Microsoft AB-100 Practice Test 49

Question 480

A facility compliance agent retrieves a low-quality OCR extract with ambiguous emergency valve numbers. A field engineer asks which valve to close. What should the agent do?

Answer confidently to avoid delaying the engineer, omitting the OCR uncertainty.
Withhold a definitive valve instruction and escalate to a verified maintenance procedure or qualified operator.
Select whichever valve number appears most often in the corrupted OCR extract.
Use the nearest numerically adjacent valve code if the scanned digits are unclear.

Explanation

Correct Answer

A misread equipment identifier is safety-critical, so incomplete source evidence must not become an actionable command.

Incorrect Answers

A rapid unsupported command is not safer than an explicit evidence and verification escalation.

Repeated OCR artifacts can be wrong and cannot establish the actual tagged field equipment.

Proximity of identifiers does not prove the correct isolation point and could create a safety hazard.

Question 481

A project management team already uses Microsoft 365 Copilot in Teams. They need read-only milestones from a separate authenticated ticketing system. What is the appropriate first architecture comparison?

Copy every ticket into all employees’ prompts to avoid connector integration.
Have the model declare ticket milestones without a supported system connection.
Assess a supported permission-aware Copilot extension in Teams against the cost and control needs of a new portal.
Build a new external portal solely to display the same milestones with a different visual theme.

Explanation

Correct Answer

Retaining the existing work surface can reduce duplicated sign-in and training when the connector honors caller permissions.

Incorrect Answers

Unscoped prompt copies create access violations and stale ticket status across teams.

An unconnected model cannot establish current private project records.

A cosmetic difference does not justify duplicating identity, operations and maintenance without an unmet requirement.

Question 482

A company wants anonymous public customers to submit claims with secure identity verification and document uploads. Its staff use Microsoft 365 Copilot internally. What should guide the app choice?

Give all anonymous customers shared employee Microsoft 365 credentials.
Route every customer claim through a staff member’s private Copilot chat and omit individual customer identity.
Disable all document validation because Microsoft 365 already supports file storage.
Use an approved external customer application and identity flow rather than assume an internal employee Copilot entitlement provides a public portal.

Explanation

Correct Answer

A public claim intake has customer authentication and interaction needs distinct from a licensed staff productivity surface.

Incorrect Answers

Sharing workforce identities destroys accountability and exposes internal tenant permissions.

An employee session does not verify the customer or provide an appropriate external submission contract.

Storage availability does not establish a permitted customer upload or claims-processing authorization.

Question 483

A finance team needs to see its own approved purchase-order status in Outlook, not create new orders. An existing extension can query orders under user identity. What is the appropriate solution?

Extend the existing Copilot experience with read-only scoped order status and verify negative entitlement cases.
Create a standalone purchasing platform with unrestricted write privileges for every user.
Index all supplier contracts under one global identity so Outlook can display every order.
Train a new foundation model to memorize current order status without querying the purchase system.

Explanation

Correct Answer

The user-facing requirement fits an existing work surface and permission-aware read contract without new transaction powers.

Incorrect Answers

A read-only status request does not justify a new high-privilege service and duplicated ownership.

Global retrieval can disclose unauthorized purchasing records despite the user’s limited business role.

Order status changes over time and cannot be reliably stored in model weights for operational answers.

Question 484

An aircraft maintenance tool must classify fault codes while disconnected from the network. A Microsoft 365 Copilot extension would require a live hosted call. Which constraint should drive the decision?

Share a remote administrator credential with the field device so cloud access works without a network.
Select an approved offline-capable solution and benchmark its local accuracy and operational update process.
Choose Microsoft 365 Copilot because users are familiar with it and ignore the offline requirement.
Cache previously generated fault decisions without checking whether the equipment condition has changed.

Explanation

Correct Answer

The no-connectivity requirement is a hard architectural constraint that a cloud-only extension cannot meet.

Incorrect Answers

Credentials do not create connectivity and would add unnecessary secret-handling risk.

Interface familiarity cannot compensate for an unavailable inference dependency during field work.

Historical responses can be stale or unsafe for a newly observed failure.

Question 485

A team can add an approved employee-directory search via a supported Copilot extension or build a custom web app. The extension meets all role and freshness tests. What lifecycle factor matters?

Prefer custom code because it exposes more implementation files even when business outcomes are identical.
Make employee directory data publicly searchable to simplify both integration options.
Compare incremental connector upkeep and compatibility testing with full custom UI, identity, accessibility and support ownership.
Assume the extension has no future maintenance because the initial connector works.

Explanation

Correct Answer

An extension may avoid rebuilding solved enterprise capabilities, but its service dependencies still require maintenance checks.

Incorrect Answers

Code quantity is not evidence of functional value or lower long-term total cost.

Removing source authorization changes the confidentiality requirement rather than resolving the architecture choice.

API versions, source permissions and consent settings can change and still require an accountable owner.

Question 486

A customer support classifier repeatedly mislabels a specialist product. Logs show its reference catalog still uses a discontinued SKU code. What should the architect test before paying for fine-tuning?

Train on the obsolete SKU labels as ground truth to make the original answers more consistent.
Increase model context without reconciling the code mapping.
Evaluate only polished response tone and assume improved fluency means better SKU classification.
Correct and version the catalog mapping, then reevaluate the base model on held-out product cases.

Explanation

Correct Answer

Bad grounding can cause apparent reasoning failure; custom model training will not reliably fix a wrong source identifier.

Incorrect Answers

Reinforcing incorrect identifiers would increase alignment with stale data rather than current catalog accuracy.

More stale product records may worsen ambiguity when canonical identity is still missing.

Presentation quality does not establish correct mapping of specialist products to the current business code.

Question 487

A security-ticket classifier has 98% benign tickets and 2% severe incidents. A tuned model achieves 98% accuracy by predicting benign every time. What should the acceptance metric emphasize?

Severe-incident recall and false-positive tradeoffs on representative held-out data, not overall accuracy alone.
Approve the model because 98% accuracy exceeds most enterprise target numbers.
Remove severe incidents from the holdout set to obtain a cleaner evaluation.
Treat model-generated justifications as substitutes for correctly classified severe incidents.

Explanation

Correct Answer

Class imbalance makes a majority-class baseline deceptively strong while missing all costly severe cases.

Incorrect Answers

The score hides a complete failure on the risk class that the classifier must identify.

Removing the critical class would conceal the intended task and make the benchmark invalid.

Fluent narratives cannot compensate for false negatives on mandatory incident escalation.

Question 488

A credit-underwriting team proposes tuning a model using customer bank statements. Its vendor requests permanent retention of all source documents in an unmanaged workspace. What must be resolved?

Train immediately and document consent and data residency after deployment.
Establish lawful training use, protected storage, retention, access limits and deletion controls before data export.
Upload all statements under an administrator share link because more training data ensures fairness.
Strip only filenames while keeping all account numbers and transactions available to every trainer.

Explanation

Correct Answer

A custom model requires explicit handling of sensitive training inputs beyond ordinary inference permission.

Incorrect Answers

Retrospective paperwork cannot make an earlier unauthorized sensitive-data transfer acceptable.

Data volume does not waive confidentiality and unrestricted links create exposure beyond the approved purpose.

Removing a filename alone does not anonymize financial records or define authorized retention.

Question 489

A parts-classification team demonstrates a tuned model, but product codes are updated monthly and nobody owns retraining or rollback. What is required before selection?

Allow production data to overwrite model parameters continuously with no reviewed holdout tests.
Rely on the agent to guess novel codes from the nearest old code indefinitely.
Assign model and dataset owners with drift monitoring, refresh triggers, versioned evaluation and rollback criteria.
Treat a one-time lab accuracy score as proof the model will never become stale.

Explanation

Correct Answer

A tuned artifact is an ongoing lifecycle obligation when the underlying product taxonomy changes routinely.

Incorrect Answers

Uncontrolled training can regress accepted decisions and destroy release reproducibility.

Nearest-neighbor similarity cannot ensure correct new-category assignment without updated governed ground truth.

New codes can change the evaluated distribution and invalidate earlier generalization.

Question 490

An aircraft-inspection agent uses a general vision model. A narrow defect class is missed repeatedly on labeled turbine images despite verified prompts and image quality. What is the strongest next step?

Choose the largest language-only model without testing whether it accepts the needed images.
Train exclusively on the same images used to report the original accuracy score.
Remove the rare defect class from the test set so the general model passes.
Benchmark an appropriately governed specialist vision model or tuning approach on independent annotated defect samples.

Explanation

Correct Answer

A reproducible task-specific gap after baseline fixes can justify further modeling investment and domain validation.

Incorrect Answers

Parameter count and text generation do not establish relevant visual inspection performance.

Evaluating on training images cannot demonstrate generalization to new turbine inspections.

Excluding observed missed defects hides the safety-critical gap rather than addressing it.

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

Microsoft AB-100 Practice Test 47 - 10 Questions (Last Reviewed 28-Sep-2026)
Microsoft AB-100 Practice Test 48 - 10 Questions (Last Reviewed 27-Sep-2026)
Microsoft AB-100 Practice Test 49 - 10 Questions (Last Reviewed 26-Sep-2026)
Microsoft AB-100 Practice Test 50 - 10 Questions (Last Reviewed 25-Sep-2026)
Microsoft AB-100 Practice Test 51 - 10 Questions (Last Reviewed 24-Sep-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 480

Explanation

Correct Answer

Incorrect Answers

Question 481

Explanation

Correct Answer

Incorrect Answers

Question 482

Explanation

Correct Answer

Incorrect Answers

Question 483

Explanation

Correct Answer

Incorrect Answers

Question 484

Explanation

Correct Answer

Incorrect Answers

Question 485

Explanation

Correct Answer

Incorrect Answers

Question 486

Explanation

Correct Answer

Incorrect Answers

Question 487

Explanation

Correct Answer

Incorrect Answers

Question 488

Explanation

Correct Answer

Incorrect Answers

Question 489

Explanation

Correct Answer

Incorrect Answers

Question 490

Explanation

Correct Answer

Incorrect Answers

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