Microsoft AB-100 Practice Test 41

Question 400

A marketplace indexes seller invoices in a shared search service. Each invoice source record also embeds a customer shipping address and internal wholesale margin that guest sellers must not see. A seller is authorized for their own invoice totals only. What permission-trimming architecture prevents the model from receiving the restricted fields?

Retrieve entire authorized invoice documents and ask the model to omit shipping addresses and wholesale margins in its final answer.
Rank visible invoice totals above restricted fields so sensitive snippets are less likely to appear in search results.
Allow full document retrieval and mask sensitive fields in the rendered user interface after the model has used the content.
Separate protected fields or passages into permission-scoped records and enforce both seller ownership and field/passage entitlements before grounding.

Explanation

Correct Answer

Enforce permission-aware field projection or chunk-level filtering at the source/retrieval boundary, and verify tenant ownership so only entitled invoice facts enter the model context.

Incorrect Answers

Document ownership alone does not authorize every field. Once a full invoice chunk enters context, output instructions cannot guarantee sensitive margin and address data remain unseen.

Ranking is not authorization: lower-ranked but retrieved chunks can still disclose protected values to the model or its traces.

Client-side masking does not revoke model-context access to restricted invoice fields, even when they do not appear visibly in the answer.

Question 401

A wholesaler receives standardized EDI orders. Product IDs, tax rules, stock checks and approval thresholds are all available as validated fields; exceptions are already routed to staff. What should the architect implement for the normal order path?

A free-form autonomous agent that can waive approval limits when order volume spikes.
A model-only flow that validates orders after inventory has already been reserved.
A deterministic orchestration flow with explicit validations and an exception queue, adding an agent only where unstructured interpretation is actually needed.
A generative agent that independently invents each tax and quantity check from order descriptions.

Explanation

Correct Answer

All normal-path decisions follow known rules; deterministic execution provides repeatable behavior, lower inference cost and transaction traceability.

Incorrect Answers

High load does not delegate permission to bypass enforceable transaction thresholds; the workflow must keep those gates.

Post-reservation checks permit invalid stock commitments before the required precondition and approval decisions have occurred.

A generative interpretation can depart from the authoritative EDI rules and is unnecessary for already typed order values.

Question 402

A shipping workflow deterministically computes late-delivery credits, but emailed freight claims include handwritten notes, photos and conflicting carrier narratives. Which capability should be assigned to an agent?

Calculate credit amounts directly from the complaint sentiment score without consulting the tariff table.
Override the shipping ledger whenever a photographed note seems more persuasive than a signed delivery record.
Auto-pay all claims that contain the words delayed delivery before the exception review is complete.
Extract and compare evidence, cite the conflicting passages and propose an exception category for a verified claims reviewer.

Explanation

Correct Answer

The model can interpret unstructured material while an authorized person and the credit engine retain the final award decision.

Incorrect Answers

Sentiment describes frustration, not the contractual amounts and delivery milestones that determine credit eligibility.

Unverified images cannot silently displace authoritative delivery milestones; conflicting evidence needs source resolution.

A matching phrase does not establish contractual coverage or authorization for a financial adjustment.

Question 403

A facilities team gets structured temperature readings and free-text reports about unusual smells near equipment. The emergency shutdown rule is fixed, but the reports need triage. Which design best separates responsibilities?

Use deterministic alarm thresholds for emergency shutdown and an evidence-linked agent to classify descriptions for an operator.
Replace calibrated thresholds with the model’s summary of the latest incident reports.
Let the model suppress sensor alarms whenever its narrative analysis suggests a benign condition.
Run the narrative classifier on readings only and skip the actual free-text reports.

Explanation

Correct Answer

Safety-critical shutdown triggers remain enforceable, while language interpretation assists but cannot overrule fixed sensor interlocks.

Incorrect Answers

A text summary is not an instrumented measurement and cannot provide a reliable interlock for imminent equipment risk.

An agent lacks authority to invalidate the fixed safety interlocks even if a narrative seems reassuring.

Sensor data alone cannot perform the requested language triage of unusual observations reported by staff.

Question 404

An accounts-payable robot matches invoices that have exact purchase-order and receipt IDs. A small group of invoices lacks a clear line-item correspondence and needs investigative judgment. What is the best split?

Route every perfectly matched invoice through an expensive full-context review before applying the existing matching rules.
Keep exact three-way matches in rules; ask an agent to assemble evidence for unmatched lines without posting payment.
Replace the exact-match payment checks with an agent that decides whether two amounts look similar.
Post unmatched invoices immediately and send their discrepancies to an agent afterward.

Explanation

Correct Answer

A limited reasoning assistant improves investigation while the payment service still checks matched identifiers and approval status.

Incorrect Answers

That ordering adds cost and latency without improving a clear deterministic decision already supported by trusted identifiers.

Amount similarity is not documentary proof of receipt, purchase authorization or an acceptable payment discrepancy.

Retroactive explanation cannot undo an unapproved settlement or prevent duplicate payment on a mismatched line.

Question 405

An agent reports that stores using a new coaching bot sell 18% more, but those stores were already the highest performers. Leadership wants evidence that the bot improved sales. What study design is more defensible?

Ask a generative model to rewrite the sales chart until it expresses the desired uplift.
Use only the bot users’ satisfaction surveys as a proxy for incremental gross revenue.
Compare change against matched or randomized control stores while accounting for baseline performance and seasonality.
Treat the raw 18% cross-store difference as attributable uplift without a baseline.

Explanation

Correct Answer

A controlled comparison helps distinguish intervention impact from selection bias and pre-existing sales differences.

Incorrect Answers

Changing narrative wording cannot establish counterfactual performance or remove systematic selection bias.

Satisfaction is valuable adoption data but does not establish how many sales would have occurred without the bot.

The selection of high-performing stores confounds the observed gap and makes a causal claim unsupported.

Question 406

A service agent processed 2,400 tickets; 300 were reopened within seven days. Its dashboard reports 2,400 resolved cases. Which verified outcome should be used to evaluate durable resolution?

Count every initial close as a successful resolution, even if the issue returns the next day.
Discard the reopened tickets from both the numerator and denominator without recording the exclusion.
Use average response length as the primary measure of ticket resolution.
Track independently confirmed resolutions without reopenings in the defined window, alongside total throughput.

Explanation

Correct Answer

A ticket closure event is not durable resolution; the reopen rate makes the reported 2,400 completions misleading.

Incorrect Answers

Reopened cases reveal that closure events can overstate the business outcome under the seven-day quality definition.

Removing failures from the evaluated population artificially inflates the score and prevents an auditable outcome rate.

Response length describes generated content, not whether the customer issue was actually and durably resolved.

Question 407

A debt-collection prediction assistant has strong overall accuracy, but it regularly misclassifies newly onboarded small businesses because the training set is dominated by large enterprises. What evidence should control release?

Measure false positives and missed cases by business segment, then remediate the failing small-business slice.
Ship the model based only on its overall accuracy across all clients.
Remove small-business cases from evaluation until the overall metric improves.
Tune the answer style to sound more confident when predicting a small-business case.

Explanation

Correct Answer

Aggregate accuracy masks a segment-specific distribution gap with potentially different financial and compliance consequences.

Incorrect Answers

The abundant enterprise records can dominate the aggregate and hide unacceptable errors for the smaller segment.

Excluding the failing population conceals the actual operational risk rather than improving prediction quality.

Confidence of prose has no direct effect on class-conditional error rates or missing small-business training evidence.

Question 408

A procurement analytics assistant predicts 1,000 units of demand next month and supplies a 95% interval of 990–1,010. Historical realized demand is volatile, often differing by 150 units. What should the architect test?

Multiply the predicted units by 95% and call the result a validated lower forecast bound.
Back-test whether stated prediction intervals actually contain held-out outcomes at approximately their claimed rate.
Treat the 95% label as guaranteed coverage because it was generated by a model.
Use the point estimate only and suppress all interval information to simplify purchasing decisions.

Explanation

Correct Answer

An unusually narrow interval may be miscalibrated even when the average point forecast has little bias.

Incorrect Answers

A confidence percentage is not a scaling factor for demand and does not calculate prediction uncertainty.

A probability label alone is not observed calibration; volatile historical errors may invalidate the interval width.

Removing uncertainty hides the potential stockout or overstock range that the purchasing decision needs.

Question 409

A payments agent shows a director a $12,000 transfer to supplier A. After approval, a retry supplies a different beneficiary account in the tool payload. Which control prevents misuse of the earlier approval?

Let the model certify the new beneficiary if its explanation uses the supplier’s legal name.
Compare only the director’s name in the chat transcript and execute the new payload.
Bind the approval to an immutable transaction ID and the reviewed beneficiary, amount and account fields, rejecting a changed payload.
Treat any previously approved transfer of $12,000 as permission for all transactions of that size.

Explanation

Correct Answer

A signed decision must apply to the exact transaction details; an altered beneficiary requires another authorized approval.

Incorrect Answers

Language similarity cannot provide the transaction-specific signer authorization required by the payment service.

A transcript identity without immutable payload binding cannot prove that the person approved the altered transfer.

Equal monetary amounts do not establish authorization for a new beneficiary or different funding account.

Question 410

A plant manager approves a high-risk maintenance action valid for the next 20 minutes. The field agent attempts to execute it two hours later using the same token. What should occur?

Extend the approval indefinitely whenever the agent’s retry counter is nonzero.
Execute first, then refresh the approval timestamp in the agent’s stored narrative.
Treat a read-only work-order view as an equivalent approval for the later execution.
Reject the expired approval and request a new authorization tied to the current maintenance request.

Explanation

Correct Answer

Short-lived approval prevents stale authority being reused after the operational conditions or responsible staff change.

Incorrect Answers

Retries do not prolong delegated authority; execution must respect the actual expiry of the approved window.

Backdating or rewriting the narrative would conceal an unauthorized operation rather than satisfy the prerequisite.

Viewing an approved request does not prove the current caller holds valid time-bound execution authority.

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

Microsoft AB-100 Practice Test 39 - 10 Questions (Last Reviewed 06-Oct-2026)
Microsoft AB-100 Practice Test 40 - 10 Questions (Last Reviewed 05-Oct-2026)
Microsoft AB-100 Practice Test 41 - 10 Questions (Last Reviewed 04-Oct-2026)
Microsoft AB-100 Practice Test 42 - 10 Questions (Last Reviewed 03-Oct-2026)
Microsoft AB-100 Practice Test 43 - 10 Questions (Last Reviewed 02-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 400

Explanation

Correct Answer

Incorrect Answers

Question 401

Explanation

Correct Answer

Incorrect Answers

Question 402

Explanation

Correct Answer

Incorrect Answers

Question 403

Explanation

Correct Answer

Incorrect Answers

Question 404

Explanation

Correct Answer

Incorrect Answers

Question 405

Explanation

Correct Answer

Incorrect Answers

Question 406

Explanation

Correct Answer

Incorrect Answers

Question 407

Explanation

Correct Answer

Incorrect Answers

Question 408

Explanation

Correct Answer

Incorrect Answers

Question 409

Explanation

Correct Answer

Incorrect Answers

Question 410

Explanation

Correct Answer

Incorrect Answers

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