Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 88

Question 870

One enterprise rollout has 2 separately testable requirements. (1) Business use: Average accuracy hides failures on complex finance questions. Existing shortcoming: Lower-cost routing degrades a business-critical subgroup. (2) Business use: Thousands of transcripts contain recurring misunderstandings. Existing shortcoming: Analysts overlook a widespread intent failure among unique phrasings. Which 2 success metrics directly satisfy these stated cases? Choose TWO.

  1. Root-cause clusters converted to verified corrections
  2. Concessions with full recommendation-to-posting audit chain
  3. Quality-adjusted cost per approved business result
  4. Cross-role cache disclosure incidents

Explanation

Correct Answers

A:

The outcome root-cause clusters converted to verified corrections measures whether the workflow where thousands of transcripts contain recurring misunderstandings is improving; the scorecard separately needs quality-adjusted cost per approved business result.

C:

The outcome quality-adjusted cost per approved business result measures whether the workflow where average accuracy hides failures on complex finance questions is improving; the scorecard separately needs root-cause clusters converted to verified corrections.

Incorrect Answers

B:

The alternative measures concessions with full recommendation-to-posting audit chain for the situation where service recommends a price adjustment requiring finance approval; it cannot substitute for performance outcomes in the separate named cases: average accuracy hides failures on complex finance questions; thousands of transcripts contain recurring misunderstandings.

D:

The alternative measures cross-role cache disclosure incidents for the situation where shared result caching makes answers fast across roles; it cannot substitute for performance outcomes in the separate named cases: average accuracy hides failures on complex finance questions; thousands of transcripts contain recurring misunderstandings.

Question 871

An investment review asks what measurable result would justify continuing an AI solution where product catalog queries mix synonyms and exact serial codes. The original issue is: Pure semantic search misses precise part identifiers. Which indicator should be tracked?

  1. Model versions with complete reproducibility evidence
  2. Case-workspace completion without navigation detours
  3. Precision at top results for identifier and natural-language queries
  4. Answer correctness on cross-section retrieval cases

Explanation

Correct Answer

C:

This scorecard needs an outcome related to product catalog queries mix synonyms and exact serial codes; precision at top results for identifier and natural-language queries. That indicator can be checked against the earlier symptom that pure semantic search misses precise part identifiers.

Incorrect Answers

A:

Retraining uses revised labels and a different source snapshot; its intended measurement is model versions with complete reproducibility evidence; substituting that result here would not capture whether pure semantic search misses precise part identifiers has been corrected.

B:

Staff need a contextual workspace assembled around one customer case; its intended measurement is case-workspace completion without navigation detours; substituting that result here would not capture whether pure semantic search misses precise part identifiers has been corrected.

D:

The alternative tracks answer correctness on cross-section retrieval cases, arising from the mechanism where semantically coherent chunks with provenance preserve context during retrieval; its movement alone says little about the failure that pure semantic search misses precise part identifiers.

Question 872

An agent-assisted service does not behave as specified: Router saves tokens but sends difficult tasks to weak models. It is used in this situation: Real task mix differs from prelaunch performance tests. Which observable diagnostic procedure is best?

  1. Compare metered usage with adoption and concurrency assumptions
  2. Compare host limitations against the requirements matrix
  3. Inspect join keys and collision rates in the ingestion pipeline
  4. Segment failures by route, token cost and difficulty

Explanation

Correct Answer

D:

The investigation directly tests the causal connection: feedback-driven thresholds rebalance accuracy, latency and spend; it collects evidence for a workflow where real task mix differs from prelaunch performance tests but the outcome is instead that router saves tokens but sends difficult tasks to weak models.

Incorrect Answers

A:

For the failure that token and action costs exceed the pilot estimate, comparing metered usage with adoption and concurrency assumptions is relevant; this question instead reports that router saves tokens but sends difficult tasks to weak models, so different evidence is needed.

B:

In the separate scenario where a workflow needs proprietary scheduling, model choice and runtime control, the same investigation could explain a predefined host cannot meet the required orchestration contract; it does not isolate the reported cause in a system where real task mix differs from prelaunch performance tests.

C:

In the process where multiple systems use inconsistent customer identifiers and duplicate cases, the diagnostic procedure could explain responses merge two unrelated customers with similar names; this particular incident instead turns on whether feedback-driven thresholds rebalance accuracy, latency and spend and why router saves tokens but sends difficult tasks to weak models.

Question 873

The architect wants evidence before altering the deployment. In a workflow where an AI triage model under-serves one customer language, overall performance masks unequal escalation rates. What should the team examine first?

  1. Compare abandonment and handoff reasons by channel
  2. Compare outcomes by locale, channel and agent version
  3. Compare error and escalation metrics across justified cohorts
  4. Compare ticket assignment confusion matrices across versions
  5. Compare resolved, abandoned and escalated session cohorts

Explanation

Correct Answer

C:

The investigation directly tests the causal connection: segment-aware evaluations surface fairness risks; it collects evidence for a workflow where an AI triage model under-serves one customer language but the outcome is instead that overall performance masks unequal escalation rates.

Incorrect Answers

A:

The evidence this method produces applies to a workflow where chat and voice agents have different interaction patterns; it cannot by itself establish why overall performance masks unequal escalation rates in the current scenario.

B:

That probe evaluates cohort telemetry supports safe staged validation in a workflow where a conversational change affects only one support geography; the reported failure concerns the path for which segment-aware evaluations surface fairness risks instead.

D:

The proposed probe concerns a support agent changes how it classifies priority intents, where controlled service evaluations validate routing before rollout; this incident instead requires evidence about segment-aware evaluations surface fairness risks and the failure that overall performance masks unequal escalation rates.

E:

The proposed inspection is associated with support leads need resolution and escalation insights over time, where agent analytics surface outcome trends and usage signals for supported sessions; it would not expose the failure mode that overall performance masks unequal escalation rates with the required specificity.

Question 874

The architecture board has evidence that seasonality produces the same change in unassisted teams. The affected situation is: Management attributes a sales lift to Copilot without a comparison. What change most directly establishes the missing behavior?

  1. Use a controlled baseline to measure causal impact
  2. Include human review and exception handling in unit economics
  3. Define autonomy tiers by consequence and reversibility
  4. Use Dataverse as the governed handoff for customer cases

Explanation

Correct Answer

A:

Management attributes a sales lift to Copilot without a comparison; its failing behavior is that seasonality produces the same change in unassisted teams; applying the underlying design principle—matched comparisons separate solution impact from unrelated shifts—connects the requirement to an implementable control.

Incorrect Answers

B:

The technical behavior of this alternative is that end-to-end unit cost includes review, escalation and retry effort; the user case being assessed is management attributes a sales lift to Copilot without a comparison with a different failure: seasonality produces the same change in unassisted teams.

C:

For another type of architecture, tiered permissions require progressively stronger approvals for riskier actions is useful; this design must instead support management attributes a sales lift to Copilot without a comparison despite the problem where seasonality produces the same change in unassisted teams.

D:

The alternative addresses an opportunity owner sees a stale service escalation status by a design in which shared authorized records support coordinated business context; it does not supply the missing relationship for management attributes a sales lift to Copilot without a comparison, where seasonality produces the same change in unassisted teams.

Question 875

Support engineers report this failure: A replaced model package introduces unknown behavior. The business context is: Deployment automation pulls model artifacts from unverified origins. What evidence should be collected to isolate the cause?

  1. Match release manifest to transformation and model checksums
  2. Inspect signatures, hashes and approved source records
  3. Inspect artifacts for training data hashes and evaluation reports
  4. Reconcile active model versions with promotion events

Explanation

Correct Answer

B:

Where a replaced model package introduces unknown behavior, the useful evidence comes from inspecting signatures, hashes and approved source records; its focus matches the architectural mechanism that artifact provenance and integrity checks guard model supply chain in the process where deployment automation pulls model artifacts from unverified origins.

Incorrect Answers

A:

A preprocessing release changes category labels incorrectly; its matching diagnostic check is reasonable for restoring code alone does not restore previous predictions; the present investigation needs evidence for the different failure where a replaced model package introduces unknown behavior.

C:

The evidence this method produces applies to a workflow where data scientists cannot explain why a newer classifier behaves differently; it cannot by itself establish why a replaced model package introduces unknown behavior in the current scenario.

D:

Reconciling active model versions with promotion events explores a separate connection, where deployment audit links actor, approved artifact and release time; the diagnosed requirement is deployment automation pulls model artifacts from unverified origins, and the actual defect is a replaced model package introduces unknown behavior.

Question 876

An investment review asks what measurable result would justify continuing an AI solution where agent must remember order context while asking for missing details. The original issue is: It substitutes a prior customer's order in a later turn. Which indicator should be tracked?

  1. Multi-turn tasks resolved without entity confusion
  2. Concessions with full recommendation-to-posting audit chain
  3. Complete onboarding cases with accurate Sales context
  4. Critical journeys passing mandatory acceptance tests

Explanation

Correct Answer

A:

The use case involves agent must remember order context while asking for missing details and the baseline failure that it substitutes a prior customer's order in a later turn; the KPI tracks multi-turn tasks resolved without entity confusion, so improvement has an interpretable operational meaning.

Incorrect Answers

B:

For the workflow where service recommends a price adjustment requiring finance approval, concessions with full recommendation-to-posting audit chain is an actionable measure; it does not establish whether it substitutes a prior customer's order in a later turn in the situation under review.

C:

A won opportunity should create a properly owned onboarding case; its intended measurement is complete onboarding cases with accurate Sales context; substituting that result here would not capture whether it substitutes a prior customer's order in a later turn has been corrected.

D:

For the workflow where agent supports refunds, order status and generic FAQs, critical journeys passing mandatory acceptance tests is an actionable measure; it does not establish whether it substitutes a prior customer's order in a later turn in the situation under review.

Question 877

The fault affects an AI-powered workflow in which sellers need current credit exposure held outside CRM. Specifically, Copilot relies on last quarter's manually imported account report. Which diagnostic investigation best distinguishes the relevant failure path?

  1. Compare source timestamps and principal mapping on connector calls
  2. Compare summaries with mandated case-handoff sections
  3. Trace action authentication and source mutation evidence
  4. Review confusion pairs against the customer-service glossary

Explanation

Correct Answer

A:

In the setting where sellers need current credit exposure held outside CRM, comparing source timestamps and principal mapping on connector calls can isolate the failure in which Copilot relies on last quarter's manually imported account report, because a supported connector can retrieve external information with explicit identity mapping.

Incorrect Answers

B:

Comparing summaries with mandated case-handoff sections explores a separate connection, where tailored instructions and contextual fields standardize case handoff content; the diagnosed requirement is sellers need current credit exposure held outside CRM, and the actual defect is Copilot relies on last quarter's manually imported account report.

C:

The proposed inspection is associated with an agent needs to update a governed ticket status, where an authorized connector action invokes an operational service; it would not expose the failure mode that Copilot relies on last quarter's manually imported account report with the required specificity.

D:

The proposed probe concerns agents must distinguish an incident, complaint, inquiry and entitlement, where business-term definitions align user vocabulary with governed service entities; this incident instead requires evidence about a supported connector can retrieve external information with explicit identity mapping and the failure that Copilot relies on last quarter's manually imported account report.

Question 878

A planned AI solution must support this situation: Several agents need the same inventory query operation. Its prototype fails because each agent builds a different unaudited HTTP wrapper. Which design best fits the constraint?

  1. Use a declarative agent for role-tailored Microsoft 365 work
  2. Use MCP tools for a governed reusable integration surface
  3. Make task actions retry-safe through idempotency
  4. Use agent reasoning for uncertain multi-step analysis with bounded tools

Explanation

Correct Answer

B:

The defect is that each agent builds a different unaudited HTTP wrapper; the expected business case is that several agents need the same inventory query operation; the design fits because a supported MCP server exposes typed tool capabilities to clients.

Incorrect Answers

A:

A custom service would duplicate built-in host orchestration; the alternate design can address that situation because declarative agents define instructions, knowledge and actions in the supported host, whereas this question needs a solution to each agent builds a different unaudited HTTP wrapper.

C:

There is a valid place for the mechanism that idempotency keys tie repeated submissions to one business operation, namely network retries can repeat a single purchase request; it does not remove the specific defect that each agent builds a different unaudited HTTP wrapper from this case.

D:

Reasoning can plan through evidence and tools within enforced boundaries; its main purpose is procurement exceptions require comparing several policies and records, not correction of the reported defect where each agent builds a different unaudited HTTP wrapper in the target process.

Question 879

The pilot needs a durable correction: Makers validate only a few manual chats. This affects the workflow in which prompt revisions cause intermittent regressions. Which design choice should the architect approve?

  1. Track successful task completion rather than conversational volume
  2. Test multi-turn context retention and clarification
  3. Create actionable SLOs and incident alerts for agent flows
  4. Automate repeatable evaluation gates before publishing

Explanation

Correct Answer

D:

The defect is that makers validate only a few manual chats; the expected business case is that prompt revisions cause intermittent regressions; the design fits because stable automated suites compare response quality across versions.

Incorrect Answers

A:

When a service bot handles many chats but few cases are resolved, outcome measures connect agent use with completed business work; in the stated workflow, however, makers validate only a few manual chats; the relevant architectural behavior is that stable automated suites compare response quality across versions.

B:

Multi-turn tests assess state tracking, entity separation and follow-up questions; its main purpose is agent must remember order context while asking for missing details, not correction of the reported defect where makers validate only a few manual chats in the target process.

C:

The alternative addresses teams discover failures through complaints several days later by a design in which service-level objectives and alert thresholds turn telemetry into response signals; it does not supply the missing relationship for prompt revisions cause intermittent regressions, where makers validate only a few manual chats.

Question 880

Operations has not established the cause of this defect: Users receive permission errors after successful package import. The associated setting is: An imported agent can authenticate in dev but not production. Which inspection should precede remediation?

  1. Review backend authorization logs against actor identities
  2. Trace token acquisition and authorization results by environment
  3. Inspect accessible tool actions and underlying API scopes
  4. Audit token use and secret inventory for agent services

Explanation

Correct Answer

B:

For the situation where an imported agent can authenticate in dev but not production, the defect is that users receive permission errors after successful package import; tracing token acquisition and authorization results by environment helps verify the cause-and-effect path: production consent, identity and connection setup are distinct release checks.

Incorrect Answers

A:

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 production consent, identity and connection setup are distinct release checks and why users receive permission errors after successful package import.

C:

The evidence this method produces applies to a workflow where an agent should summarize account data but not alter it; it cannot by itself establish why users receive permission errors after successful package import in the current scenario.

D:

Auditing token use and secret inventory for agent services explores a separate connection, where managed identity provides credential-less workload authentication where supported; the diagnosed requirement is an imported agent can authenticate in dev but not production, and the actual defect is users receive permission errors after successful package import.

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