Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 23

Question 220

Before production promotion, the team needs evidence for this setting: The system cannot always find evidence for regulated guidance. A prior regression caused the following: An agent guesses a policy instead of escalating. Which validation is most persuasive?

  1. Submit edge cases and validate response shape
  2. Remove required sources and check that the agent declines
  3. Run multi-hop evaluation cases for the strong-model route
  4. Attempt low-risk and high-risk actions and inspect approval enforcement

Explanation

Correct Answer

B:

The scenario is that the system cannot always find evidence for regulated guidance; the defect to rule out is that an agent guesses a policy instead of escalating; testing via removing required sources and checking that the agent declines challenges the design dependency where confidence and evidence checks route insufficiently supported cases to humans.

Incorrect Answers

A:

For agents return inconsistent fields for downstream integrations, the alternate test is about whether clear instructions and structured outputs reduce avoidable ambiguity; it cannot prove the outcome needed after the reported regression where an agent guesses a policy instead of escalating.

C:

The tested behavior is that complexity classification selects a capable model for difficult tasks; the necessary evidence here concerns confidence and evidence checks route insufficiently supported cases to humans and the previously observed an agent guesses a policy instead of escalating.

D:

A pass on that verification would speak to teams propose that an agent perform both low-risk lookups and financial postings; it would not show that confidence and evidence checks route insufficiently supported cases to humans, which is needed to avoid the observed defect an agent guesses a policy instead of escalating.

Question 221

A proposed deployment supports this business case: A support agent changes how it classifies priority intents. Reviewers also warn that a new support agent might route urgent cases to a general queue. Which restriction mitigates that concrete risk?

  1. Enforce approval before production import and publication
  2. Block release when production variables or connections are unresolved
  3. Require a tested human escalation path for unmatched cases
  4. Prevent duplicate case creation on repeated event delivery

Explanation

Correct Answer

C:

When a new support agent might route urgent cases to a general queue, the use case a support agent changes how it classifies priority intents requires a guardrail aligned with the behavior that controlled service evaluations validate routing before rollout.

Incorrect Answers

A:

That protection is tied to the mechanism that solution-based ALM transports supported components under controlled releases; it does not prevent the distinct risk where a new support agent might route urgent cases to a general queue in the system under review.

B:

This restriction is associated with a flow moved to production still points to a test endpoint and addresses the exposure that imported actions might retain staging endpoints after production deployment; it does not specifically protect the target workflow against a new support agent might route urgent cases to a general queue.

D:

A won opportunity should create a properly owned onboarding case; its guardrail addresses the possibility that duplicate Dynamics events might create redundant onboarding cases; the present architecture must constrain the different threat that a new support agent might route urgent cases to a general queue.

Question 222

During integration testing, a network fault leaves the two apps inconsistent. The associated business context is: A blocked invoice triggers a hold on related procurement actions. Which architectural remedy is the strongest match?

  1. Test Finance-to-Supply-Chain exception handling after partial failure
  2. Define end-to-end correlation across agent and business APIs
  3. Test resilience when a grounding dependency fails
  4. Test Sales-to-Service customer context and case handoff

Explanation

Correct Answer

A:

The defect is that a network fault leaves the two apps inconsistent; the expected business case is that a blocked invoice triggers a hold on related procurement actions; the design fits because failure-injection tests verify retries and compensation between systems.

Incorrect Answers

B:

For a request crosses Copilot Studio, a flow and the CRM backend, this other approach relies on the fact that correlation identifiers link the distributed operational path; the issue here is a network fault leaves the two apps inconsistent; that mechanism does not establish the required failure-injection tests verify retries and compensation between systems.

C:

The agent invents answers rather than saying evidence is missing; the alternate design can address that situation because failure injection checks controlled degradation and abstention, whereas this question needs a solution to a network fault leaves the two apps inconsistent.

D:

For a won opportunity should create a properly owned onboarding case, this other approach relies on the fact that end-to-end tests verify identity, mapping and assignment across apps; the issue here is a network fault leaves the two apps inconsistent; that mechanism does not establish the required failure-injection tests verify retries and compensation between systems.

Question 223

Release reviewers will not accept a generic success message. The business scenario is that a preprocessing release changes category labels incorrectly, and the historical failure is: Restoring code alone does not restore previous predictions. Which test is appropriately targeted?

  1. Reconstruct one released model's source and threshold decisions
  2. Roll back a staged faulty transformation and compare output
  3. Introduce a known distribution shift and verify review triggers
  4. Rebuild a staged agent from recorded versions and run smoke tests

Explanation

Correct Answer

B:

For the required workflow where a preprocessing release changes category labels incorrectly, the test needs to reproduce or prevent the failure where restoring code alone does not restore previous predictions; rolling back a staged faulty transformation and comparing output exercises the path where versioned data transformations and model artifacts enable consistent rollback.

Incorrect Answers

A:

Although the exercise might catch models move to production without an audit of training inputs, it would not directly validate the mechanism that versioned data transformations and model artifacts enable consistent rollback in the target implementation.

C:

Introducing a known distribution shift and verifying review triggers verifies an adjacent objective where customer terminology changes over time; it leaves untested the causal path in which versioned data transformations and model artifacts enable consistent rollback for this case.

D:

This check exercises a different system behavior: a complete agent release manifest pins related runtime artifacts; a pass there cannot establish that restoring code alone does not restore previous predictions no longer occurs in the required workflow.

Question 224

An agent-assisted service does not behave as specified: Team retrains monthly despite unchanged model behavior. It is used in this situation: Customer terminology changes over time. Which observable diagnostic procedure is best?

  1. Reconcile active model versions with promotion events
  2. Compare input distributions and labeled error trends
  3. Compare embedding versions and chunking metadata
  4. Inspect solution dependencies and common entity schema

Explanation

Correct Answer

B:

The investigation directly tests the causal connection: drift monitoring and outcomes justify retraining when performance shifts; it collects evidence for a workflow where customer terminology changes over time but the outcome is instead that team retrains monthly despite unchanged model behavior.

Incorrect Answers

A:

The evidence this method produces applies to a workflow where a model is replaced without a traceable change ticket; it cannot by itself establish why team retrains monthly despite unchanged model behavior in the current scenario.

C:

The proposed probe concerns a search index changes embedding model and chunking approach, where explicit index versions support compatible retrieval and rollback; this incident instead requires evidence about drift monitoring and outcomes justify retraining when performance shifts and the failure that team retrains monthly despite unchanged model behavior.

D:

The proposed probe concerns a customer handoff depends on shared record and term mappings, where cross-app release plans validate shared contracts and timing; this incident instead requires evidence about drift monitoring and outcomes justify retraining when performance shifts and the failure that team retrains monthly despite unchanged model behavior.

Question 225

A test plan focuses on a previously missed failure: Candidate improves one metric while violating another. It occurred when teams promote a custom model because it feels better. Which check directly validates that behavior?

  1. Try injection, invalid inputs and out-of-scope actions
  2. Execute refund and cross-account misuse cases in staging
  3. Run the same corpus in a CI release candidate
  4. Submit a document without classification and verify it stays excluded
  5. Evaluate candidate and baseline on identical frozen examples

Explanation

Correct Answer

E:

The scenario is that teams promote a custom model because it feels better; the defect to rule out is that candidate improves one metric while violating another; testing via evaluating candidate and baseline on identical frozen examples challenges the design dependency where predefined gates weigh accuracy, latency and business risk.

Incorrect Answers

A:

Trying injection, invalid inputs and out-of-scope actions verifies an adjacent objective where benchmarks cover only cooperative user wording; it leaves untested the causal path in which predefined gates weigh accuracy, latency and business risk for this case.

B:

A pass on that verification would speak to agent supports refunds, order status and generic FAQs; it would not show that predefined gates weigh accuracy, latency and business risk, which is needed to avoid the observed defect candidate improves one metric while violating another.

C:

This check exercises a different system behavior: stable automated suites compare response quality across versions; a pass there cannot establish that candidate improves one metric while violating another no longer occurs in the required workflow.

D:

Submitting a document without classification and verifying it stays excluded verifies an adjacent objective where new documents enter a production source automatically; it leaves untested the causal path in which predefined gates weigh accuracy, latency and business risk for this case.

Question 226

Governance approves the business purpose but requires control of this exposure: A low-capability routed model might quietly misclassify difficult cases. The agent will be used when high-volume narrow extraction needs predictable latency and cost. Which boundary is most defensible?

  1. Use approved extension points rather than unsupported modifications
  2. Fall back to a stronger model on low-confidence cases
  3. Budget for independent hosting, safety testing and monitoring
  4. Operate independent identity, safety and observability controls

Explanation

Correct Answer

B:

Because a low-capability routed model might quietly misclassify difficult cases, a guardrail must constrain the actual system behavior: smaller models can be evaluated for narrow tasks with lower cost; protecting a response alone cannot enforce that boundary.

Incorrect Answers

A:

The proposed control targets an unsupported product customization might break after vendor updates, while the given risk scenario is a low-capability routed model might quietly misclassify difficult cases; these exposures arise in different operational contexts.

C:

For the use case where a workflow needs proprietary scheduling, model choice and runtime control, the control addresses the threat that a bespoke runtime might be shipped without an accountable security or support team; the specified agent must be protected against a low-capability routed model might quietly misclassify difficult cases.

D:

The solution demands its own model selection and hosting policies; its guardrail addresses the possibility that an owned custom runtime might lack the host's identity and safety governance; the present architecture must constrain the different threat that a low-capability routed model might quietly misclassify difficult cases.

Question 227

A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: A workflow needs proprietary scheduling, model choice and runtime control. Known defect: A predefined host cannot meet the required orchestration contract. Which KPI should be selected?

  1. Outbound notices released with recorded approval
  2. Unsupported-host requirements met by the custom design
  3. Cost per accepted classification at required accuracy
  4. Authorized tool calls completed without excess privileges

Explanation

Correct Answer

B:

For the use case where a workflow needs proprietary scheduling, model choice and runtime control, the indicator unsupported-host requirements met by the custom design is worth baselining against the earlier issue where a predefined host cannot meet the required orchestration contract.

Incorrect Answers

A:

For the situation where standard capabilities draft customer communications but policy requires approval, outbound notices released with recorded approval is a sensible indicator; the problem that a predefined host cannot meet the required orchestration contract requires measurement of unsupported-host requirements met by the custom design.

C:

Although cost per accepted classification at required accuracy is observable, the affected process involves a workflow needs proprietary scheduling, model choice and runtime control; its acceptance outcome is unsupported-host requirements met by the custom design, not the alternative measurement.

D:

For the situation where a Copilot Studio agent needs a reusable interface to governed operations, authorized tool calls completed without excess privileges is a sensible indicator; the problem that a predefined host cannot meet the required orchestration contract requires measurement of unsupported-host requirements met by the custom design.

Question 228

A product owner must measure whether an AI-enabled change actually improves this process: A research agent repeatedly queries sources on difficult cases. The earlier problem was: A single case exhausts model and tool quotas. Which outcome indicator is most relevant?

  1. Unauthorized ancillary actions per task
  2. Eligible exception events creating one investigation
  3. Grounding recall for critical enterprise questions
  4. Agent runs terminating within predefined budgets

Explanation

Correct Answer

D:

For the use case where a research agent repeatedly queries sources on difficult cases, the indicator agent runs terminating within predefined budgets is worth baselining against the earlier issue where a single case exhausts model and tool quotas.

Incorrect Answers

A:

Although unauthorized ancillary actions per task is observable, the affected process involves a research agent repeatedly queries sources on difficult cases; its acceptance outcome is agent runs terminating within predefined budgets, not the alternative measurement.

B:

For the situation where shipment exceptions should initiate investigation without a chat message, eligible exception events creating one investigation is a sensible indicator; the problem that a single case exhausts model and tool quotas requires measurement of agent runs terminating within predefined budgets.

C:

The alternative tracks grounding recall for critical enterprise questions, arising from the mechanism where labeled queries and expected evidence distinguish retrieval from synthesis; its movement alone says little about the failure that a single case exhausts model and tool quotas.

Question 229

A proposed deployment supports this business case: Teams promote a custom model because it feels better. Reviewers also warn that a candidate might pass overall accuracy while failing a mandatory critical metric. Which restriction mitigates that concrete risk?

  1. Block release on any high-severity authorization regression
  2. Restore the previous prompt if critical intent selection regresses
  3. Enforce owner sign-off for high-risk grounding collections
  4. Block promotion when any mandatory critical threshold fails

Explanation

Correct Answer

D:

Teams promote a custom model because it feels better; its risk scenario is that a candidate might pass overall accuracy while failing a mandatory critical metric; the safeguard fits because predefined gates weigh accuracy, latency and business risk and the restriction addresses that specific exposure.

Incorrect Answers

A:

That protection is tied to the mechanism that risk-weighting prioritizes high-impact actions and failure paths; it does not prevent the distinct risk where a candidate might pass overall accuracy while failing a mandatory critical metric in the system under review.

B:

A rewritten prompt improves a single demonstration; its guardrail addresses the possibility that a prompt revision might break critical intent handling without a rollback; the present architecture must constrain the different threat that a candidate might pass overall accuracy while failing a mandatory critical metric.

C:

The proposed control targets sensitive ownerless documents might enter production grounding, while the given risk scenario is a candidate might pass overall accuracy while failing a mandatory critical metric; these exposures arise in different operational contexts.

Question 230

Operations has not established the cause of this defect: Users cannot tell whether a value was generated or saved. The associated setting is: Staff reject suggestions because approval controls are unclear. Which inspection should precede remediation?

  1. Observe error-prone approval and revision actions in usability tests
  2. Review target host support and publishing settings
  3. Trace page data bindings and invoked record APIs
  4. Compare generated recommendations with recorded dispositions

Explanation

Correct Answer

A:

The investigation directly tests the causal connection: clear review interfaces surface provenance and editable proposals; it collects evidence for a workflow where staff reject suggestions because approval controls are unclear but the outcome is instead that users cannot tell whether a value was generated or saved.

Incorrect Answers

B:

For the failure that an agent package is assumed to work in every host automatically, reviewing target host support and publishing settings is relevant; this question instead reports that users cannot tell whether a value was generated or saved, so different evidence is needed.

C:

Tracing page data bindings and invoked record apis explores a separate connection, where generated interface components can be grounded in governed app data and defined actions; the diagnosed requirement is staff reject suggestions because approval controls are unclear, and the actual defect is users cannot tell whether a value was generated or saved.

D:

That probe evaluates a persistent agent activity feed supports review, assignment and disposition in a workflow where operations teams review multiple proposed next actions daily; the reported failure concerns the path for which clear review interfaces surface provenance and editable proposals instead.

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