Microsoft AB-100 Practice Test 5

Question 40

A distributor wants one agent to recommend shipment plans and another to reserve warehouse capacity. Reservations require exact units and idempotent tool calls. How should the architect divide them?

Let both agents independently reserve the same capacity without a shared transaction identifier.
Exchange only free-text promises between agents with no schema for units.
Allow the planning agent to bypass warehouse availability during peak demand.
Use a planning agent that proposes a typed reservation and a specialized execution agent/tool that validates capacity and authorization.

Explanation

Correct Answer

Separating reasoning from controlled execution avoids duplicated writes and supports typed handoff and compensating actions.

Incorrect Answers

Concurrent independent reservations can double-book limited resources.

Unstructured instructions can confuse units and account IDs.

A proposal should not be treated as validated inventory state.

Question 41

A Dynamics 365 service team needs a common FAQ agent whose content and supported actions are already provided by its business application. What is the most efficient solution choice?

Build a new foundation model even though the workflow and knowledge already fit a supported agent.
Build a new custom API before testing whether the prebuilt Customer Service agent already supports the required action and escalation controls.
Assess and configure the prebuilt app agent before approving a custom replacement.
Copy FAQs into every individual employee prompt and remove maintenance ownership.

Explanation

Correct Answer

Prebuilt capabilities can reduce cost and governance burden when they meet the tested requirement and permission model.

Incorrect Answers

Custom modeling creates training and support costs without a demonstrated gap.

Premature custom integration increases support cost and maintenance without first measuring a genuine capability gap.

Unmanaged prompt copies drift from approved answers.

Question 42

A Copilot Studio agent can look up warranty status and issue returns. The return action must never accept an unverified order ID. Which rule belongs in its tool design?

Require typed order identifiers and back-end policy validation before exposing the write operation.
Accept any free-text order number whenever the agent says it is plausible.
Put “only approved returns” in an instruction but omit transaction authorization checks.
Grant the tool service administrator rights to every order to avoid failed calls.

Explanation

Correct Answer

Tool schemas and back-end authorization define enforceable contracts; agent prose must not control entitlement.

Incorrect Answers

Model plausibility is not a verified order identity.

A prompt alone does not validate the real business transaction.

Broad rights violate least privilege and increase cross-account risk.

Question 43

An HR assistant must answer vacation-policy questions from an approved intranet page while avoiding outdated departmental drafts. What knowledge-source design fits?

Connect the published policy source with permission trimming and effective-date filtering.
Embed a superseded PDF into the system prompt so it is always visible.
Teach the model to fabricate policies when retrieval is sparse.
Ground on every document tagged vacation, including rejected drafts.

Explanation

Correct Answer

Retrieval must distinguish current approved sources from semantically similar draft material and preserve access rights.

Incorrect Answers

Hard-coding an obsolete policy prevents timely governance updates.

Fabrication is not an acceptable fallback for HR policy.

Keyword matches cannot prove publication authority.

Question 44

A sales team already works in Microsoft 365 Copilot and wants limited account insights from a governed CRM integration. When is extending Copilot preferable?

Only after retraining the base Microsoft 365 model on all tenant emails.
When the existing user experience can meet needs through a scoped extension and permission-aware data access.
Only if every CRM record is copied permanently into every user conversation.
Whenever the extension needs broader privileges than a caller possesses.

Explanation

Correct Answer

An extension can reuse the familiar interface and add authorized business context without a separate application.

Incorrect Answers

Base-model retraining is not required for connected, governed data access.

Permanent full-data copies cause exposure and freshness problems.

An extension cannot override service-side entitlements.

Question 45

A small clinic has ten labeled examples of a rare discharge code but thousands of approved policy paragraphs. A stakeholder proposes tuning a large custom model. What should be tested first?

Choose a larger model solely for its context window without evaluating grounded policy accuracy on representative discharge-code questions.
Assume ten labels prove a custom model will generalize safely.
Ground a suitable existing model in verified policies and evaluate whether a custom model is actually needed.
Train on unconsented patient records to increase sample size quickly.

Explanation

Correct Answer

Prompting and retrieval are lower-risk baselines; custom training needs adequate data, measurable benefit and governance.

Incorrect Answers

More context capacity does not prove task correctness and may simply expose more irrelevant patient policy material.

Tiny labeled sets rarely establish robust generalization.

Consent and protected-data limits remain mandatory during experimentation.

Question 46

An AI Center of Excellence maintains reusable prompts for finance, HR and procurement. What library feature most improves safe reuse?

A list of prompts ranked only by how often users click them.
A folder of anonymous prompts with no usage conditions.
One editable prompt shared anonymously among all departments.
Versioned templates with declared inputs, owners, test examples and review status.

Explanation

Correct Answer

Version control and acceptance examples make prompts auditable and limit silent drift between use cases.

Incorrect Answers

Popularity does not measure factual accuracy or authorized use.

Unowned instructions cannot be maintained when policy changes.

Shared uncontrolled editing obscures who introduced a risky change.

Question 47

A field-service device has strict offline compute limits and must classify a small finite set of fault labels. Which model strategy deserves comparison with a large hosted LLM?

An unrestricted cloud-only model despite the required offline operating mode.
A random large model with no on-device latency evaluation.
A rule that all compact models are inherently less accurate without benchmarking.
A task-specific smaller language model validated on field samples and resource budgets.

Explanation

Correct Answer

Bounded tasks may fit an optimized small model; evaluate accuracy, memory, latency and update requirements.

Incorrect Answers

An unavailable cloud dependency cannot satisfy an offline contract.

Model parameters alone do not establish usable performance on field hardware.

Performance must be measured for the actual task and devices.

Question 48

Customer-service agents often invent warranty durations when the knowledge base is silent. Which prompt standard should the architects adopt?

Tell staff to rely on model confidence scores without record-level citation.
Tune generation temperature to produce consistent-looking warranty terms without requiring a citation to a binding current policy.
Require cited approved policy evidence, clear uncertainty and escalation instead of unsupported answers.
Ask the agent to sound certain even when no source was found.

Explanation

Correct Answer

Grounding and explicit abstention constrain generation to evidence; users need a reliable exception path.

Incorrect Answers

Confidence scores do not establish source authorization or correctness.

Generation settings affect variability, not the authority or effective period of a warranty rule.

Confidence tone can disguise absence of supporting policy.

Question 49

An enterprise AI Center of Excellence must govern multiple Copilot Studio teams without writing every agent itself. Which operating model is appropriate?

Publish shared security patterns, review gates and reusable components while delegating approved domain ownership.
Allow each team to invent unreviewed connector permissions.
Make the center the only author of every prompt and eliminate business subject experts.
Standardize each agent’s prompt, interface and connector naming conventions while leaving permission policy and release accountability entirely with individual makers.

Explanation

Correct Answer

A center of excellence sets cross-team standards and oversight while domain owners provide relevant business expertise.

Incorrect Answers

Uncontrolled integrations create inconsistent data and identity risk.

Central bottlenecks can block legitimate improvements and discard essential domain input.

Consistent names and prompts do not enforce minimum security or release gates across teams; a CoE needs shared controls and delegated accountable owners.

Question 50

A sales opportunity agent draws data from Dynamics 365 Sales and customer payment status from Dynamics 365 Finance. What design prevents contradictory customer identities?

Have the model guess which Finance entity corresponds to each sales contact.
Use governed identity mapping and versioned integration contracts between the applications.
Link customers by normalized account display name without an authoritative CRM-to-Finance identifier and currency-aware contract.
Copy Finance balances into a free-text prompt and ignore reconciliation.

Explanation

Correct Answer

Cross-app workflows require stable keys, ownership and tested mappings before AI conclusions can be trusted.

Incorrect Answers

Identity inference cannot replace authoritative business system joins.

Display names can collide and change; financial data integration needs stable keys and explicit currency semantics.

Unreconciled prompt copies can misstate receivables.

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

Microsoft AB-100 Practice Test 3 - 10 Questions (Last Reviewed 11-Nov-2026)
Microsoft AB-100 Practice Test 4 - 10 Questions (Last Reviewed 10-Nov-2026)
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)

Pass Your Microsoft AB-100 Exam Easy!

Question 40

Explanation

Correct Answer

Incorrect Answers

Question 41

Explanation

Correct Answer

Incorrect Answers

Question 42

Explanation

Correct Answer

Incorrect Answers

Question 43

Explanation

Correct Answer

Incorrect Answers

Question 44

Explanation

Correct Answer

Incorrect Answers

Question 45

Explanation

Correct Answer

Incorrect Answers

Question 46

Explanation

Correct Answer

Incorrect Answers

Question 47

Explanation

Correct Answer

Incorrect Answers

Question 48

Explanation

Correct Answer

Incorrect Answers

Question 49

Explanation

Correct Answer

Incorrect Answers

Question 50

Explanation

Correct Answer

Incorrect Answers

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