Microsoft AB-100 Practice Test 17

Question 160

A prompt action returns a JSON object with valid syntax but contains an unsupported refund reason code. What validation is still required?

Use the first free-text value as an approved code.
Ignore schema version mismatches because the parser succeeded.
Assume any syntactically valid JSON is semantically authorized.
Check the enum against current business rules and reject or request correction before updating Finance.

Explanation

Correct Answer

Syntax validation alone does not prove that values belong to permitted business categories.

Incorrect Answers

Free text does not establish an approved enumerated reason.

Version changes can alter the accepted contract.

Valid JSON can still carry invalid or unauthorized values.

Question 161

Regional sales teams use different words for a Dynamics 365 “Qualified” lead. How should Copilot understand them without rewriting lead stages?

Create a different Dynamics 365 lead-stage value for each regional synonym instead of mapping each term to the existing Qualified status.
Maintain an approved business-vocabulary mapping of regional synonyms to the fixed CRM stage.
Give the model permission to alter qualification policy per language.
Rewrite the underlying CRM lead-stage values for each regional synonym instead of mapping the terms to existing approved codes.

Explanation

Correct Answer

Semantic synonyms should map consistently to a stable system-of-record status.

Incorrect Answers

Fragmenting one business stage into regional variants breaks pipeline analytics and downstream qualification rules.

Language variation does not authorize policy changes.

Stored stage values are a shared system contract; ad hoc renaming can break reporting, automation and audit consistency.

Question 162

A service Copilot generates an article recommendation for a case but has access to both public and internal-only articles. What customization prevents disclosure?

Share the same high-privilege knowledge connector with all public customers.
Trust the model to know which employee is allowed to see each article.
Filter candidate knowledge by article publication status and caller/recipient permissions before generation.
Retrieve all articles and remove the phrase “internal” from the final response.

Explanation

Correct Answer

Access-controlled source selection is the reliable way to keep internal articles private.

Incorrect Answers

Shared broad credentials bypass source entitlement boundaries.

The model does not independently enforce resource permissions.

Removing a word does not remove restricted content.

Question 163

A Dynamics 365 Sales agent must read paginated opportunity data without exposing records outside the account manager’s permission scope. Select TWO connector requirements.

Assume one HTTP 200 response always contains the complete opportunity collection.
Write every retrieved record into a public shared file for later model use.
Enforce record-level authorization with the requesting user’s permitted scope.
Honor pagination and continuation tokens until authorized results are complete.
Use the organization administrator token for all user queries to maximize results.

Explanation

Correct Answers

Scope-sensitive authorization must precede sharing data with the model.

Pagination handling is necessary for complete business calculations and list retrieval.

Incorrect Answers

A successful page request does not indicate the collection has no further pages.

Public copies defeat confidentiality and data governance.

Global admin identity can disclose opportunities outside the caller’s role.

Question 164

A Contact Center messaging agent handles a vulnerable customer, then switches the session to a human operator. What is the most important human-handoff control?

Send full unrelated caller history with each specialist transfer in an effort to maximize contextual information.
Mark the issue closed simply because a live agent has been assigned.
Transfer the authenticated case context and escalation reason without exposing unrelated conversation history.
Transfer a summary without the existing authenticated case identifier, requiring the human agent to search for the customer again.

Explanation

Correct Answer

A responsible handoff balances continuity, minimal disclosure and correct work ownership.

Incorrect Answers

The next specialist needs only the verified relevant customer case; unrelated call records violate data minimization and privacy.

Assignment is not proof of resolution.

Unlinked handoffs interrupt continuity and increase mistaken-record exposure; preserve verified case identity and relevant escalation state.

Question 165

A task agent updates an internal case and sends a customer notification. The notification provider fails. Which state is accurate?

Invent a delivered-message receipt using the agent’s summary.
The case update may have succeeded, but notification is pending; record partial state and retry safely.
Treat the entire workflow as completed simply because the update succeeded.
Recreate the case record to trigger duplicate notifications.

Explanation

Correct Answer

Reliable task automation records per-step outcomes and compensates or retries failed dependencies.

Incorrect Answers

Generated receipts cannot prove delivery.

A successful first action does not imply downstream success.

Duplicate records confuse service teams.

Question 166

A financial reporting agent runs autonomously each morning. It detects inconsistent inventory revenue but cannot reconcile the cause. Which autonomous behavior is acceptable?

Open a traceable investigation task and withhold automated publication until reconciliation.
Quietly edit source ERP ledgers to force agreement.
Publish the higher number because it makes the report look successful.
Archive or delete the disagreeing finance source records to make the next AI report internally consistent.

Explanation

Correct Answer

Autonomy should stop at its verified authority and escalate unresolved material financial discrepancies.

Incorrect Answers

Agent edits to ledgers can violate financial controls.

Optimistic selection is not reconciliation.

Eliminating contrary evidence does not reconcile the ledger; it destroys the basis for an investigation into the mismatch.

Question 167

A product knowledge bot provides read-only warranty guidance. A user asks it to update the product serial number. What should the agent do?

Grant itself database write permissions to honor the request.
Claim the serial number changed when no write tool exists.
Persist the proposed serial number only in conversation context while leaving the authenticated registration database unchanged.
Explain the authenticated device-registration process and hand off to its authorized update workflow.

Explanation

Correct Answer

A prompt-response agent should distinguish advice from actual transactional authority.

Incorrect Answers

Permissions must come from explicit tool and user authorization.

Generated claims cannot change the source record.

A prompt can preserve a proposed value but cannot change a protected registration record or provide an authoritative write receipt.

Question 168

An agent must extract key requirements from a multilingual meeting recording and create an audit-ready summary. Which processing sequence fits?

Use speech transcription with appropriate language support, then summarize with timestamped evidence and review uncertainty.
Use text extraction on an autogenerated meeting-summary PDF instead of processing the recorded audio and its time segments.
Use machine translation of prior public statements about the project rather than translating what was actually said in the meeting recording.
Summarize a calendar event’s agenda as though it records the meeting decisions, without checking the meeting audio.

Explanation

Correct Answer

Audio must be converted to reliable text and linked to recording intervals for traceability.

Incorrect Answers

A downstream summary cannot reconstruct source utterances, speaker identities or timestamped requirements missing from the original transcription.

Public statements may provide background context but do not prove that a specific participant made a particular meeting commitment.

An agenda records intended topics, not which requirements were spoken or accepted; the summary needs transcribed utterances with timestamps.

Question 169

A generative canvas page drafts approval guidance from policy data. Which constraint should prevent hidden changes to source records?

Automatically execute any command embedded in the generated page.
Keep generated output read-only until an authenticated explicit user action invokes a validated update service.
Generate an app page that submits the draft update in the background as soon as the user opens its review panel.
Treat a formatted page as consent to update every related record.

Explanation

Correct Answer

User intent and record mutation must be confirmed through trusted application controls.

Incorrect Answers

Generated code is not trusted execution authority.

Opening a review screen does not establish informed permission to mutate source records; consequential writes need an explicit authenticated action.

Presentation alone does not establish transactional approval.

Question 170

A Copilot Studio support topic gets two equally plausible intent matches. Which fallback behavior should be used?

Execute both tool actions and keep whichever returns first.
Ask the user to choose the intended topic or escalate without executing either risky action.
Randomly select one action because the model must always proceed.
Execute the first matching intent by topic registration order even when two high-impact actions score equally.

Explanation

Correct Answer

A clarification step prevents ambiguity from causing unintended service changes.

Incorrect Answers

Parallel side effects can produce multiple transactions.

Random selection can trigger the wrong workflow.

Topic ordering is not user confirmation and can run the wrong transaction; clarification must precede side effects.

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

Microsoft AB-100 Practice Test 15 - 10 Questions (Last Reviewed 30-Oct-2026)
Microsoft AB-100 Practice Test 16 - 10 Questions (Last Reviewed 29-Oct-2026)
Microsoft AB-100 Practice Test 17 - 10 Questions (Last Reviewed 28-Oct-2026)
Microsoft AB-100 Practice Test 18 - 10 Questions (Last Reviewed 27-Oct-2026)
Microsoft AB-100 Practice Test 19 - 10 Questions (Last Reviewed 26-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 160

Explanation

Correct Answer

Incorrect Answers

Question 161

Explanation

Correct Answer

Incorrect Answers

Question 162

Explanation

Correct Answer

Incorrect Answers

Question 163

Explanation

Correct Answers

Incorrect Answers

Question 164

Explanation

Correct Answer

Incorrect Answers

Question 165

Explanation

Correct Answer

Incorrect Answers

Question 166

Explanation

Correct Answer

Incorrect Answers

Question 167

Explanation

Correct Answer

Incorrect Answers

Question 168

Explanation

Correct Answer

Incorrect Answers

Question 169

Explanation

Correct Answer

Incorrect Answers

Question 170

Explanation

Correct Answer

Incorrect Answers

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