Microsoft AB-100 Practice Test 83

Question 820

A new invoice agent release introduces a payment-approval action. The rollout team copied a broad test role into production. What must change before promotion?

Ask the model whether the user sounds like a manager.
Use approved finance role mappings and test permitted and denied approval actions with production-like identities.
Allow every Finance viewer to approve because the action is inside the Finance app.
Hide the approval button but leave the endpoint accessible under a global token.

Explanation

Correct Answer

New business capabilities need explicit user-role authorization, not inherited development convenience.

Incorrect Answers

Conversation style cannot authenticate an approver’s delegated role.

Read permission does not imply transaction signoff authority.

Backend access remains exploitable without service-side checks.

Question 821

A release adds a new vendor risk classification used by both Finance and Supply Chain agents. After cutover, one consumer rejects the new enum value. What rollback plan is sound?

Remove the supplier master entirely until the next release.
Let each agent interpret the enum in its own prompt.
Retain a compatible prior schema/version and apply a coordinated data/connector rollback or compatibility adapter.
Change all unsupported values to “approved” so transactions continue.

Explanation

Correct Answer

A shared field must preserve consumer compatibility to prevent blocking purchasing and payable workflows.

Incorrect Answers

Destructive removal can disrupt unrelated operations and loses source evidence.

Undocumented per-agent meaning leads to inconsistent risk decisions.

Silent fallback can authorize risky suppliers and break governance.

Question 822

A new Dynamics 365 service workflow introduces `Pending Customer` status, but its copilot still treats any non-open case as resolved. What must release planning include?

Keep the old binary status mapping because most cases are open or closed.
Ask the model to choose a status from sentiment in support notes.
Automatically close Pending Customer cases after generating a response.
Update and test status mappings and response rules across the copilot and the service application before activation.

Explanation

Correct Answer

A new operational state can materially change customer messaging and downstream actions.

Incorrect Answers

The new pending state must not be falsely reported as resolved.

Sentiment cannot replace the authoritative case-state enum.

Closing unresolved cases undermines customer service control and may discard needed replies.

Question 823

A new Sales copilot can draft and send renewal emails. The old feature only drafted them. What deployment guardrail is required?

Keep send permissions separately gated and require tested recipient/consent validation before enabling the new action.
Enable automatic sending for all sellers because drafting tests passed.
Send test messages to actual customers without review.
Assume email address autocomplete is sufficient identity verification.

Explanation

Correct Answer

Changing from drafting to external communication creates a new high-impact side effect.

Incorrect Answers

Draft quality does not establish correct recipients or permission to contact them.

Production outreach is not an appropriate substitute for authorized sandbox testing.

Autocomplete can select similarly named or outdated customer contacts.

Question 824

A service chatbot release changes escalation categories but the human support queue still accepts the old codes. Cases disappear after handoff. What ALM coordination is needed?

Ask the model to invent a new queue name from the customer’s issue.
Update both the agent handoff schema and queue routing rules, testing accepted/rejected categories end to end.
Retry the same unknown category indefinitely.
Mark escalations successful when the agent emits a routing phrase.

Explanation

Correct Answer

Topic naming changes can become operational interface breaks when queues depend on exact codes.

Incorrect Answers

Unrecognized identifiers can drop cases and break accountability.

Unsupported queue codes will not become valid merely through repetition.

Language does not prove the staffed service queue accepted the case.

Question 825

A sales AI solution is promoted with copied marketing contacts from a test dataset. The new production environment has different opt-out flags. What must release validation do?

Allow the model to infer consent from friendly conversation history.
Disable opt-out filtering until the first campaign completes.
Rebind to authoritative production consent data and verify outreach is blocked for opted-out records.
Assume test consent flags remain valid for every production contact.

Explanation

Correct Answer

Environment promotion must not reuse stale or synthetic consent values for real customer communication.

Incorrect Answers

Tone and prior interaction do not establish current marketing permissions.

Post-contact remediation cannot undo an unauthorized message.

Test records and production legal preferences are distinct datasets.

Question 826

A new customer-service agent can suggest and execute refunds. Which TWO checks must precede production release? (Select TWO.)

Test caller-level and order-level refund authorization including denial paths.
Validate committed refund amount and duplicate-retry behavior against Finance receipts.
Approve release when the agent demonstrates friendly conversation style.
Give the agent one unrestricted refund administrator token.
Count successful draft messages as verified refund completions.

Explanation

Correct Answers

A customer must not cause writes against orders beyond their entitlements.

Transaction correctness requires reconciliation and idempotency, not just an action success phrase.

Incorrect Answers

Style is not evidence of correct financial writes or backend permissions.

Broad credentials bypass the required requester authorization boundary.

A generated proposal is not a confirmed payment transaction.

Question 827

A new opportunity-scoring model changes risk categories from Low/Medium/High to numerical bands, but dashboards still expect strings. What should the release package include?

A versioned score-to-category mapping or coordinated dashboard contract update tested on historical and new examples.
Let the dashboards display raw numbers under the old text headers.
Ask the agent to invent narrative categories after the score is committed.
Disable data lineage so old and new scores appear identical.

Explanation

Correct Answer

Consumers must interpret the same score semantics across model and reporting releases.

Incorrect Answers

Mismatched labels can mislead users and invalidate existing thresholds.

Unstructured inference cannot guarantee consistent reporting definitions.

Removing version evidence conceals change and prevents accurate comparisons.

Question 828

A deployed service bot version starts duplicating return labels, but hundreds of sessions are in flight. What is the safest rollback approach?

Treat the first label-related chat message as proof a label was issued.
Pin new sessions to the known-good version and reconcile pending return-label requests with durable IDs.
Delete all session IDs and force users to restart label creation.
Leave the defective version active until sessions naturally end.

Explanation

Correct Answer

Rollback must avoid reissuing transactions and preserve customer case continuity.

Incorrect Answers

Conversational intention does not establish downstream label creation.

Restarting without transaction state can create more duplicate labels.

Known duplicate-output behavior should be contained and repaired.

Question 829

A warranty agent release reduces response time but increases incorrect warranty exclusions. What should trigger rollback?

Only a rise in average token count.
One positive executive demo of the faster response.
A breached factual/policy error threshold on representative verified warranty decisions.
Only an outage of the chat UI.

Explanation

Correct Answer

Speed improvements cannot justify materially incorrect policy outcomes.

Incorrect Answers

Inference consumption is not the same as customer harm.

A favorable demonstration cannot negate verified factual regressions.

Release health includes content and business correctness, not just availability.

Question 830

A new agent build binds to a payment processor. What production smoke test is appropriate before enabling real customer transactions?

Create a real $1,000 payment and refund it later.
Assume the connection works because its credential field is populated.
Ask the model to name the processor from its prompt.
Exercise a processor-supported non-mutating health/status or dedicated sandbox transaction path with verified endpoint identity.

Explanation

Correct Answer

Connectivity checks should not risk an unauthorized real financial posting.

Incorrect Answers

An avoidable live charge is not a safe verification prerequisite.

Secret presence does not prove network reachability or permitted actions.

Generated text cannot validate the bound endpoint and transaction contract.

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

Microsoft AB-100 Practice Test 81 - 10 Questions (Last Reviewed 21-Oct-2026)
Microsoft AB-100 Practice Test 82 - 10 Questions (Last Reviewed 20-Oct-2026)
Microsoft AB-100 Practice Test 83 - 10 Questions (Last Reviewed 19-Oct-2026)
Microsoft AB-100 Practice Test 84 - 10 Questions (Last Reviewed 18-Oct-2026)
Microsoft AB-100 Practice Test 85 - 10 Questions (Last Reviewed 17-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 820

Explanation

Correct Answer

Incorrect Answers

Question 821

Explanation

Correct Answer

Incorrect Answers

Question 822

Explanation

Correct Answer

Incorrect Answers

Question 823

Explanation

Correct Answer

Incorrect Answers

Question 824

Explanation

Correct Answer

Incorrect Answers

Question 825

Explanation

Correct Answer

Incorrect Answers

Question 826

Explanation

Correct Answers

Incorrect Answers

Question 827

Explanation

Correct Answer

Incorrect Answers

Question 828

Explanation

Correct Answer

Incorrect Answers

Question 829

Explanation

Correct Answer

Incorrect Answers

Question 830

Explanation

Correct Answer

Incorrect Answers

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