Microsoft AB-100 Practice Test 7

Question 60

A service company has a well-performing grounded model but experiences errors on internal product codes. What is an appropriate signal that custom tuning might be justified?

Measured failure on a representative held-out product-code task despite strong prompting and grounding baselines.
A single inaccurate conversation without reproducing its cause.
Assume a larger GPU allocation justifies fine-tuning even though held-out product-code accuracy already meets the business threshold.
Treat the model vendor’s published average benchmark as proof custom tuning is necessary without testing company-specific product-code errors.

Explanation

Correct Answer

A targeted validated gap and quality baseline can justify extra training and governance cost.

Incorrect Answers

One example may be bad retrieval rather than model deficiency.

Hardware availability does not prove a model-capability gap; training is justified by representative measured failures after grounding and prompting have been tested.

General vendor scores do not show that retrieval and prompting baselines fail on the actual internal code distribution.

Question 61

A reusable prompt library contains dozens of templates that silently change after publication. How should change management work?

Store the latest approved prompt as an unversioned file in the team repository and retain its newest copy only.
Promote whichever prompt template passes adoption testing, without running its required-output and tool-schema regression cases.
Use reviewed version identifiers, test cases, approvers and rollback to known-good prompt releases.
Allow untracked edits in production if users say the prompt sounds better.

Explanation

Correct Answer

Controlled prompt releases prevent surprising behavior shifts and make testing reproducible.

Incorrect Answers

A repository copy without immutable release versions or dependency records cannot reconstruct which prompt and tool contract ran before a regression.

Popularity and usability are meaningful adoption evidence but cannot establish that a changed prompt still meets business output, data access and tool requirements.

Unreviewed edits remove accountability and comparison baselines.

Question 62

A regional claims team needs on-prem inference for a short fixed classification vocabulary. Which comparison establishes whether a compact model is adequate?

Judge quality only by the smaller model’s file size.
Benchmark a suitable small model against labeled claims, privacy and hardware latency requirements.
Assume a cloud-scale model will work without network access.
Infer accuracy from vocabulary size without testing actual claim language.

Explanation

Correct Answer

Measured task performance and deployment constraints determine fitness, not parameter count alone.

Incorrect Answers

File size does not establish classification precision.

Offline systems cannot depend on unavailable hosted inference.

Small label sets can still require sophisticated language understanding.

Question 63

A financial advice agent tends to omit uncertainties when retrieval returns contradictory guidance. Which prompt practice should be required?

Choose the strongest wording and erase the other source.
Increase generation temperature so the model chooses one contradictory policy consistently rather than exposing unresolved legal authority.
Use model self-confidence as a replacement for authoritative policy resolution.
Surface the disagreement with source citations and withhold definitive advice until authority is resolved.

Explanation

Correct Answer

Conflicts between evidence sources should be explicit and auditable before high-impact advice.

Incorrect Answers

Silencing contradictions produces unsupported certainty.

Temperature changes sampling behavior; it cannot establish which conflicting policy source is binding for the scenario.

Model confidence is not external evidence of authority.

Question 64

A decentralized agent program suffers from inconsistent approval gates and unclear privacy responsibilities. What should its AI Center of Excellence publish?

Publish a uniform prompt and model-selection standard but allow every team to approve its own unrestricted tool credentials.
A list of preferred model names without owner or release accountability.
A reusable governance baseline covering data classification, tool permissions, evaluation and escalation.
A direction to ignore domain-specific regulatory obligations.

Explanation

Correct Answer

Shared patterns and controls make distributed agent teams safer without blocking domain expertise.

Incorrect Answers

Prompt consistency does not create privacy ownership or access enforcement. Governance must include validated approval and scoped connector permissions.

Model catalogs alone cannot define review or approval responsibility.

Local law and policy still apply within each use case.

Question 65

A Dynamics 365 Sales agent sends a quote to Finance for credit checking. The two applications disagree about the customer currency. Which integration control addresses the mismatch?

Copy the quote as text with no typed currency code or business key.
Ignore credit validation if the sales agent offers a polished explanation.
Resolve currency from the first symbol in a quote without validating its ISO code, exchange rate, or customer context.
Use an explicit currency and customer-identity contract with conversion rules and reconciliation tests.

Explanation

Correct Answer

Typed cross-application contracts prevent silent money and identity errors.

Incorrect Answers

Losing currency codes prevents verifiable financial calculations.

Narrative fluency cannot replace a credit check.

The same symbol can denote different currencies; structured codes and approved rates are needed for safe credit decisions.

Question 66

An autonomous returns agent may propose goodwill credits but the amount above $1,000 needs supervisor approval. Which state-machine step is correct?

Post the credit immediately and mark it pending afterward.
Let an LLM self-approve when the supervisor is offline.
Use a prompt saying “be conservative” instead of checking the threshold.
Pause in a pending-approval state and commit the credit only after verified supervisor signoff.

Explanation

Correct Answer

A controlled state transition keeps financial authority separate from agent-generated proposals.

Incorrect Answers

Post-commit labels cannot substitute for prior approval.

Models do not possess delegated supervisor authority.

Language caution is not an enforceable transaction limit.

Question 67

An international enterprise has many AI pilots but no policy for classifying model data. Which Cloud Adoption Framework transition is needed before fleet-wide scale?

Adopt a single unrestricted connector credential for every team.
Defer cost and risk ownership until after models have generated revenue.
Treat pilot subscriptions as automatically approved production zones.
Establish readiness, governance and secure landing-zone patterns with named service owners.

Explanation

Correct Answer

Production adoption requires governed environments, identity, security and cost controls rather than merely more pilots.

Incorrect Answers

Broad connector access weakens least privilege across business domains.

Unowned risk and spending make scaling financially and operationally unsafe.

Prototype hosting is not a documented enterprise operating model.

Question 68

A bank compares an account-document extraction agent with an executive brainstorming bot. The extraction process has 80,000 monthly documents and a measurable manual error rate. Which prioritization evidence matters?

Prioritize the brainstorming bot because its pilot scored higher on creativity, without measuring the extraction workflow’s current error-rework cost.
A quantified baseline and a risk-adjusted forecast of cycle-time and error savings.
Choose the extraction agent solely because it recognizes the highest number of document types in a demo, without testing audited processing outcomes.
Prioritize whichever pilot has the highest projected adoption before validating access controls or correction effort.

Explanation

Correct Answer

High-volume measurable work allows pilot value and control cost to be compared with a known baseline.

Incorrect Answers

A creativity score measures the brainstorming use case, not document-processing savings. The high-volume extraction baseline supplies a measurable error and labor opportunity.

Demo coverage is not a financial or risk-adjusted outcome. Selection needs verified throughput, exception and manual correction measures.

Adoption is one input, but cannot establish a safe benefit when error-rework cost and regulated data handling are unmeasured.

Question 69

A company already uses a deterministic approval flow for payroll changes, but employees want natural-language help assembling required documents. What approach balances automation and control?

Add a preparatory Copilot experience while keeping payroll changes in the authoritative approval workflow.
Let a conversational agent commit payroll changes from its own unverified summary.
Replace compliance approval records with the assistant’s conversation transcript.
Make document upload automatically change payroll without identity verification.

Explanation

Correct Answer

Natural-language assistance improves intake; payment and identity controls remain enforced by the existing transaction system.

Incorrect Answers

Free text is not sufficient authority to alter salary records.

A transcript cannot replace governed approval state.

An upload event alone does not verify an authorized change request.

Question 70

A claims pipeline has a document-reading agent and a fraud-triage agent. They disagree over whether the same case is already processed. What multi-agent design resolves that race?

Ask both agents to decide independently whether the other one already completed work.
Delete all processing logs after a case appears successful.
Let each agent generate a new case identifier during every retry.
Carry a unique case ID and shared state transition contract with idempotent processing.

Explanation

Correct Answer

Explicit shared identity and state transitions prevent duplicated actions and allow deterministic recovery.

Incorrect Answers

Model guesses cannot substitute for durable workflow state.

Removing logs hides errors and frustrates audit recovery.

New retry IDs make the same case look like new work.

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

Microsoft AB-100 Practice Test 5 - 10 Questions (Last Reviewed 09-Nov-2026)
Microsoft AB-100 Practice Test 6 - 10 Questions (Last Reviewed 08-Nov-2026)
Microsoft AB-100 Practice Test 7 - 10 Questions (Last Reviewed 07-Nov-2026)
Microsoft AB-100 Practice Test 8 - 10 Questions (Last Reviewed 06-Nov-2026)
Microsoft AB-100 Practice Test 9 - 10 Questions (Last Reviewed 05-Nov-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 60

Explanation

Correct Answer

Incorrect Answers

Question 61

Explanation

Correct Answer

Incorrect Answers

Question 62

Explanation

Correct Answer

Incorrect Answers

Question 63

Explanation

Correct Answer

Incorrect Answers

Question 64

Explanation

Correct Answer

Incorrect Answers

Question 65

Explanation

Correct Answer

Incorrect Answers

Question 66

Explanation

Correct Answer

Incorrect Answers

Question 67

Explanation

Correct Answer

Incorrect Answers

Question 68

Explanation

Correct Answer

Incorrect Answers

Question 69

Explanation

Correct Answer

Incorrect Answers

Question 70

Explanation

Correct Answer

Incorrect Answers

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