Measure whether the agent formatted the proposed purchase order correctly without validating the supplier acknowledgment or durable posted state.

Explanation

Correct Answer

Operational success requires the intended transaction to be confirmed at the system of record.

Incorrect Answers

An accepted request may later fail validation or commit; durable order ID and posted value are the business outcome.

Recommendations without commitment do not satisfy the posting workflow.

A well-formatted recommendation can be followed by a failed transaction; successful business completion requires persisted order evidence.

Question 235

An insurer’s agent answers routine claims well but becomes inaccurate for a newly launched product. What tuning trigger is defensible?

Train on all unlabeled claims immediately with no controls.
Wait for a general aggregate accuracy decline while ignoring product segmentation.
Segmented postlaunch error evaluation against labeled new-product claims and a verified remediation plan.
Retune the model using historical product cases before isolating errors or label coverage for the newly launched product.

Explanation

Correct Answer

New distributions require targeted evidence and safe evaluation before model or retrieval changes.

Incorrect Answers

Unvalidated training can amplify errors.

Aggregate metrics can hide a concentrated high-impact failure.

Broad tuning can leave the new-product failure unchanged; first measure segmented errors against labeled current examples.

Question 236

A router sends model requests based on estimated complexity. Easy requests are successful, but a small subset of legal inquiries are routed to the cheap model and fail. Which metric should the router use?

Route solely by estimated token length, treating short legal requests as low risk without a policy-risk classification.
Failure and escalation rate by risk class, then refine the routing threshold for legal requests.
Overall token saving only.
Count of model deployments irrespective of task performance.

Explanation

Correct Answer

A model router should preserve quality on the high-risk tails, not merely minimize average cost.

Incorrect Answers

Short legal questions can require high-accuracy interpretation; measured downstream failures should define when the route escalates.

Savings are misleading when legal errors rise.

Model count does not establish routing fitness.

Question 237

An agent’s payment tool sometimes times out after the provider commits the charge. Select TWO trace fields that are essential for safe incident analysis.

No record of tool arguments or outcome because the provider is external.
The provider transaction status or receipt retrieved independently after the timeout.
A fresh random payment key for each retry without storing prior attempts.
A stable idempotency/correlation key linking attempts to the same intended payment.
Only the assistant’s positive completion sentence.

Explanation

Correct Answers

An independent provider record distinguishes a committed charge from a genuine failure.

A stable key allows retries and traces to be linked to one business intention.

Incorrect Answers

Missing transaction evidence defeats root-cause and compensation analysis.

New keys can duplicate charges during uncertain retries.

Generated success prose is not financial evidence.

Question 238

A case agent receives 400 “wrong category” complaints. Which analysis best informs corrective work?

Label a representative set by expected category and inspect confusion patterns across case types.
Delete rare categories because they cause difficult examples.
Choose the most frequently used response template as a universal fix.
Count total complaints without linking them to actual classifications.

Explanation

Correct Answer

Ground-truth labels and class-level errors identify where routing needs refinement.

Incorrect Answers

Deleting real categories sacrifices legitimate business requirements.

Template popularity does not explain category confusion.

Volume alone cannot locate wrong decisions.

Question 239

A business solution has reliable model latency but a payment API p95 jumps above its SLA. Which operational diagnosis is supported?

The model must be the bottleneck because all calls occur in an agent.
The dependency stage is slow; investigate payment API saturation and retry behavior using traces.
Hide payment API timing to keep the dashboard simple.
Scale the language-model deployment before checking payment API service saturation despite traces showing the tool is the slow stage.

Explanation

Correct Answer

Segmented telemetry isolates the external service rather than misattributing end-to-end delay.

Incorrect Answers

Agent membership does not make model inference responsible for every stage.

Missing timing makes troubleshooting harder.

Additional model capacity will not reduce a payment-service p95 regression; diagnose payment-side concurrency, retries and errors.

Question 240

An AI-led service process completes 1,000 requests, but 150 are later reversed for wrong eligibility. Which metric is needed for management?

Initial request count only, because reversals happen later.
Optimize average inference cost per initial request while excluding the cost and count of eligibility reversals.
Treat the ERP’s initial acceptance response as final eligibility success without reconciling later reversals.
Correct sustained completion after reversals, with error-rework and exception costs.

Explanation

Correct Answer

Business value depends on durable correct outcomes, not temporary successful-looking transitions.

Incorrect Answers

Ignoring reversals inflates completion quality.

Token efficiency is an operating-cost component but cannot substitute for sustained correct outcomes and remediation cost.

A provisional submission or acceptance can be reversed; sustained correct fulfillment and rework cost determine success.

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

Microsoft AB-100 Practice Test 22 - 10 Questions (Last Reviewed 23-Oct-2026)
Microsoft AB-100 Practice Test 23 - 10 Questions (Last Reviewed 22-Oct-2026)
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)

Pass Your Microsoft AB-100 Exam Easy!

Question 230

Explanation

Correct Answer

Incorrect Answers

Question 231

Explanation

Correct Answer

Incorrect Answers

Question 232

Explanation

Correct Answer

Incorrect Answers

Question 233

Explanation

Correct Answer

Incorrect Answers

Question 234

Explanation

Correct Answer

Incorrect Answers

Question 235

Explanation

Correct Answer

Incorrect Answers

Question 236

Explanation

Correct Answer

Incorrect Answers

Question 237

Explanation

Correct Answers

Incorrect Answers

Question 238

Explanation

Correct Answer

Incorrect Answers

Question 239

Explanation

Correct Answer

Incorrect Answers

Question 240

Explanation

Correct Answer

Incorrect Answers

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