Question 450
A healthcare provider finished a chatbot proof of concept. Security assessment, incident ownership and business continuity tests are unfinished, but leaders request immediate rollout to all clinics. What should the adoption roadmap require?
- Roll out to all clinics and defer safety controls until model usage patterns are known.
- Treat user satisfaction in the pilot as equivalent to tested business continuity.
- Move the same exploratory credentials to production to avoid creating separate access policies.
- Complete defined governance, operational recovery and safety acceptance gates before phased clinic rollout.
Explanation
Correct Answer
D:A successful pilot establishes feasibility but does not replace production security, monitoring and recovery readiness.
Incorrect Answers
A:Widespread release would multiply the impact of unresolved privacy, reliability and support defects.
B:Positive feedback cannot prove operation during connectivity loss, downstream service errors or credential revocation.
C:Prototype permissions may be excessive or unmanaged and should not be carried into patient-facing production service.
Question 451
A city agency compares processing 40,000 standardized permit queries with automating a rare discretionary appeal. The permit FAQ is approved; the appeals process has disputed legal ownership. Which pilot should rank higher?
- The high-volume permit-query assistant with clear factual tests, approved sources and measured deflection outcomes.
- The appeals agent because it appears more autonomous during a prototype demonstration.
- Both at full production scale before either workload has an outcome baseline.
- Whichever project has the longest list of model features rather than process evidence.
Explanation
Correct Answer
A:It offers repeatable evidence of value and a defined source authority while the discretionary appeal lacks an accountable rule owner.
Incorrect Answers
B:Autonomy is not evidence of feasibility when legal review authority and measurable case volume remain unresolved.
C:Simultaneous deployment without acceptance criteria hides whether any cost or service improvement was achieved.
D:Capability count cannot substitute for data readiness, throughput, risk and measurable process value.
Question 452
A manufacturer can automate an occasional report that saves 20 minutes weekly or a defect-triage process that avoids costly rework on thousands of parts. What comparison should guide investment?
- Rank by which department submitted its proposal first.
- Calculate expected verified rework reduction and labor savings per process, adjusting for implementation risk and operating cost.
- Choose the report assistant because its demo response is shorter and more polished.
- Choose the process with the most model tokens generated per user interaction.
Explanation
Correct Answer
B:Different workloads must be compared through business impact, confidence in evidence and full deployment cost.
Incorrect Answers
A:Submission timing says nothing about feasibility or financial consequences of each automation candidate.
C:Narrative quality does not establish greater savings than preventing documented high-cost manufacturing defects.
D:Token consumption measures inference expense rather than value created or errors prevented.
Question 453
Two proposed agents each save an estimated 1,000 staff hours yearly. One summarizes internal meeting notes; the other recommends benefit denials without a reviewed appeal process. What should change the ranking?
- Prioritize the denial agent because a final decision is more autonomous than a meeting summary.
- Exclude any expected appeal and remediation expenses from the financial analysis.
- Apply risk-adjusted value that includes accountable human review, error severity and mitigation cost for denials.
- Treat the two as equal because the labor-saving numbers match.
Explanation
Correct Answer
C:Equal hour savings do not imply equal outcomes when one task can unlawfully affect a person’s benefits.
Incorrect Answers
A:Independence from human review can increase risk and is not itself a business advantage.
B:Leaving out downstream costs inflates the apparent savings of decisions that produce costly errors.
D:A common efficiency metric omits the denial workflow’s much higher accuracy, fairness and accountability requirements.
Question 454
A retailer estimates $2 million annual benefit from a replenishment agent, but item-level inventory is reconciled only once per month. A lower-value support agent already has reliable live data. What should the portfolio committee do?
- Fund the full replenishment rollout because projected benefits dominate every implementation constraint.
- Let the model generate missing interim stock movements from recent product trends.
- Discard the support candidate without recording any feasibility or risk comparison.
- Treat inventory freshness as a material dependency and fund a limited data-readiness validation before full replenishment rollout.
Explanation
Correct Answer
D:A large headline ROI is not reliable when the core stock input cannot support time-sensitive purchase decisions.
Incorrect Answers
A:The benefit estimate may be unattainable without timely stock data, creating costly or unauthorized orders.
B:Predictions cannot substitute for authoritative transaction events when executing inventory changes.
C:An investment decision should retain the alternatives and explain how data readiness and risk affected priority.
Question 455
Three departments propose agent pilots with incompatible success claims: one counts replies, one counts time saved, and one counts eliminated complaints. What should a central program impose before ranking?
- Document a baseline, business outcome denominator, measurement window and adverse-error metric for each use case.
- Rank initiatives solely by the greatest count of generated agent messages.
- Select the pilot with the highest satisfaction narrative written by its project sponsor.
- Require every agent to share one generic accuracy percentage without defining the evaluated population.
Explanation
Correct Answer
A:Comparable decision evidence requires explicit outcome definitions and risks rather than choosing the most flattering raw number.
Incorrect Answers
B:A message count is not comparable to labor saved or complaint reductions and may reward unnecessary interactions.
C:Promotional prose cannot establish reproducible results across dissimilar business processes.
D:A single score without denominators and harm severity cannot support portfolio tradeoffs across distinct tasks.
Question 456
A legal team wants AI to draft contract redlines while only licensed counsel may sign terms. The drafting service has no validated signing tool. Which solution boundary fits?
- Publish contract changes automatically whenever an agent generates fewer than three edits.
- Provide an assistive drafting Copilot with citations and leave execution and signature to the approved counsel workflow.
- Grant the drafting model direct signature rights because it can summarize the clauses accurately.
- Have the model add a textual “signed by counsel” footer before counsel reviews the terms.
Explanation
Correct Answer
B:Editing and explanation are assistive; the separate signature system retains the enforceable legal commitment boundary.
Incorrect Answers
A:Edit count does not establish approved legal risk or permission for contract execution.
C:Draft accuracy does not confer legal capacity or authenticated authority to bind the organization.
D:A generated signature statement is not evidence of consent from the authorized signatory.
Question 457
A service operations bot routinely classifies non-sensitive equipment tickets and assigns them to maintenance queues. The mapping is documented and reversible. When can an agentic flow be appropriate?
- Only if model instructions replace the ticketing service’s authenticated permissions.
- Only if it uses the largest available language model for every structured ticket.
- When its actions are limited to approved queue assignments, monitored against misroutes and reversible by staff.
- Only if the agent can permanently delete any incident it classifies as low priority.
Explanation
Correct Answer
C:A bounded low-impact workflow can benefit from autonomous orchestration while retaining measurable quality and recovery.
Incorrect Answers
A:Permissions must still be checked on each assignment regardless of the selected reasoning architecture.
B:Autonomy is a workflow design choice, not a requirement to use an expensive general-purpose model for simple routing.
D:Irreversible deletion exceeds the stated assignment scope and removes evidence needed to correct misclassification.
Question 458
An airline uses a deterministic form to reset a loyalty PIN, but irregular-disruption cases require collecting customer alternatives, comparing travel policy and proposing a rebooking. Which design is appropriate?
- Use a free-form autonomous agent to reset every PIN without verifying identity.
- Treat every rebooking proposal as a completed ticket before airline inventory confirms availability.
- Avoid all policy retrieval because a general model can infer the airline’s current disruption rules.
- Keep PIN reset deterministic and use a policy-constrained agent to plan options before a ticketing service confirms the rebooking.
Explanation
Correct Answer
D:The irregular case needs interpretation and multi-step planning, while the booking commit remains independently validated.
Incorrect Answers
A:The existing structured reset process already provides a stronger identity boundary than conversational inference.
B:A proposed itinerary is not a booked seat; the reservation service must confirm live inventory and permitted changes.
C:Current airline-specific policy terms require grounded evidence rather than an unsupported model assumption.
Question 459
An insurer wants an agent to rank suspicious claims for investigation. It may organize case evidence, but policy bars automated fraud accusations. What should the architect propose?
- An assistive risk-ranking workflow that exposes evidence and requires investigator validation before labeling or action.
- An autonomous fraud agent that suspends policies for every case above an arbitrary model threshold.
- A conversational bot that privately assigns guilt without retaining supporting evidence.
- Automatically erase low-score claims from the evidence archive to reduce storage.
Explanation
Correct Answer
A:A score may prioritize review but must not become an unverified allegation or enforcement outcome.
Incorrect Answers
B:A high score is not a validated fraud determination or legal authority to change coverage.
C:Opaque unreviewable classifications cannot satisfy the stated human review and accountability requirements.
D:Model predictions do not replace retention obligations or justify removing material from regulated records.
Question 460
A HR assistant currently drafts onboarding checklists. Management wants it to create real payroll records automatically, but the employee-ID system and approvals are not integrated. What is the appropriate expansion decision?
- Treat a successful checklist generation as proof that the payroll database write path is safe.
- Retain assistive drafting until typed employee identity, payroll authorization and verified commit workflows are implemented and tested.
- Let the model infer payroll IDs from employee names and issue records in the background.
- Add a prompt telling the assistant to be cautious, then immediately grant payroll administrator privileges.
Explanation
Correct Answer
B:Execution autonomy requires transaction-level controls that a checklist-only assistant never needed.
Incorrect Answers
A:Draft generation and durable payroll record creation are different capabilities with different failure and control risks.
C:Names may collide; inferred identifiers cannot establish the identity required for salary transactions.
D:Natural-language caution is not a credential, scope or approval enforcement layer.
Agentic AI Business Solutions Architect (AB-100) Practice Tests