Microsoft AB-100 Practice Test 73

Question 720

A service assistant encounters new repair codes and proposes tuning itself on all recent conversations, including restricted customer notes. How should adaptation be governed?

Disable model-version records after tuning to reduce administrative load.
Route novel-code errors into an approved data-review and evaluation cycle with consent and scoped training access.
Automatically train on every conversation after each misclassification.
Ignore the new repair codes while reporting no model errors.

Explanation

Correct Answer

Model improvement must preserve data rights, reproducible tests and promotion approval.

Incorrect Answers

Without version provenance future regressions cannot be traced or rolled back.

Unreviewed online training can ingest protected data and cause uncontrolled behavior drift.

Silent coverage gaps can lead to incorrect service guidance.

Question 721

A support agent routes simple order-status lookups to a high-cost reasoning model. A smaller model is proposed. Which experiment justifies the change?

Use larger context as the only measure of model suitability.
Route every status request to a random model and aggregate results.
Benchmark both models on representative authenticated status tasks for accuracy, p95 latency and cost per correct answer.
Choose the cheapest model with no real transaction test.

Explanation

Correct Answer

Lower price is useful only if the required retrieval and response behavior remains reliable.

Incorrect Answers

Context capacity is not an outcome metric for a small structured lookup.

Uncontrolled routing makes correctness and incident attribution more difficult.

A low token price does not guarantee correct status interpretation.

Question 722

A central router sends financial analysis to one model and simple FAQ to another. The dashboard only reports tenant-wide average token price. What breakdown is needed?

Use one average token price for all workloads.
Count only the number of configured models.
Remove model-deployment labels to simplify usage statistics.
Report model/deployment, workload class, answer correctness, tokens, latency and task completion per route.

Explanation

Correct Answer

Cost optimization requires knowing which tasks benefit from each model and whether quality changes.

Incorrect Answers

A mix of expensive and inexpensive requests can conceal poor route-specific economics.

Available model count does not describe routing effectiveness or cost.

Without route identity the team cannot attribute cost and quality to a model choice.

Question 723

A customer agent tries a small model, then always calls a large model even when the small response passed validation. Its costs double with no quality gain. What routing change is appropriate?

Invoke the fallback model only when defined quality, safety or capability criteria fail on the first route.
Keep calling both and retain whichever answer is longer.
Switch all tasks to the large model without measuring value.
Disable all response validation so only the first model is used.

Explanation

Correct Answer

Unconditional double inference is redundant when the initial answer already meets verified acceptance rules.

Incorrect Answers

Answer length is not evidence of better correctness or lower risk.

Universal escalation may increase cost for simple tasks without benefit.

Removing safety checks can reduce calls while allowing incorrect answers through.

Question 724

A model-routing change reduces average inference cost by 30% but increases human review from 5% to 18% of invoice cases. What economic metric should guide the decision?

Ignore the new review rate because humans are already employed.
Total cost per correctly completed invoice including inference, retries, correction and human-review labor.
Choose the route with the lowest price per thousand input tokens.
Compare only average generated response length.

Explanation

Correct Answer

A cheaper call may be more costly overall when it causes extra exceptions.

Incorrect Answers

Additional labor and delayed invoices have real operational opportunity cost.

Token price excludes substantial downstream handling costs.

Length does not capture invoice correctness or review effort.

Question 725

A global agent routes customer data to models hosted in multiple regions. The lowest-cost deployment resides outside the customer’s approved processing region. What routing restriction is required?

Assume encryption permits unrestricted international processing.
Let the model decide its own permitted geography from prompt phrasing.
Filter eligible model deployments by data residency and approved processing boundaries before applying cost optimization.
Select the cheapest endpoint then erase the region from telemetry.

Explanation

Correct Answer

A cheaper route is unacceptable when it violates the customer’s data-processing constraints.

Incorrect Answers

Encryption protects data in transit but does not automatically satisfy residency restrictions.

Regional policy must be enforced in infrastructure, not inferred from text.

Concealing geography does not change where protected data was processed.

Question 726

A primary model endpoint returns 503 during a busy period. A secondary supported model exists, but it has not been evaluated on regulated claims. What should the router do?

Route every regulated claim to the untested model automatically.
Generate decisions locally from stale chat history without an endpoint.
Mark the failed primary calls successful because an alternate route is configured.
Use the secondary only for approved evaluated workloads; queue or escalate regulated claims until a validated route exists.

Explanation

Correct Answer

Availability fallback must respect task-specific accuracy and risk acceptance, not only endpoint reachability.

Incorrect Answers

Technical availability does not prove compliance and claim quality on a different model.

Unsupported inference cannot replace validated claim rules or verified sources.

A configured fallback is not a committed outcome until an eligible task is actually processed.

Question 727

A team compresses retrieved troubleshooting text to reduce model tokens. Savings improve, but generated repair procedures omit safety warnings. What should govern promotion?

Reject or revise compression until held-out procedure tests retain mandatory safety details and acceptable task correctness.
Promote solely on token savings because concise context is always better.
Add a generic “be safe” instruction instead of restoring source warnings.
Ignore missing warnings if output grammar remains clear.

Explanation

Correct Answer

Input reduction is not beneficial if it removes evidence needed for safe technical recommendations.

Incorrect Answers

Lower cost cannot excuse omission of required safety steps.

An unsourced prompt cannot reliably reconstruct omitted procedure requirements.

Fluent text can still be technically unsafe or materially incomplete.

Question 728

A team proposes directing all VIP customer questions to the most expensive model and others to a small model, regardless of task complexity. What is a defensible routing design?

Always choose the largest model because it is inherently safer.
Choose a model by evaluated task requirements, approved data handling, latency and quality targets; apply service policies separately.
Use account status alone as a replacement for task-level evaluation.
Give every high-cost route global data-access privileges.

Explanation

Correct Answer

Customer service tier does not prove a particular model is needed for a simple transaction lookup.

Incorrect Answers

Large models still need task-specific validation and may incur needless cost.

VIP labeling says little about factual complexity or grounding requirements.

Model selection does not confer permission to view protected records.

Question 729

A procurement agent can autonomously submit orders below $500 but must request human signoff at $500 or above. Which test best checks the boundary?

Use one generated purchase summary with no actual transaction status.
Submit a $700 purchase once and infer all lower cases are correct.
Test $499.99, $500.00 and $500.01 with expected tool permissions and approval audit records.
Test only an order for $1 because it is safely below the limit.

Explanation

Correct Answer

Values around the exact threshold reveal off-by-one comparisons and unauthorized commits.

Incorrect Answers

Prose validation does not test the approval service’s decision or committed write.

A high-value sample cannot detect incorrect equality handling at $500.

A low amount cannot expose behavior at the approval boundary.

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.

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

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)
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)

Pass Your Microsoft AB-100 Exam Easy!

Question 720

Explanation

Correct Answer

Incorrect Answers

Question 721

Explanation

Correct Answer

Incorrect Answers

Question 722

Explanation

Correct Answer

Incorrect Answers

Question 723

Explanation

Correct Answer

Incorrect Answers

Question 724

Explanation

Correct Answer

Incorrect Answers

Question 725

Explanation

Correct Answer

Incorrect Answers

Question 726

Explanation

Correct Answer

Incorrect Answers

Question 727

Explanation

Correct Answer

Incorrect Answers

Question 728

Explanation

Correct Answer

Incorrect Answers

Question 729

Explanation

Correct Answer

Incorrect Answers

Question 730

Explanation

Correct Answer

Incorrect Answers

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