Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 18

Question 170

A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: Some cases need cross-document reasoning and detailed justification. Known defect: Small-model outputs omit critical contractual exceptions. Which KPI should be selected?

  1. High-complexity case pass rate by model route
  2. Unsupported-host requirements met by the custom design
  3. New-policy answer accuracy after ingestion
  4. Accuracy-adjusted cost per completed business task

Explanation

Correct Answer

A:

Because complexity classification selects a capable model for difficult tasks, the outcome worth following is high-complexity case pass rate by model route; the metric is tied to the reported issue—small-model outputs omit critical contractual exceptions—for the scenario where some cases need cross-document reasoning and detailed justification.

Incorrect Answers

B:

Unsupported-host requirements met by the custom design; that indicator describes a distinct workflow where a workflow needs proprietary scheduling, model choice and runtime control; it does not quantify high-complexity case pass rate by model route, which is needed for the stated case.

C:

The premise of this indicator is retrieval separates volatile knowledge from model parameters; the architecture being evaluated instead serves some cases need cross-document reasoning and detailed justification, and its success is measured by high-complexity case pass rate by model route.

D:

For the workflow where real task mix differs from prelaunch performance tests, accuracy-adjusted cost per completed business task is an actionable measure; it does not establish whether small-model outputs omit critical contractual exceptions in the situation under review.

Question 171

A design adds autonomous behavior to a process where a pilot reports gross labor savings but omits operating expenses. This creates a documented risk: Unverified ROI assumptions might be published as approved financial forecasts. Which enforced control is the best fit?

  1. Keep a known-good model and agent configuration for rollback
  2. Require finance approval of assumptions and sensitivity ranges
  3. Exclude use cases lacking a safe data-access path from early rollout
  4. Require risk sign-off from accountable stakeholders
  5. Fall back to a stronger model on low-confidence cases

Explanation

Correct Answer

B:

In the context where a pilot reports gross labor savings but omits operating expenses, the exposure is that unverified ROI assumptions might be published as approved financial forecasts; containing it requires a guardrail in the operation where net benefit compares realized outcomes with implementation and recurring costs.

Incorrect Answers

A:

Where new behavior must be evaluated without exposing the full workforce, the guard can address a new agent release could regress behavior across the entire employee base; the exposure that unverified ROI assumptions might be published as approved financial forecasts remains relevant to the present assessment.

C:

For the use case where candidate processes are attractive but permissions and data quality are weak, the control addresses the threat that early adopters might receive access to an ungoverned or unreliable data source; the specified agent must be protected against unverified ROI assumptions might be published as approved financial forecasts.

D:

The safeguard for business, platform, security and legal teams make conflicting AI decisions deals with a different threat: cross-function objections might be overruled without an accountable approval decision; even after it is enforced, the exposure that unverified ROI assumptions might be published as approved financial forecasts remains possible in the target workflow where net benefit compares realized outcomes with implementation and recurring costs.

E:

This restriction is associated with high-volume narrow extraction needs predictable latency and cost and addresses the exposure that a low-capability routed model might quietly misclassify difficult cases; it does not specifically protect the target workflow against unverified ROI assumptions might be published as approved financial forecasts.

Question 172

A staging test has to detect the specific defect that a boxed product cannot express the execution and ownership rules. In the workflow where competitive requirements depend on unique multi-model control, which exercise verifies the corrective behavior?

  1. Try prompts requesting disallowed API operations
  2. Request a high-value refund and verify approval blocks execution
  3. Prototype the distinguishing behavior under production constraints
  4. Simulate a production regression and locate the accountable response team

Explanation

Correct Answer

C:

The scenario is that competitive requirements depend on unique multi-model control; the defect to rule out is that a boxed product cannot express the execution and ownership rules; testing via prototyping the distinguishing behavior under production constraints challenges the design dependency where custom engineering trades more control for more operating responsibility.

Incorrect Answers

A:

A pass on that verification would speak to agent instructions mention many APIs but only two are permitted; it would not show that custom engineering trades more control for more operating responsibility, which is needed to avoid the observed defect a boxed product cannot express the execution and ownership rules.

B:

The verification concerning the agent may refund large amounts or deactivate accounts checks confirmation and human approval separate advice from execution authority against the conversation alone authorizes a permanent business change; its result says little about the observed issue where a boxed product cannot express the execution and ownership rules.

D:

Although the exercise might catch the pilot has no team for monitoring, incidents, or iteration, it would not directly validate the mechanism that custom engineering trades more control for more operating responsibility in the target implementation.

Question 173

An agent-assisted service does not behave as specified: The agent follows lower-trust text as an order. It is used in this situation: A retrieved email contains instructions to bypass approval. Which observable diagnostic procedure is best?

  1. Inspect tool invocations following injected content
  2. Correlate outage windows with unsupported generated claims
  3. Inspect backend denials and attempted tool arguments
  4. Compare denial logs and executed actions for attack samples

Explanation

Correct Answer

A:

Where the agent follows lower-trust text as an order, the useful evidence comes from inspecting tool invocations following injected content; its focus matches the architectural mechanism that adversarial fixtures verify separation of sources and instructions in the process where a retrieved email contains instructions to bypass approval.

Incorrect Answers

B:

That probe evaluates failure injection checks controlled degradation and abstention in a workflow where knowledge retrieval is temporarily unavailable; the reported failure concerns the path for which adversarial fixtures verify separation of sources and instructions instead.

C:

The proposed probe concerns a user asks the agent to read an unrelated executive record, where authorization evaluations verify server-side deny behavior; this incident instead requires evidence about adversarial fixtures verify separation of sources and instructions and the failure that the agent follows lower-trust text as an order.

D:

Benchmarks cover only cooperative user wording; its matching diagnostic check is reasonable for a malicious instruction triggers a privileged tool; the present investigation needs evidence for the different failure where the agent follows lower-trust text as an order.

Question 174

One enterprise rollout has 2 separately testable requirements. (1) Business use: New documents enter a production source automatically. Existing shortcoming: Unreviewed confidential drafts appear in agent answers. (2) Business use: A preprocessing release changes category labels incorrectly. Existing shortcoming: Restoring code alone does not restore previous predictions. Which 2 success metrics directly satisfy these stated cases? Choose TWO.

  1. True performance drift detected before critical SLO breach
  2. Ingested documents passing quality and access gates
  3. Retrieved chunks traceable to authoritative source versions
  4. Time needed to recover a known-good model/data pair

Explanation

Correct Answers

B:

The outcome ingested documents passing quality and access gates measures whether the workflow where new documents enter a production source automatically is improving; the scorecard separately needs time needed to recover a known-good model/data pair.

D:

The outcome time needed to recover a known-good model/data pair measures whether the workflow where a preprocessing release changes category labels incorrectly is improving; the scorecard separately needs ingested documents passing quality and access gates.

Incorrect Answers

A:

The alternative measures true performance drift detected before critical SLO breach for the situation where customer terminology changes over time; it cannot substitute for performance outcomes in the separate named cases: new documents enter a production source automatically; a preprocessing release changes category labels incorrectly.

C:

The alternative measures retrieved chunks traceable to authoritative source versions for the situation where agents begin citing an obsolete article after an ingestion job; it cannot substitute for performance outcomes in the separate named cases: new documents enter a production source automatically; a preprocessing release changes category labels incorrectly.

Question 175

A planned AI solution must support this situation: A canvas app passes customer records to a model action. Its prototype fails because hidden fields from the UI appear in the prompt payload. Which design best fits the constraint?

  1. Apply security and privacy controls to app-held AI context
  2. Extend service assistance with an approved escalation action
  3. Use a prompt action for bounded text transformation
  4. Invoke an AI-backed action from a canvas app with typed inputs

Explanation

Correct Answer

A:

Because server-side authorization and minimized inputs protect data flow, this approach corrects the reported breakdown where hidden fields from the UI appear in the prompt payload; the affected situation is that a canvas app passes customer records to a model action.

Incorrect Answers

B:

For a representative must initiate an exception workflow from the case, this other approach relies on the fact that an authorized action creates a governed workflow item from the current case; the issue here is hidden fields from the UI appear in the prompt payload; that mechanism does not establish the required server-side authorization and minimized inputs protect data flow.

C:

When an agent must turn service notes into a fixed JSON summary, prompt actions encapsulate reusable inputs and expected outputs; in the stated workflow, however, hidden fields from the UI appear in the prompt payload; the relevant architectural behavior is that server-side authorization and minimized inputs protect data flow.

D:

When inspectors submit a short description and need suggested categorization, an app action validates inputs and structured results before saving; in the stated workflow, however, hidden fields from the UI appear in the prompt payload; the relevant architectural behavior is that server-side authorization and minimized inputs protect data flow.

Question 176

A quality engineer is writing a regression case after the incident where customers repeat authentication and issue details after transfer. The underlying business scenario is: A voice bot cannot resolve a complex eligibility dispute. What should the test exercise?

  1. Ask unrelated questions and verify no supplier action is triggered
  2. Test edge cases with conflicting documents and incomplete facts
  3. Create an escalation from a test case and inspect audit records
  4. Escalate a multi-step voice conversation and check context continuity

Explanation

Correct Answer

D:

The check is targeted rather than generic: structured handoff passes allowed intent, transcript summary and case context; its pass/fail evidence applies to the workflow where a voice bot cannot resolve a complex eligibility dispute and to the prior defect that customers repeat authentication and issue details after transfer.

Incorrect Answers

A:

Although the exercise might catch it autonomously contacts a supplier without approval, it would not directly validate the mechanism that structured handoff passes allowed intent, transcript summary and case context in the target implementation.

B:

The suggested test serves procurement exceptions require comparing several policies and records, in which a simple scripted topic cannot handle legitimate variants; the present scenario instead concerns the reported problem where customers repeat authentication and issue details after transfer.

C:

A pass on that verification would speak to a representative must initiate an exception workflow from the case; it would not show that structured handoff passes allowed intent, transcript summary and case context, which is needed to avoid the observed defect customers repeat authentication and issue details after transfer.

Question 177

Governance approves the business purpose but requires control of this exposure: A scripted compliance topic might finish without every mandatory disclosure. The agent will be used when a regulated process requires fixed questions in exact sequence. Which boundary is most defensible?

  1. Prevent automatic approvals when the credit source is unavailable
  2. Reject generated fields outside the approved schema
  3. Block completion until required questions are answered
  4. Require human sign-off when regulated language is modified

Explanation

Correct Answer

C:

Avoiding the consequence where a scripted compliance topic might finish without every mandatory disclosure requires a control for a regulated process requires fixed questions in exact sequence; in this architecture, deterministic topic logic controls known prompts and branching, which is the point at which the boundary can be applied.

Incorrect Answers

A:

This restriction is associated with external credit service responds slowly during regional outages and addresses the exposure that credit-dependent approvals might proceed while the live credit source is unavailable; it does not specifically protect the target workflow against a scripted compliance topic might finish without every mandatory disclosure.

B:

The safeguard for an agent must turn service notes into a fixed JSON summary deals with a different threat: generated response keys might be trusted even when they violate the schema; even after it is enforced, the exposure that a scripted compliance topic might finish without every mandatory disclosure remains possible in the target workflow where deterministic topic logic controls known prompts and branching.

D:

The proposed control targets a model might alter mandatory legal wording in customer correspondence, while the given risk scenario is a scripted compliance topic might finish without every mandatory disclosure; these exposures arise in different operational contexts.

Question 178

A design adds autonomous behavior to a process where required knowledge stores are intermittently unreachable. This creates a documented risk: A retrieval outage could make the model invent unsupported factual claims. Which enforced control is the best fit?

  1. Restrict access through source permissions, not prompt wording
  2. Suppress assertions that require evidence when no source was retrieved
  3. Never grant tools based on instructions found in retrieved text
  4. Roll back route changes if quality guardrails regress

Explanation

Correct Answer

B:

In the scenario where required knowledge stores are intermittently unreachable, the explicit risk is that a retrieval outage could make the model invent unsupported factual claims; enforcement must follow the mechanism by which health checks and explicit no-evidence responses handle missing context safely.

Incorrect Answers

A:

For the use case where employee policies reside in governed Microsoft 365 documents, the control addresses the threat that an agent prompt could claim permission to read a private SharePoint document; the specified agent must be protected against a retrieval outage could make the model invent unsupported factual claims.

C:

The alternate boundary mitigates hostile instructions in retrieved content might be treated as tool-authorizing commands, which is relevant to third-party documents include sentences telling the model to ignore rules; it leaves the current workflow vulnerable to a retrieval outage could make the model invent unsupported factual claims.

D:

That protection is tied to the mechanism that feedback-driven thresholds rebalance accuracy, latency and spend; it does not prevent the distinct risk where a retrieval outage could make the model invent unsupported factual claims in the system under review.

Question 179

The acceptance board asks how to verify a corrected design. The setting is: Users report stale answers after a policy refresh. Last time, feedback tickets lack the evidence needed to reproduce them. Which scenario should the team execute?

  1. Try injection, invalid inputs and out-of-scope actions
  2. Reproduce the intent using a captured test trace
  3. Reproduce a reported answer from recorded non-sensitive context
  4. Use an annotated query with a known authoritative paragraph

Explanation

Correct Answer

C:

Feedback tickets lack the evidence needed to reproduce them; a release test must establish that this no longer occurs; reproducing a reported answer from recorded non-sensitive context measures the specific behavior for which trace and content-version linkage makes failures testable.

Incorrect Answers

A:

The verification concerning benchmarks cover only cooperative user wording checks negative tests expose policy and validation boundaries against a malicious instruction triggers a privileged tool; its result says little about the observed issue where feedback tickets lack the evidence needed to reproduce them.

B:

In a solution where the planner calls an order-cancel tool for status requests, reproducing the intent using a captured test trace could catch the defect that final answer text conceals the erroneous tool choice; the current release gate must challenge the failure where feedback tickets lack the evidence needed to reproduce them.

D:

For an assistant responds fluently but cites unrelated support articles, the alternate test is about whether retrieval telemetry distinguishes evidence search from answer generation; it cannot prove the outcome needed after the reported regression where feedback tickets lack the evidence needed to reproduce them.

Question 180

An investment review asks what measurable result would justify continuing an AI solution where users need to distinguish verified fields from AI suggestions. The original issue is: Staff mistake speculative recommendations for recorded facts. Which indicator should be tracked?

  1. Drafts meeting structure and source-accuracy criteria
  2. Complex exception tasks resolved with supported reasoning traces
  3. Numeric calculations matching the governed reference formula
  4. Compound requests completed with correct tool sequence
  5. User understanding of verified versus suggested content

Explanation

Correct Answer

E:

This scorecard needs an outcome related to users need to distinguish verified fields from AI suggestions; user understanding of verified versus suggested content. That indicator can be checked against the earlier symptom that staff mistake speculative recommendations for recorded facts.

Incorrect Answers

A:

A change in drafts meeting structure and source-accuracy criteria is informative for marketing specialists need tailored descriptions from existing briefings; it does not establish whether the current process achieves user understanding of verified versus suggested content.

B:

Although complex exception tasks resolved with supported reasoning traces is observable, the affected process involves users need to distinguish verified fields from AI suggestions; its acceptance outcome is user understanding of verified versus suggested content, not the alternative measurement.

C:

Although numeric calculations matching the governed reference formula is observable, the affected process involves users need to distinguish verified fields from AI suggestions; its acceptance outcome is user understanding of verified versus suggested content, not the alternative measurement.

D:

That scorecard concerns the activity in which users request a status check and an authorized follow-up action and its result of compound requests completed with correct tool sequence; the reported shortcoming here is staff mistake speculative recommendations for recorded facts.

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