Microsoft AB-100 Practice Test 2

Question 10

An AI program combines HR and Finance datasets. The data team cannot decide who approves corrections to shared employee-cost fields. Which organizational control is missing?

A prompt telling users to edit source data when an answer seems unexpected.
A nightly export into an unmanaged spreadsheet accessible to every team.
Assume that using a stronger embedding model eliminates the need to assign a data steward for shared HR and Finance records.
A named data steward and escalation workflow for ownership, quality and authorized updates.

Explanation

Correct Answer

Clear stewardship assigns accountable owners who can resolve conflicting records and approve changes at the system of record.

Incorrect Answers

Ad hoc end-user edits risk corrupting shared records and bypassing review.

An unmanaged copy increases access and reconciliation problems instead of defining ownership.

Better retrieval does not provide organizational authority over quality corrections, access permissions or lawful use of employee data.

Question 11

A contact-center process has fixed identity checks but callers often explain complex exceptions in free text. Which agent strategy best separates responsibilities?

Replace mandatory identity checks with sentiment analysis of caller frustration.
Use the agent’s fluent exception summary as evidence that the customer passed the mandatory identity and refund approval checks.
Use fixed authentication and policy actions, then let an agent summarize the exception for an authorized specialist.
Let the agent infer caller identity from conversational confidence before exposing account data.

Explanation

Correct Answer

Structured controls remain deterministic while language understanding helps staff process ambiguous narrative evidence safely.

Incorrect Answers

Sentiment is unrelated to the required account authentication boundary.

A well-formed summary can still lack verified caller identity and policy eligibility; enforce those controls at the transactional boundary.

Natural-language confidence does not verify identity or permissions.

Question 12

Before moving a warehouse maintenance assistant into production, the sponsor asks how to judge its pilot beyond a successful demo. Which evidence is most defensible?

Use the proportion of sessions where the agent declares a task successful, without reconciling the underlying business transaction.
Require zero unanswered questions by instructing the model never to abstain.
Treat one favorable executive conversation as proof of enterprise reliability.
Measure task completion, factual-error rate and human-escalation outcomes against a documented manual baseline.

Explanation

Correct Answer

A baseline and test cases reveal whether the system improves real workflows without hiding incorrect answers or costly escalations.

Incorrect Answers

Self-reported completion is not a trustworthy outcome metric; it can be positive when a downstream write failed.

Forcing guesses makes apparent coverage rise while harming factual trustworthiness.

One demo cannot establish reliability across ordinary and failure cases.

Question 13

A travel-expense team has deterministic rate and eligibility checks, but a minority of receipts have ambiguous free-text descriptions. Running a large model on every claim would exceed the project budget. Which scope decision keeps expense policy enforceable and avoids unnecessary inference cost?

Run the rules engine for ordinary claims and invoke a grounded narrative-analysis agent only for uncertain receipts, escalating policy exceptions for review.
Run every claim through a large agent before applying rate rules, even when the structured fields already determine eligibility.
Turn off deterministic eligibility checks and let the model calculate rates in one inference call to reduce integration work.
Process uncertain receipts with no human review because skipping escalation reduces the average cost per claim.

Explanation

Correct Answer

Conditional agent invocation uses language interpretation where ambiguity exists, while fixed rates and binding approvals remain outside the model. This reduces variable inference expense without weakening policy.

Incorrect Answers

Always invoking a large model adds latency and token cost to claims whose eligibility is already determined by structured rules; there is no demonstrated added value.

A single prompt can appear cheaper to integrate but does not enforce the statutory rate and category checks or provide a reliable audit trail.

Avoiding review may reduce immediate handling cost but lets unsupported reimbursements commit; exception risk and rework belong in the cost decision.

Question 14

A retailer proposes an analytics assistant that recommends staffing by store. Branch managers distrust its forecasts because holiday closures are absent from the data. What improvement is primary?

Include closure calendars and evaluate forecasts per store and time period before optimization.
Train only on the busiest store and apply that result to every location.
Rank the staffing proposals by manager satisfaction with the narrative without comparing forecasted and observed labor requirements.
Grant all managers the ability to change historical transactions to fit the forecast.

Explanation

Correct Answer

Operational covariates and segment-specific validation prevent systematic errors from misleading staffing decisions.

Incorrect Answers

A single store can have atypical demand and closure patterns.

User sentiment can complement adoption metrics, but cannot replace validation of operational staffing decisions and demand forecasting.

Rewriting history for model performance corrupts the source of truth.

Question 15

A procurement agent drafts contracts and can call a payment tool. Two controls are required before a high-value invoice can be paid. Select TWO.

Require a separate approved human signoff for invoices above the policy threshold.
Give the model a permanent payment administrator token so all requests execute.
Validate payment authorization in the transactional service against the caller’s verified role.
Treat the agent’s generated rationale as equivalent to the approver’s authenticated signature.
Accept the payment whenever an invoice PDF contains the phrase “pre-approved”.

Explanation

Correct Answers

An independent human approval must be checked before the transaction is committed.

Back-end authorization prevents a model or prompt from bypassing the user’s permitted transaction scope.

Incorrect Answers

Standing privileged credentials destroy least privilege and increase misuse or injection impact.

Generated text is not authenticated delegated authority from an actual approver.

The invoice itself is untrusted input and cannot grant a payment permission.

Question 16

A CRM dataset contains customer opt-out flags that are occasionally omitted by one export job. The AI marketing assistant must not contact customers who opted out. What correctness gate is necessary?

Reconcile consent flags with the authoritative customer record and fail closed when consent is unknown.
Ask the generator to make a marketing-friendly guess about opt-out status.
Use the newest CSV regardless of whether it includes the consent column.
Assume missing consent means permission to contact until a user complains.

Explanation

Correct Answer

Consent must be explicitly established from a source of record; incomplete exports require blocking unsafe outreach.

Incorrect Answers

Generative language cannot establish customer legal preferences.

File recency does not compensate for omitted required fields.

Absence of consent evidence is not affirmative permission.

Question 17

An inventory copilot must recommend purchases from a nightly stock snapshot. Several fast-moving items have same-day stock changes. How should timeliness be designed?

Hide retrieval timestamps from users to avoid concern about incomplete updates.
Treat the nightly snapshot as real-time because it uses a familiar schema.
Ask the model to estimate stock from how confident its explanation sounds.
Read authoritative inventory updates or flag the snapshot timestamp and stop automatic purchasing when freshness is insufficient.

Explanation

Correct Answer

A freshness policy ties high-impact purchase decisions to actual stock state rather than stale contextual grounding.

Incorrect Answers

Suppressing timestamps removes evidence of an important decision limitation.

Schema correctness does not establish current quantity.

Model confidence is not a substitute for a verified inventory event stream.

Question 18

An HR knowledge base has current policies but a corrupted identity mapping causes employee salaries to join the wrong staff ID. What must be fixed before grounding?

Join HR records by matching normalized employee names when the authoritative employee ID is unavailable.
Correct the identity-key join and verify record-to-person matches against the HR system of record.
Apply a friendlier conversational prompt without reconciling the data mapping.
Increase embedding dimensions to force the model to select the intended employee.

Explanation

Correct Answer

Correct identifiers and join lineage are prerequisites for truthful personalized AI answers, especially for sensitive HR data.

Incorrect Answers

Different employees can share names, and one person can have aliases. A name-based join can expose the wrong personnel record.

Tone does not fix false row associations or privacy exposure.

Vector dimensions cannot repair incorrect relational keys.

Question 19

Three business units want a common product catalog for their agents, but each expects different optional fields. Which architecture supports controlled data reuse?

Require all services to share one broad administrator identity to access every field.
Duplicate each catalog independently so a correction can be made in only one unit.
Publish a versioned shared schema with documented optional extensions and consumer compatibility testing.
Let every agent use undocumented field meanings and reconcile discrepancies in its prompt.

Explanation

Correct Answer

A stable contract and explicit extension governance let consumers evolve without silently changing common product semantics.

Incorrect Answers

Shared privileged access violates least-privilege boundaries and makes auditing difficult.

Independent copies drift and hinder authoritative updates.

Unspecified field meaning creates inconsistent answers and fragile integration.

Question 20

A shared department-policy index still returns restricted HR documents to staff after they transfer to another department. The chat user interface shows their new department correctly, but the indexed results are cached under a privileged service credential. What must be corrected to stop this specific leak?

Hide the restricted search hits after model generation using the department displayed by the browser.
Keep the privileged index results but instruct the model to avoid citing policies from the employee’s former department.
Enforce current employee ACLs before retrieval reaches the model, invalidate stale cached authorizations, and run role-change denial tests.
Give every employee the same service index credential so results are consistent across departments.

Explanation

Correct Answer

The defect is an outdated authorization boundary at query time. Revoking cached rights and testing the transferred employee ensures retired entitlements are actually denied.

Incorrect Answers

UI-level filtering is too late: privileged cached passages could already have entered the model context, logs or generated answer.

A prompt cannot revoke an employee’s former department rights or prevent sensitive passages from being retrieved under an overprivileged identity.

A shared privileged credential reproduces the root cause by bypassing the requesting user’s current department membership.

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

Microsoft AB-100 Practice Test 1 - 10 Questions (Last Reviewed 13-Nov-2026)
Microsoft AB-100 Practice Test 2 - 10 Questions (Last Reviewed 12-Nov-2026)
Microsoft AB-100 Practice Test 3 - 10 Questions (Last Reviewed 11-Nov-2026)
Microsoft AB-100 Practice Test 4 - 10 Questions (Last Reviewed 10-Nov-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 10

Explanation

Correct Answer

Incorrect Answers

Question 11

Explanation

Correct Answer

Incorrect Answers

Question 12

Explanation

Correct Answer

Incorrect Answers

Question 13

Explanation

Correct Answer

Incorrect Answers

Question 14

Explanation

Correct Answer

Incorrect Answers

Question 15

Explanation

Correct Answers

Incorrect Answers

Question 16

Explanation

Correct Answer

Incorrect Answers

Question 17

Explanation

Correct Answer

Incorrect Answers

Question 18

Explanation

Correct Answer

Incorrect Answers

Question 19

Explanation

Correct Answer

Incorrect Answers

Question 20

Explanation

Correct Answer

Incorrect Answers

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