100% Latest & Updated Microsoft AB-100 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 1
A benefits agent cites an old benefits handbook after HR published a new policy yesterday; document access works but the indexed snapshot is stale. Which design decision should the architect prioritize?
- Refresh and reindex the authoritative knowledge corpus on a defined cadence.
- Normalize entity identifiers and deduplicate conflicting source records before indexing.
- Document timestamp semantics and freshness service-level objectives in the data contract.
- Apply caller-scoped access permissions when retrieving protected knowledge documents.
Explanation
Correct Answer
A:Reindex the authoritative updated HR handbook so the benefits agent retrieves yesterday's policy instead of an older indexed snapshot; access permissions are already functioning.
Incorrect Answers
B:Resolving duplicate entity IDs prevents conflicting entries for one record, but does not refresh the out-of-date handbook that the benefits agent currently cites.
C:Freshness SLAs and timestamp definitions clarify how recent data should be, yet they cannot replace stale indexed passages without an actual ingestion or reindex operation.
D:Caller-scoped retrieval permissions would prevent unauthorized disclosures. The agent can already read its handbook, so narrowing access does not update the stale policy text.
Question 2
A service agent quotes a confidential regional pay table to an employee outside that region even though the source itself is current. Which controlled change is most appropriate for this situation?
- Enforce approved data-region placement and verify every cross-region data flow.
- Apply caller-scoped access permissions when retrieving protected knowledge documents.
- Curate authoritative policy sources and explicit precedence for approved versions.
- Normalize entity identifiers and deduplicate conflicting source records before indexing.
Explanation
Correct Answer
B:Filter protected pay-table retrieval by the requesting employee's identity and permissions before ranking or generating an answer; the correct current table should not be visible to out-of-region staff.
Incorrect Answers
A:Storing documents in an approved region addresses data residency, but a regional employee can still retrieve confidential pay data unless each caller's access rights are enforced.
C:Source precedence selects which policy is authoritative, but even the newest authoritative pay table may be confidential to employees outside its designated region.
D:Canonical record IDs remove duplicates, not inappropriate entitlement. Deduplicating the pay table would still allow unauthorized retrieval without caller-specific permissions.
Question 3
A claims agent retrieves contradictory duplicate records with different field names for the same claim; every record is recent. Which design choice targets the observed problem?
- Curate authoritative policy sources and explicit precedence for approved versions.
- Collect classifications and effective access policies for sensitive attributes.
- Normalize entity identifiers and deduplicate conflicting source records before indexing.
- Refresh and reindex the authoritative knowledge corpus on a defined cadence.
Explanation
Correct Answer
C:Normalize the claim identifier and field schema, then reconcile duplicates before indexing so the agent receives one consistent representation of each claim.
Incorrect Answers
A:Document authority rules can choose between signed and draft policies; this case instead contains duplicate claim records with inconsistent field names requiring identity and schema reconciliation.
B:Classifying sensitive fields protects restricted attributes, but does not reconcile two recent claims records that represent the same case under different identifiers.
D:Refreshing the index would simply ingest both recent contradictory claim records again unless they are normalized and deduplicated first.
Question 4
A compliance agent uses an unofficial draft policy that ranks above the signed policy in retrieved results, despite both being readable. Which concrete solution decision should the team make?
- Define one governed knowledge-and-decision service reused through approved channel adapters.
- Apply caller-scoped access permissions when retrieving protected knowledge documents.
- Refresh and reindex the authoritative knowledge corpus on a defined cadence.
- Curate authoritative policy sources and explicit precedence for approved versions.
Explanation
Correct Answer
D:Curate approved policy sources and enforce signed-versus-draft precedence in retrieval so an unofficial copy cannot displace governing signed policy evidence.
Incorrect Answers
A:A shared decision service can apply common policy to several channels, but it cannot rank a signed policy above a draft unless the source-selection rules identify authority and precedence.
B:Both documents are currently readable, so a caller access filter does not distinguish an approved signed policy from an unofficial draft.
C:Reindexing recent content helps freshness, but a newly indexed draft can still outrank the signed policy if the retrieval system lacks explicit authority precedence.
Question 5
A vendor invoice must be matched to an exact purchase-order key and rejected using fixed tolerance arithmetic; outcomes must be identical for the same inputs. Which recommendation addresses the identified failure most directly?
- Design a bounded agent with tools and explicit approval gates for consequential steps.
- Implement a deterministic validation workflow with fixed business rules.
- Use an AI assistant to synthesize evidence and require a human final buying decision.
- Use fixed classification rules and expose exceptions for manual review.
Explanation
Correct Answer
B:Implement fixed validation rules for exact PO identifiers and tolerance calculations, with consistent rejection and audit outcomes for identical invoice inputs.
Incorrect Answers
A:An agent with tools adds flexible planning where the requirement instead calls for exact purchase-order key matching and deterministic tolerance arithmetic; extra autonomy makes results less reproducible.
C:A human purchasing recommendation can be appropriate for qualitative supplier trade-offs, but deterministic invoice acceptance must follow the same arithmetic and keys every time.
D:A fixed classifier can label cases, but it does not by itself check matching purchase-order fields and perform the specified numerical tolerances that determine payment eligibility.
Question 6
A procurement analyst needs to compare several supplier narratives and draft a recommendation with cited trade-offs for a human buyer. What should the solution architect recommend for this specific need?
- Design a bounded agent with tools and explicit approval gates for consequential steps.
- Build a governed analytics report over the ticket dataset.
- Use an AI assistant to synthesize evidence and require a human final buying decision.
- Use AI to recommend with reasons while reserving eligibility decisions for authorized staff.
Explanation
Correct Answer
C:Use an AI assistant to extract and cite supplier claims, compare their trade-offs and draft a recommendation while leaving the final purchasing judgment with an accountable human.
Incorrect Answers
A:A tool-using agent may be useful for executing approved procurement steps, but the analyst currently needs evidence synthesis and a recommendation, not autonomous transactional action.
B:Aggregate ticket reporting produces counts and trends, whereas this decision requires comparing qualitative supplier narratives and writing cited trade-offs for a buyer.
D:Human ownership of eligibility decisions is important in benefits workflows, but no eligibility determination is described; the procurement buyer needs a sourced comparison.
Question 7
A service mailbox receives multi-step requests that require choosing approved tools to gather context and propose next actions across systems. What is the most defensible intervention for this case?
- Implement a deterministic validation workflow with fixed business rules.
- Build a governed analytics report over the ticket dataset.
- Publish a canonical customer schema with mapped identifiers across both systems.
- Design a bounded agent with tools and explicit approval gates for consequential steps.
Explanation
Correct Answer
D:Implement a bounded agent that invokes only approved tools, tracks their results, and routes consequential steps through an independent approval gate before execution.
Incorrect Answers
A:Fixed business rules work for deterministic validations, but cannot flexibly select approved tools and sequence information gathering across several changing mailbox requests.
B:A governed ticket dashboard summarizes historical cases, not the context-sensitive tool selection and cross-system follow-up needed for incoming service requests.
C:Canonical customer identifiers help different systems agree on identity, but do not coordinate the selection and ordering of mailbox actions across tools.
Question 8
A manager wants weekly counts of resolved tickets by queue, with no language interpretation or tool planning. Which architecture response most directly addresses the risk?
- Build a governed analytics report over the ticket dataset.
- Use an AI assistant to synthesize evidence and require a human final buying decision.
- Separate read-only planning tools from a booking action that requires explicit user confirmation.
- Implement a deterministic validation workflow with fixed business rules.
Explanation
Correct Answer
A:Use a governed analytics report with a weekly queue-by-queue count of resolved tickets; ordinary aggregation meets the request without language generation or agent planning.
Incorrect Answers
B:Supplier-narrative synthesis is valuable when comparing qualitative evidence, but the manager asks for fixed counts from structured ticket data.
C:Planning-versus-booking controls protect a travel transaction, not a weekly resolved-ticket count calculated from an existing database.
D:Deterministic validation checks individual records against rules; the requested output is an aggregated management report grouped by week and ticket queue.
Question 9
A legal team permits an AI tool to summarize precedents but prohibits it from sending communications to clients. Which solution approach best meets the stated requirement?
- Use fixed classification rules and expose exceptions for manual review.
- Insert an authenticated human approval step before high-value refund execution.
- Give the summarization component read-only data and no outbound communication tool.
- Curate authoritative policy sources and explicit precedence for approved versions.
Explanation
Correct Answer
C:Grant the precedent summarizer read-only evidence access and no email or messaging tool, preventing its generated text from initiating unauthorized client communications.
Incorrect Answers
A:A fixed classifier could categorize precedents, but would not enforce that a legal summarization component lacks a client-email or messaging capability.
B:Approval before a large refund protects financial transactions; the prohibited operation here is any outbound communication from the legal research tool.
D:Approved document precedence helps select governing precedents but cannot block the program from sending messages if outbound tools remain available.
Question 10
A billing team wants an agent to propose refunds, but individual refunds above a threshold must be approved. Which option most directly closes the documented gap?
- Design a bounded agent with tools and explicit approval gates for consequential steps.
- Use fixed classification rules and expose exceptions for manual review.
- Separate read-only planning tools from a booking action that requires explicit user confirmation.
- Insert an authenticated human approval step before high-value refund execution.
Explanation
Correct Answer
D:Validate the refund amount server-side and require authenticated human approval before the execution API processes any refund above the configured threshold.
Incorrect Answers
A:A general high-impact approval framework is helpful, but the stated control must specifically enforce the refund monetary threshold at the execution boundary.
B:A deterministic case classifier can label refund requests but does not obtain human authorization before money is actually returned.
C:Separating travel quote gathering from ticket booking concerns a different purchase flow; refund approval must be attached to the billing execution API.
Agentic AI Business Solutions Architect (AB-100) Practice Tests