Microsoft AB-100 Practice Test 13

Question 120

A business app must read invoice totals from scanned PDFs before a downstream agent checks them. Which Microsoft AI capability category should be considered?

A generic model instructed to invent totals when scans are unreadable.
A document extraction service designed to return fields and supporting document evidence.
Run generic OCR text extraction without preserving table and invoice-field structure needed to validate totals and taxes.
Classify the document’s overall intent from extracted text without extracting the authoritative amount, tax and supplier fields.

Explanation

Correct Answer

Document-specific extraction should expose fields and uncertainty for validation before business actions.

Incorrect Answers

Invented totals cannot support financial reconciliation.

OCR text alone may omit invoice field relationships, confidence and source spans needed for downstream reconciliation.

An invoice type label does not supply structured amounts or document evidence required for financial checks.

Question 121

A Power Apps canvas app uses a generated page to summarize cases. The designer proposes executing arbitrary JavaScript returned by an LLM. What is the safer design?

Give the generated page direct table-wide Dataverse access instead of using scoped queries to fetch approved case details.
Skip field validation because the model generated the schema.
Render validated typed data through trusted application components and keep executable actions governed.
Evaluate every model response as privileged app code.

Explanation

Correct Answer

Generated content should not silently become trusted executable code or cross-user authority.

Incorrect Answers

Unrestricted table access can expose payroll or unrelated customer records that the signed-in user is not authorized to inspect.

Generated schemas still need contract validation.

Untrusted code execution creates injection and permission risk.

Question 122

A Copilot Studio assistant has a billing topic, but an unusual question falls outside every configured intent. What should the fallback topic do?

Discard all prior user context and claim the issue is resolved.
Select the most frequent billing-topic response when the user’s issue is outside its supported topics, without clarification or handoff.
Execute a refund tool to signal the conversation succeeded.
Offer a controlled clarification or handoff while preserving user context and avoiding invented policy.

Explanation

Correct Answer

Fallback routes ambiguous requests safely when no specific topic provides sufficient support.

Incorrect Answers

False completion hides unresolved customer needs.

The most frequent answer may be unrelated and can falsely reassure the customer that an unresolved case was completed.

A transactional action is unrelated to intent resolution.

Question 123

An internal agent ingests PDF policies with repeated headers and outdated revisions. What preprocessing improves its grounding index?

Remove all source labels so chunks look uniform.
Index every PDF page as one unstructured blob regardless of layout.
Extract meaningful sections with source/revision metadata, deduplicate boilerplate and exclude obsolete policy versions.
Replace unreadable clauses with model-generated text during ingestion.

Explanation

Correct Answer

Useful retrieval needs clean chunks, version context and verifiable source references.

Incorrect Answers

Missing labels prevent provenance checks.

Unstructured bulk pages can lower precision and citation quality.

Synthetic replacement text is not authenticated policy evidence.

Question 124

A Power Apps canvas app helps sales agents draft quotes. Users must inspect and approve the proposed discount before it becomes an order. Which design is appropriate?

Save the AI-proposed discount to a staging Dataverse field as an approved quote before the salesperson reviews it.
Let the canvas app automatically commit the proposed discount when the model confidence exceeds the configured threshold.
Give every viewer administrator access to the order Dataverse table.
Display a reviewable AI suggestion, then use the authenticated app action to validate and save the approved quote.

Explanation

Correct Answer

A canvas app can embed assistance while keeping the real transaction behind explicit validated user intent.

Incorrect Answers

A staging field may safely hold a draft, but marking it approved or triggering an order before verified user consent crosses the transaction boundary.

Confidence can rank a suggestion but cannot express a salesperson’s approved commercial decision or satisfy server-side discount authority.

Broad Dataverse rights exceed a quote reviewer’s role.

Question 125

A Power Platform workload requires predictable service during a regional connector outage. Which Well-Architected design concern should dominate?

Reliability: bounded retries, fallback paths, continuity expectations and recovery tests.
Experience optimization alone, with no service recovery plan.
Increase the generative model capacity to lower response times without designing a connector-outage fallback or testing recovery.
Use interface load animations as the sole proof of reliability while ignoring connector failure rates and recovery behavior.

Explanation

Correct Answer

The reliability pillar addresses availability and recovery under failure.

Incorrect Answers

Usability does not restore a failed dependency.

Model throughput is a performance consideration, not a remedy for an unavailable connector; continuity needs bounded retries and a defined degraded path.

A polished loading state does not establish availability, recovery time or correct business operations when a dependency fails.

Question 126

An organization has two business-language tasks: routing utterances to one of five fixed support categories, and composing a grounded answer that synthesizes several policy paragraphs. Select TWO suitable design choices.

Use unrestricted free-form generation to create a new routing category for each support utterance, instead of enforcing one of the five defined intents.
Use only a fixed-intent classification label to compose the policy answer, without retrieving the paragraphs that support its details.
Use a trained language-understanding or classification approach for the fixed intent labels.
Use a fixed keyword list as proof that an answer is grounded in multiple current policies.
Use generative orchestration with authorized grounding for the open-ended policy synthesis.

Explanation

Correct Answers

A bounded intent set can be recognized with trained language classification and evaluated against labeled utterances.

Synthesizing grounded narrative across documents is a legitimate generative-agent use with source checks.

Incorrect Answers

Open-ended category invention breaks a closed-label routing contract and prevents reliable confusion-matrix evaluation against the five categories.

A category label can route a request but does not contain the policy evidence needed to support an open-ended synthesized answer.

Keyword presence does not prove the cited passages support the answer.

Question 127

A Copilot Studio flow submits a stock reservation, but the next carrier-booking call times out. What should the workflow design include?

An idempotent reservation key, status reconciliation and compensating release or operator escalation.
Erase intermediate state so the agent can start from memory.
Blindly retry both writes with new IDs until a success message appears.
Declare completion based on the first successful reservation response.

Explanation

Correct Answer

Multi-step agent flows need durable state and a way to handle partial failure without duplicated commitment.

Incorrect Answers

Lost state makes safe recovery and audit impossible.

New IDs can create duplicate reservations and bookings.

One step success does not prove the whole workflow finished.

Question 128

A Copilot Studio prompt action produces a proposed support priority. The downstream automation accepts only P1, P2 or P3. What output contract should be used?

Any free-form adjective because the workflow will infer priority.
An unconstrained value that the agent may rename on each run.
A validated structured priority enumeration with missing/invalid-value handling.
A model-generated database command with no schema review.

Explanation

Correct Answer

Typed constrained outputs keep prompt assistance interoperable with deterministic downstream logic.

Incorrect Answers

Ambiguous adjectives cannot reliably map to accepted priority codes.

Unstable values break workflow rules and reporting.

Executable generated commands pose unnecessary risk.

Question 129

A customer service team uses the phrase “active policy” but the CRM field status=Open also includes archived items. How should Copilot business terms be corrected?

Define an approved semantic filter that excludes archived records and test representative user queries.
Search all statuses and ask the model not to mention archived records afterward.
Rename every archived item as active to match the phrase.
Use response confidence as the only policy-status filter.

Explanation

Correct Answer

Business terminology needs a verifiable mapping to actual lifecycle fields, not casual string matching.

Incorrect Answers

Late hiding cannot prevent stale grounding or citation.

Changing source data to fit vocabulary corrupts records.

Confidence does not define policy eligibility.

Question 130

A service Copilot customization should suggest case replies but supervisors must review all customer-facing legal assurances. Which customization boundary fits?

Publish generated legal assurances automatically whenever the response is polite.
Generate a draft with cited approved knowledge and require supervisor release before sending.
Bypass the CRM case permissions because the draft has not yet been sent.
Use an unrestricted system prompt to claim all contracts are identical.

Explanation

Correct Answer

Draft generation can be separated from the authenticated release action for consequential communications.

Incorrect Answers

Tone does not establish legal or contractual correctness.

Draft content may still expose restricted customer data.

Unsupported universal claims violate case-specific evidence.

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

Microsoft AB-100 Practice Test 11 - 10 Questions (Last Reviewed 03-Nov-2026)
Microsoft AB-100 Practice Test 12 - 10 Questions (Last Reviewed 02-Nov-2026)
Microsoft AB-100 Practice Test 13 - 10 Questions (Last Reviewed 01-Nov-2026)
Microsoft AB-100 Practice Test 14 - 10 Questions (Last Reviewed 31-Oct-2026)
Microsoft AB-100 Practice Test 15 - 10 Questions (Last Reviewed 30-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 120

Explanation

Correct Answer

Incorrect Answers

Question 121

Explanation

Correct Answer

Incorrect Answers

Question 122

Explanation

Correct Answer

Incorrect Answers

Question 123

Explanation

Correct Answer

Incorrect Answers

Question 124

Explanation

Correct Answer

Incorrect Answers

Question 125

Explanation

Correct Answer

Incorrect Answers

Question 126

Explanation

Correct Answers

Incorrect Answers

Question 127

Explanation

Correct Answer

Incorrect Answers

Question 128

Explanation

Correct Answer

Incorrect Answers

Question 129

Explanation

Correct Answer

Incorrect Answers

Question 130

Explanation

Correct Answer

Incorrect Answers

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