PMI and Prosci AI adoption certification (pilot exam)

Last checked November 10, 2026 against the official PMI exam page.

1. Question 1Define AI Success

   A regional insurer plans to give claims handlers a generative AI assistant. Before the tool is selected, the steering committee asks how success should be defined. Which approach is best?

   1. AMeasure success by the number of licenses purchased, the share of employees who log in during the first month, and the total number of prompts sent through the tool each week.
   2. BDefine measurable business outcomes, such as claim cycle time and error rates, capture baselines before rollout, and agree targets with the sponsor.
   3. CDeclare success when the model reaches high accuracy on a public benchmark dataset that the vendor has chosen for its marketing materials.
   4. DTreat success as completing the deployment on schedule and within the original budget, regardless of how staff actually use the tool.
   Show answer and explanationHide answer

   Correct answer: B

   Success should be tied to measurable business outcomes with baselines and agreed targets, which option B describes. Option A counts inputs and logins rather than results, option C relies on a vendor-chosen benchmark that may not reflect claims work, and option D ignores whether the tool creates any value.

2. Question 2Integrating Technical Delivery and Change Management

   A team has built a document-search model that performs well in testing. The change lead notes that end users were not involved in the design and the training plan is still a draft. The delivery lead wants to release on the planned date. What is the best recommendation?

   1. ARelease on the planned date, because strong performance in testing is the main driver of user adoption and the training can be arranged after the launch.
   2. BHold the release until the change plan, training and user involvement are ready, and link the technical release gate to those readiness criteria.
   3. CRelease to a small group with no training, on the reasoning that users will learn the tool quickly by trying it out in their everyday work.
   4. DDrop the change plan from the project, because it duplicates planning work that the delivery team has already completed for the technical release.
   Show answer and explanationHide answer

   Correct answer: B

   Technical readiness alone does not ensure adoption, so the release gate should include change readiness such as training and user involvement. Options A and C assume performance alone drives use, and option D removes the change work that the scenario shows is missing.

3. Question 3Organizational Alignment and Engagement

   A manufacturer plans to introduce an AI scheduling tool that changes shift patterns. Floor supervisors have not been consulted and are openly skeptical. The program lead wants adoption to move faster. What should the lead do first?

   1. ASend a mandatory training schedule so that supervisors learn the tool before they have time to form their own views on it.
   2. BCarry out a stakeholder analysis and involve supervisors early, so that their concerns shape how the scheduling change is designed.
   3. CMake use of the tool a condition of supervisors' annual performance reviews, so that compliance is enforced from the very first week.
   4. DWait until the tool is fully built and tested, then present the finished scheduling plan to supervisors at the next all-hands meeting.
   Show answer and explanationHide answer

   Correct answer: B

   Early stakeholder analysis and involvement address skepticism at its source and make the design more realistic. Options A and C try to force adoption without addressing concerns, and option D delays engagement until the plan is fixed, which leaves supervisors no real influence.

4. Question 4Assess and Validate

   A team evaluating a support-ticket classifier reports 97 percent accuracy. A reviewer notices that 95 percent of the test tickets belong to one category, and the model assigns that category to nearly every ticket. What is the best next step?

   1. AAccept the accuracy figure, since it exceeds the target the sponsor agreed to and the result is therefore sufficient for release.
   2. BReview per-class measures such as precision and recall, and check whether the test set reflects the real mix of incoming tickets.
   3. CIncrease the size of the test set by copying the same tickets, which reduces random error in the accuracy estimate for the model.
   4. DReport the overall accuracy figure to the board and proceed, because overall accuracy is the standard measure for every classification task in the organization.
   Show answer and explanationHide answer

   Correct answer: B

   When one class dominates, overall accuracy hides poor performance on the other classes, so per-class measures and a representative test set are needed. Options A and D rely on the misleading headline figure, and option C adds duplicate tickets without adding any new information.

5. Question 5Organizational Governance and Accountability

   A health system is deploying an AI tool that suggests discharge priorities for ward patients. The chief medical officer asks who is accountable if a recommendation contributes to a patient safety incident. Which arrangement best answers this?

   1. AThe software vendor, because its contract describes the model's output and its service terms place responsibility for results on the supplier.
   2. BThe AI tool itself, because it generated the recommendation and the audit log records the decision that was made by the system.
   3. CA named accountable executive and a clinical owner, set out in the governance charter with clear decision rights for approval and escalation.
   4. DWhichever clinician happened to accept the recommendation at the time, with no further assignment of responsibility needed in the charter or at the executive level.
   Show answer and explanationHide answer

   Correct answer: C

   Governance requires named accountable owners with defined decision rights, which option C provides. Option A shifts accountability to a vendor, option B attributes it to a tool that cannot be held accountable, and option D leaves responsibility to whoever happened to act at the time.