Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 112

Question 1110

A product owner must measure whether an AI-enabled change actually improves this process: Sellers need current credit exposure held outside CRM. The earlier problem was: Copilot relies on last quarter's manually imported account report. Which outcome indicator is most relevant?

  1. Sales task completion under external API impairment
  2. Fresh authorized account insights from external systems
  3. Handoffs with complete approved case context
  4. Correct identification of governed service intents

Explanation

Correct Answer

B:

The use case involves sellers need current credit exposure held outside CRM and the baseline failure that Copilot relies on last quarter's manually imported account report; the KPI tracks fresh authorized account insights from external systems, so improvement has an interpretable operational meaning.

Incorrect Answers

A:

The premise of this indicator is bounded timeouts and graceful fallback preserve essential user work; the architecture being evaluated instead serves sellers need current credit exposure held outside CRM, and its success is measured by fresh authorized account insights from external systems.

C:

Agents need structured case summaries at handoff; its intended measurement is handoffs with complete approved case context; substituting that result here would not capture whether Copilot relies on last quarter's manually imported account report has been corrected.

D:

For the workflow where agents must distinguish an incident, complaint, inquiry and entitlement, correct identification of governed service intents is an actionable measure; it does not establish whether Copilot relies on last quarter's manually imported account report in the situation under review.

Question 1111

One enterprise rollout has 3 separately testable requirements. (1) Business use: Customer terminology changes over time. Known defect: Team retrains monthly despite unchanged model behavior. (2) Business use: A retraining release changes the tokenizer and label order. Known defect: Downstream apps interpret correct predictions as the wrong category. (3) Business use: A tuned model claims higher accuracy but adds latency. Known defect: Direct replacement removes the dependable baseline. Which 3 regression tests directly satisfy these stated cases? Choose THREE.

  1. Send an identical holdout to champion and challenger
  2. Simulate an abandoned interaction and inspect outcome reporting
  3. Invoke both versions with boundary category samples
  4. Introduce a known distribution shift and verify review triggers

Explanation

Correct Answers

A:

For the business workflow where a tuned model claims higher accuracy but adds latency, sending an identical holdout to champion and challenger challenges the failure that direct replacement removes the dependable baseline; the release also needs a separate check for customer terminology changes over time; a retraining release changes the tokenizer and label order.

C:

For the business workflow where a retraining release changes the tokenizer and label order, invoking both versions with boundary category samples challenges the failure that downstream apps interpret correct predictions as the wrong category; the release also needs a separate check for customer terminology changes over time; a tuned model claims higher accuracy but adds latency.

D:

For the business workflow where customer terminology changes over time, introducing a known distribution shift and verifying review triggers challenges the failure that team retrains monthly despite unchanged model behavior; the release also needs a separate check for a retraining release changes the tokenizer and label order; a tuned model claims higher accuracy but adds latency.

Incorrect Answers

B:

This regression test is associated with the different failure where team measures chat volume but cannot see failed conversations in a workflow where support leads need resolution and escalation insights over time; it does not directly solve the listed conditions: team retrains monthly despite unchanged model behavior; downstream apps interpret correct predictions as the wrong category; direct replacement removes the dependable baseline.

Question 1112

A quality engineer is writing a regression case after the incident where the agent invents discretionary promises. The underlying business scenario is: Compliance letters must preserve approved language and disclaimers. What should the test exercise?

  1. Change a test case status and inspect audit events
  2. Search for an external document as permitted and denied users
  3. Request the same account help by voice and chat
  4. Generate stressful and adversarial user examples

Explanation

Correct Answer

D:

The agent invents discretionary promises; a release test must establish that this no longer occurs; generating stressful and adversarial user examples measures the specific behavior for which templating and approved clauses limit generated variation.

Incorrect Answers

A:

A pass on that verification would speak to an agent needs to update a governed ticket status; it would not show that templating and approved clauses limit generated variation, which is needed to avoid the observed defect the agent invents discretionary promises.

B:

The verification concerning workers need internal wiki articles from outside Microsoft 365 checks supported connectors expose external content with discoverability and security metadata against Copilot cannot find approved wiki pages by natural language; its result says little about the observed issue where the agent invents discretionary promises.

C:

Requesting the same account help by voice and chat verifies an adjacent objective where the same agent is offered in voice and asynchronous chat; it leaves untested the causal path in which templating and approved clauses limit generated variation for this case.

Question 1113

A proposed deployment supports this business case: A small portion of calls time out during peak demand. Reviewers also warn that tail latency might exceed the business SLO during model or tool delays. Which restriction mitigates that concrete risk?

  1. Disable the candidate for a cohort on critical regression
  2. Apply timeouts, scaling and graceful degradation to slow paths
  3. Remove unused privileged actions from the agent
  4. Keep rollback available until all critical criteria pass

Explanation

Correct Answer

B:

In the context where a small portion of calls time out during peak demand, the exposure is that tail latency might exceed the business SLO during model or tool delays; containing it requires a guardrail in the operation where tail percentiles surface extreme experiences affecting users.

Incorrect Answers

A:

This restriction is associated with a conversational change affects only one support geography and addresses the exposure that a regional pilot might produce a severe local regression hidden by global averages; it does not specifically protect the target workflow against tail latency might exceed the business SLO during model or tool delays.

C:

Where planner calls the wrong tool among overlapping capabilities, the guard can address a planner might choose an unused privileged tool that remains exposed; the exposure that tail latency might exceed the business SLO during model or tool delays remains relevant to the present assessment.

D:

This restriction is associated with a tuned model claims higher accuracy but adds latency and addresses the exposure that a new classifier might replace its safe baseline before comparisons finish; it does not specifically protect the target workflow against tail latency might exceed the business SLO during model or tool delays.

Question 1114

A risk analysis highlights a threat in this workflow: Agent supports refunds, order status and generic FAQs. Exposure: A release might ship with an exploitable record-access regression. Which enforceable safeguard addresses it at the relevant boundary?

  1. Block release on any high-severity authorization regression
  2. Require heightened review for autonomous privileged agents
  3. Block promotion when any mandatory critical threshold fails
  4. Treat generated expectations as suggestions until verified

Explanation

Correct Answer

A:

When a release might ship with an exploitable record-access regression, the use case agent supports refunds, order status and generic FAQs requires a guardrail aligned with the behavior that risk-weighting prioritizes high-impact actions and failure paths.

Incorrect Answers

B:

The alternate boundary mitigates a privileged autonomous agent might receive the same review as a private helper, which is relevant to personal read-only assistants and payment agents have different risk; it leaves the current workflow vulnerable to a release might ship with an exploitable record-access regression.

C:

For the use case where teams promote a custom model because it feels better, the control addresses the threat that a candidate might pass overall accuracy while failing a mandatory critical metric; the specified agent must be protected against a release might ship with an exploitable record-access regression.

D:

That protection is tied to the mechanism that specification-guided generation covers missing fields and denied roles; it does not prevent the distinct risk where a release might ship with an exploitable record-access regression in the system under review.

Question 1115

A proposed deployment supports this business case: Planner calls the wrong tool among overlapping capabilities. Reviewers also warn that a planner might choose an unused privileged tool that remains exposed. Which restriction mitigates that concrete risk?

  1. Remove unused privileged actions from the agent
  2. Reject cached results that omit original security context
  3. Remove personal details before using feedback as test data
  4. Deny external sends unless independently approved

Explanation

Correct Answer

A:

Because a planner might choose an unused privileged tool that remains exposed, a guardrail must constrain the actual system behavior: clear capabilities and restricted sets improve selection reliability; protecting a response alone cannot enforce that boundary.

Incorrect Answers

B:

The proposed control targets shared answer caches might drop source ACLs and leak a private document, while the given risk scenario is a planner might choose an unused privileged tool that remains exposed; these exposures arise in different operational contexts.

C:

The alternate guard addresses real customer details might leak into evaluation corpora derived from tickets for support tickets reveal an unusual multistep conversation failure; the present threat, a planner might choose an unused privileged tool that remains exposed, needs a different enforcement boundary.

D:

Where a prompt injection requests data export to an outside address, the guard can address an injected tool instruction might export customer records externally; the exposure that a planner might choose an unused privileged tool that remains exposed remains relevant to the present assessment.

Question 1116

In a business situation where third-party documents include sentences telling the model to ignore rules, a previous build showed this defect: Answers adopt an instruction embedded in a source document. Which reproducible test belongs in the release gate?

  1. Inject hostile instructions into a test document and check refusal
  2. Add a confidential draft and verify it is excluded for unauthorized users
  3. Test an out-of-corpus question and expect a no-evidence response
  4. Prototype the distinguishing behavior under production constraints
  5. Disconnect the source and verify a transparent failure response

Explanation

Correct Answer

A:

The scenario is that third-party documents include sentences telling the model to ignore rules; the defect to rule out is that answers adopt an instruction embedded in a source document; testing via injecting hostile instructions into a test document and checking refusal challenges the design dependency where trust boundaries treat source text as evidence rather than authority.

Incorrect Answers

B:

The verification concerning search spans internal manuals with different owners and permissions checks metadata enables filtering, review, retention and traceable answers against a model retrieves a draft manual without any accountable owner; its result says little about the observed issue where answers adopt an instruction embedded in a source document.

C:

The tested behavior is that cited retrieval separates factual support from free-form generation; the necessary evidence here concerns trust boundaries treat source text as evidence rather than authority and the previously observed answers adopt an instruction embedded in a source document.

D:

For competitive requirements depend on unique multi-model control, the alternate test is about whether custom engineering trades more control for more operating responsibility; it cannot prove the outcome needed after the reported regression where answers adopt an instruction embedded in a source document.

E:

For required knowledge stores are intermittently unreachable, the alternate test is about whether health checks and explicit no-evidence responses handle missing context safely; it cannot prove the outcome needed after the reported regression where answers adopt an instruction embedded in a source document.

Question 1117

Before introducing a fix, the team must explain why agent actions fail only after the connector reaches production. The affected setting is: An external API changes a required field used by an agent. What should be inspected?

  1. Review virtual-entity metadata and source-field changes
  2. Review backend authorization logs against actor identities
  3. Review connector groups and blocked action combinations
  4. Compare schema differences and failed action bindings

Explanation

Correct Answer

D:

In the setting where an external API changes a required field used by an agent, comparing schema differences and failed action bindings can isolate the failure in which agent actions fail only after the connector reaches production, because versioned API contracts and consumer checks detect breaking changes.

Incorrect Answers

A:

The evidence this method produces applies to a workflow where AI assistance reads operational records through Dataverse; it cannot by itself establish why agent actions fail only after the connector reaches production in the current scenario.

B:

In the process where prompt wording persuades the agent to submit a privileged operation, the diagnostic procedure could explain API trusts the model's text assertion of permission; this particular incident instead turns on whether versioned API contracts and consumer checks detect breaking changes and why agent actions fail only after the connector reaches production.

C:

That probe evaluates Power Platform data policies constrain connector use and movement pathways in a workflow where a Copilot Studio agent can read corporate data and use public messaging; the reported failure concerns the path for which versioned API contracts and consumer checks detect breaking changes instead.

Question 1118

An agent-assisted service does not behave as specified: Host-managed orchestration cannot satisfy a unique execution plan. It is used in this situation: The solution demands its own model selection and hosting policies. Which observable diagnostic procedure is best?

  1. Compare orchestration requirements with host extensibility limits
  2. Compare gap size to supported extension interfaces
  3. Compare baseline model errors by language and domain
  4. Inspect task envelopes and agent capability declarations

Explanation

Correct Answer

A:

For the case where the solution demands its own model selection and hosting policies, the diagnostic path must explain the failure that host-managed orchestration cannot satisfy a unique execution plan; comparing orchestration requirements with host extensibility limits targets the mechanism by which custom-engine architecture controls runtime, model calls and integrations.

Incorrect Answers

B:

The proposed probe concerns current users need one governed custom action inside their workflow, where extension retains familiar user experience and adds bounded custom behavior; this incident instead requires evidence about custom-engine architecture controls runtime, model calls and integrations and the failure that host-managed orchestration cannot satisfy a unique execution plan.

C:

The proposed probe concerns a unique domain vocabulary produces persistent measured errors, where a measured capability gap, appropriate dataset and MLOps plan justify customization; this incident instead requires evidence about custom-engine architecture controls runtime, model calls and integrations and the failure that host-managed orchestration cannot satisfy a unique execution plan.

D:

In the process where independently hosted agents must exchange tasks and results, the diagnostic procedure could explain cross-team agents rely on hard-coded, vendor-specific chat formats; this particular incident instead turns on whether custom-engine architecture controls runtime, model calls and integrations and why host-managed orchestration cannot satisfy a unique execution plan.

Question 1119

Governance approves the business purpose but requires control of this exposure: A transferred voice conversation might disclose protected context before verifying identity. The agent will be used when a voice bot cannot resolve a complex eligibility dispute. Which boundary is most defensible?

  1. Limit deployment to approved users and surfaces
  2. Ensure the backend rejects repeat writes for the same intent
  3. Confirm identity before sharing protected case details
  4. Cap tool depth and require evidence for consequential conclusions

Explanation

Correct Answer

C:

In the context where a voice bot cannot resolve a complex eligibility dispute, the exposure is that a transferred voice conversation might disclose protected context before verifying identity; containing it requires a guardrail in the operation where structured handoff passes allowed intent, transcript summary and case context.

Incorrect Answers

A:

The safeguard for agent experience must appear in selected Microsoft 365 contexts deals with a different threat: a Teams-targeted agent package might be made available on unapproved hosts; even after it is enforced, the exposure that a transferred voice conversation might disclose protected context before verifying identity remains possible in the target workflow where structured handoff passes allowed intent, transcript summary and case context.

B:

For the use case where network retries can repeat a single purchase request, the control addresses the threat that an API retry might create a second purchase for the same user instruction; the specified agent must be protected against a transferred voice conversation might disclose protected context before verifying identity.

D:

The safeguard for procurement exceptions require comparing several policies and records deals with a different threat: a reasoning agent might continue invoking tools without evidence or budget limits; even after it is enforced, the exposure that a transferred voice conversation might disclose protected context before verifying identity remains possible in the target workflow where structured handoff passes allowed intent, transcript summary and case context.

Question 1120

An agent rollout is intended for this situation: Business, platform, security and legal teams make conflicting AI decisions. The known problem is that deployments stall because approvals and accountability are unclear. Which success measure connects to that specific outcome?

  1. Median elapsed time from design decision to approval
  2. High-value concessions with complete approval trace
  3. Percentage of production agents with staffed owners and SLOs
  4. Error-rate delta between canary and control cohorts

Explanation

Correct Answer

A:

Because a named strategy team assigns decision rights and resolves cross-function tradeoffs, the outcome worth following is median elapsed time from design decision to approval; the metric is tied to the reported issue—deployments stall because approvals and accountability are unclear—for the scenario where business, platform, security and legal teams make conflicting AI decisions.

Incorrect Answers

B:

For the situation where service, sales and finance jointly decide high-value concessions, high-value concessions with complete approval trace is a sensible indicator; the problem that deployments stall because approvals and accountability are unclear requires measurement of median elapsed time from design decision to approval.

C:

Although percentage of production agents with staffed owners and SLOs is observable, the affected process involves business, platform, security and legal teams make conflicting AI decisions; its acceptance outcome is median elapsed time from design decision to approval, not the alternative measurement.

D:

The premise of this indicator is controlled release gates restrict blast radius and support rollback; the architecture being evaluated instead serves business, platform, security and legal teams make conflicting AI decisions, and its success is measured by median elapsed time from design decision to approval.

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