CertArea
All PMI exams

Free PMI and Prosci AI adoption certification (pilot exam) practice test: 30 questions with answers

30 original questions for PMI and Prosci AI adoption certification (pilot exam), drawn from every topic area. Answer each one to see whether you got it right and why. No sign-up.

Free questions
30
Full bank
112 questions
Real exam
110 questions, 120 min

PMI and Prosci AI adoption certification (pilot exam) exam guide: topics, cost and format

All 30 PMI and Prosci AI adoption certification (pilot exam) questions with answers

The same questions as the test above, grouped by exam topic. Open any question to see the answer and explanation.

Define AI Success (5 questions)

  1. Question 1

    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 explanation
    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 2

    Which statement best describes a baseline in AI adoption success measurement?

    1. AA benchmark score published by a vendor so that different AI models can be compared with each other.
    2. BA forecast of future benefits prepared by the finance team for the investment case presented to executives.
    3. CA measured starting value of a business metric, captured before the AI is introduced, that later results can be compared against.
    4. DThe minimum accuracy a model must reach before it is allowed into production use for live decisions that affect customers and staff members.
    Show answer and explanation
    Correct answer: C

    A baseline is the measured starting point of a business metric taken before the change, so later results can show improvement. Option A describes a vendor benchmark, option B describes a forecast, and option D describes an acceptance threshold; none of these is a pre-change measurement.

  3. Question 3

    A sales team pilots an AI tool that drafts follow-up emails. After a full quarter, users say the tool is helpful and rate it highly, but leaders cannot show any change in win rate or sales cycle length. What is the most likely gap in how success was defined?

    1. AThe pilot used too small a group of users to generate any feedback that leaders could read or act on at all.
    2. BThe success measures relied on user satisfaction rather than on the business outcomes the initiative was meant to improve.
    3. CThe vendor's model should have been replaced with a model that scores higher on a generic language benchmark test.
    4. DThe drafting feature should have been paused until the legal team had approved every email template in the library.
    Show answer and explanation
    Correct answer: B

    The scenario shows positive satisfaction but no movement in the outcomes that matter, which points to success measures that were not linked to business results. Option A contradicts the scenario because users did give feedback, and options C and D address model choice and legal review, which the scenario does not raise.

  4. Question 4

    Which two elements should be defined before an AI initiative is funded, so that its success can be judged later? (Choose two.)

    1. AThe specific business outcomes the AI is expected to influence, and how each one will be measured.
    2. BThe number of graphics processors the organization will purchase for the training environment.
    3. CTarget values for the agreed measures, and the point in time by which those targets should be reached.
    4. DThe programming language preferred by the vendor's developers for building the model interface.
    5. EThe vendor's list of marketing claims about the model's capabilities compared with its competitors.
    Show answer and explanation
    Correct answer: A, C

    Judging success later requires defined outcomes, measures, targets and a timeline, which options A and C provide. Options B, D and E cover infrastructure, tooling and vendor marketing, none of which states whether the initiative achieved its purpose.

  5. Question 5

    What is the main purpose of a leading indicator in an AI initiative?

    1. ATo report the final financial return once the investment has fully matured and all of its costs are known.
    2. BTo prove that the training data has been fully cleansed of errors and bias before the model is used anywhere.
    3. CTo assign accountability for any budget overrun to a single named sponsor recorded in the governance charter.
    4. DTo show early signs of whether adoption and behavior are moving toward the intended outcome before lagging results appear.
    Show answer and explanation
    Correct answer: D

    A leading indicator gives an early signal of whether behavior and adoption are heading toward the target, while lagging indicators confirm final results. Option A describes a lagging measure of return, and options B and C describe data quality and accountability, which are not the purpose of a leading indicator.

Integrating Technical Delivery and Change Management (5 questions)

  1. Question 6

    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 explanation
    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.

  2. Question 7

    In phased AI deployment, what does a pilot usually involve?

    1. AA permanent release to all users, with every feature enabled from the first day of operation across the whole organization.
    2. BA limited, controlled use of the solution with a defined user group, to test the technology and the change approach before wider rollout.
    3. CA procurement step in which the vendor is chosen through a formal competitive tender process with several bidders.
    4. DA governance meeting in which the sponsor approves the final budget for the full program and its ongoing operating costs.
    Show answer and explanation
    Correct answer: B

    A pilot is a limited, controlled use with a defined group that tests both the technology and the change approach before wider rollout. Options A and C describe full release and procurement, and option D describes a funding decision, so none of them is a pilot.

  3. Question 8

    A bank automated its model retraining pipeline, but business owners were not told when the model's outputs would change. After an update, call center staff saw recommendations that looked different from what their training had shown. Which practice was missing?

    1. ACoordinating model release changes with change management, including communication to users and updates to training and procedures.
    2. BReplacing the model with a version from a different vendor, so that no future updates will change the recommendations staff see.
    3. CAdding more graphics processors to production inference so that retraining and redeployment cycles can run considerably faster.
    4. DShrinking the training dataset so that the model can be updated more often, without the update process needing any coordination.
    Show answer and explanation
    Correct answer: A

    Model updates change what users see, so release changes should be coordinated with communication, training and procedures. Option B avoids the issue rather than managing it, while options C and D address speed of retraining, which was not the problem in the scenario.

  4. Question 9

    Which three practices help integrate technical delivery with change management for an AI solution? (Choose three.)

    1. AInvolving end users in testing and capturing their feedback before the solution is released to the wider population.
    2. BAligning the training plan with the release schedule and with each new version of the model as it is deployed.
    3. CTreating the model as finished at go-live, with no monitoring plan or retraining schedule in place for later use.
    4. DDefining process changes, roles and new responsibilities alongside the technical design work, not after it is complete.
    5. EKeeping the upcoming change from frontline staff until the technical team has signed off on the build of the solution.
    Show answer and explanation
    Correct answer: A, B, D

    Involving users in testing, aligning training with each release, and defining process and role changes alongside the design all integrate delivery with change. Options C and E work against integration, since they ignore monitoring and hide the change from the people who must adopt it.

  5. Question 10

    In AI operations, what term describes the gradual loss of predictive accuracy that happens when live data shifts away from the data the model was trained on?

    1. AOverfitting, where the model memorizes noise and incidental details in its training examples instead of general patterns.
    2. BFeature scaling, where numeric inputs are transformed to a common range before the model is trained on the prepared dataset.
    3. CModel drift, where the relationship between inputs and outcomes changes after deployment and the model's accuracy declines.
    4. DData labeling, where human reviewers attach the correct outcome to each training example before the model is trained on it.
    Show answer and explanation
    Correct answer: C

    Model drift describes accuracy declining after deployment as live data or input-outcome relationships change. Overfitting happens during training when the model memorizes noise, while feature scaling and data labeling are preparation steps, not descriptions of accuracy loss over time.

Organizational Alignment and Engagement (5 questions)

  1. Question 11

    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 explanation
    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.

  2. Question 12

    In the Prosci ADKAR model, which element follows Awareness and concerns an individual's willingness to support and take part in the change?

    1. ADesire to participate in the change and support it, rather than resisting or waiting to see what happens.
    2. BKnowledge of how to use the new tools and processes that the change introduces into daily work.
    3. CAbility to perform the new tasks and behaviors that the change requires in day-to-day work over time.
    4. DReinforcement that sustains the change after the project has formally ended and the team has moved on.
    Show answer and explanation
    Correct answer: A

    In ADKAR, Desire follows Awareness and refers to the individual's willingness to support and participate in the change. Knowledge and Ability concern how to change and the skills to change, and Reinforcement concerns sustaining the change after it is made.

  3. Question 13

    A stakeholder map for an AI quality-inspection project places plant managers as high influence with low current support. Which engagement approach best fits this group?

    1. AKeep them informed through a monthly newsletter, since they are busy and have limited direct influence over the project's final outcome.
    2. BInvolve them closely in key decisions, address their specific concerns directly, and plan concrete steps that move them toward active support.
    3. CExclude them from the design work so that the project team can move faster without political delays slowing the schedule down.
    4. DMonitor their position passively and avoid direct contact until the project reaches go-live, so that the schedule stays protected.
    Show answer and explanation
    Correct answer: B

    High-influence stakeholders with low support need close involvement and direct attention to their concerns, with a plan to build support. Option A treats them as low priority, while options C and D exclude or ignore them, which tends to lower support further.

  4. Question 14

    Which two actions help align a group of senior executives behind an AI adoption goal? (Choose two.)

    1. AAgreeing on a shared statement of the business problem and the outcomes the AI is expected to influence over time.
    2. BAssigning each executive a separate AI tool to champion, without a common roadmap that links the tools together.
    3. CPublishing one consistent account of why the change matters and what it means for each function involved in it.
    4. DLetting each business unit define its own success measures, so that no leader feels constrained by shared goals.
    Show answer and explanation
    Correct answer: A, C

    A shared problem statement and one consistent narrative give executives a common direction, which options A and C provide. Options B and D fragment the effort, since separate tools and separate success measures pull executives in different directions.

  5. Question 15

    A retailer is preparing an AI assistant for store managers. Frontline staff are hearing rumors that the tool will replace jobs. The change team wants to address this. Which action is most effective?

    1. AAsk the communications team to publish a short notice that details about the tool stay confidential until the day of launch.
    2. BHold direct, two-way sessions with store staff to explain the tool's intended role and what changes for them, and invite their questions.
    3. CTell store managers to stop discussing the tool, so that the rumors fade away without the company needing to respond in any formal way.
    4. DShare a corporate video that praises the technology's capabilities without mentioning staff roles or the specific concerns being raised.
    Show answer and explanation
    Correct answer: B

    Two-way sessions that explain the tool's role and invite questions address rumors directly and build trust. Options A, C and D either withhold information, suppress discussion or avoid the concern, which usually lets rumors grow.

Assess and Validate (5 questions)

  1. Question 16

    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 explanation
    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.

  2. Question 17

    What is the main purpose of a holdout test set when validating a model?

    1. ATo supply the examples the model learns from during its training runs, so that it sees a broad range of cases during fitting.
    2. BTo estimate how well the model generalizes to data it did not see during training.
    3. CTo store production logs that are later used to bill the cloud provider for the inference usage recorded during the period.
    4. DTo replace the need for business review of the model's outputs, since the test set is assumed to represent what the business wants.
    Show answer and explanation
    Correct answer: B

    A holdout set is kept apart from training so it can estimate performance on unseen data. Option A describes the training set, options C and D describe billing and governance, and none of them describes evaluating generalization.

  3. Question 18

    An organization wants to know whether a new AI-assisted workflow actually reduces handling time compared with the current process. Which validation approach gives the strongest evidence of impact?

    1. AAsk team leaders to estimate, from memory, how much handling time the tool saved after using it for about a month in their teams.
    2. BRun a controlled comparison, such as an A/B test or matched groups, and measure handling time against the pre-change baseline.
    3. CCompare this year's average handling time with the vendor's published case-study claims for similar customers in other industries.
    4. DSurvey users on whether they feel the tool is faster, and report the percentage of positive responses as the main result.
    Show answer and explanation
    Correct answer: B

    A controlled comparison against a baseline isolates the effect of the workflow change. Options A and D rely on perception, and option C compares against a vendor's claims about other organizations, which cannot show impact in this one.

  4. Question 19

    Which two checks are part of validating an AI system for fairness and reliability before wide use? (Choose two.)

    1. AComparing error rates across relevant user groups to detect whether the system performs unequally for some of them.
    2. BReviewing only the average accuracy on the development dataset, which was also used when the model was tuned and selected.
    3. CTesting performance on edge cases and on data that reflects the conditions the system will face in real operation.
    4. DConfirming that the model was trained on the largest dataset the team could obtain, without examining how that data was collected.
    Show answer and explanation
    Correct answer: A, C

    Comparing error rates across relevant groups tests fairness, and testing edge cases and realistic data tests reliability. Options B and D rely on development-set accuracy or sheer data volume, which can hide unequal or unreliable performance.

  5. Question 20

    What is the difference between verification and validation when assessing an AI solution?

    1. AVerification checks that the system is built correctly against its specification, while validation checks that it meets real user and business needs.
    2. BVerification happens only after the system has gone live in production, while validation happens only during the early design phase of the project.
    3. CVerification measures the financial return of the solution, while validation measures the latency of individual model responses under load.
    4. DThey are two names for the same activity, and the terms are used interchangeably by different departments across the organization.
    Show answer and explanation
    Correct answer: A

    Verification asks whether the system was built right against its specification, and validation asks whether it is the right system for the users and the business. Option B reverses the timing, option C assigns the wrong measures to each term, and option D denies that the two activities differ.

Organizational Governance and Accountability (5 questions)

  1. Question 21

    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 explanation
    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.

  2. Question 22

    What is the primary purpose of a RACI matrix in AI governance?

    1. ATo rank AI models by accuracy and retire any model that falls below an agreed threshold on the test benchmark.
    2. BTo calculate the total cost of ownership for cloud compute across all AI projects in the organization's portfolio.
    3. CTo list the technical dependencies between data pipelines, feature stores and the model-serving services they feed.
    4. DTo clarify who is Responsible, Accountable, Consulted and Informed for each governance decision or activity.
    Show answer and explanation
    Correct answer: D

    A RACI matrix maps roles to decisions and activities so that responsibility and accountability are clear. Options A, B and C describe model ranking, cost analysis and technical dependency mapping, which are different tools with different purposes.

  3. Question 23

    An AI steering committee has approved a risk-tiering policy. Low-impact internal tools can be approved by a team lead, while tools that influence decisions about customers need review by the ethics and risk board. A team wants to launch a customer-facing chatbot that recommends loan terms and proposes treating it as low risk because it is only a chatbot. What should governance do?

    1. AApprove the chatbot through the team lead's normal sign-off, because the policy does not mention chatbots by name and the team has described the tool as low risk.
    2. BClassify the tool by its impact on customers and decisions rather than by its interface, and route it to the formal ethics and risk board review.
    3. CRemove the risk-tiering policy, because it slows down teams that are working on customer-facing products and creates unnecessary delay in launches.
    4. DLet the vendor certify the chatbot as low risk before the internal team decides whether the launch should go ahead at all in this quarter.
    Show answer and explanation
    Correct answer: B

    Risk tiers should follow the impact of a tool on customers and decisions, not its interface, so a loan-terms chatbot belongs in formal review. Option A relies on the team's self-assessment, option C removes a control, and option D hands the decision to the vendor.

  4. Question 24

    Which two controls help an organization keep meaningful human oversight of AI-supported decisions? (Choose two.)

    1. ARequiring a trained reviewer to confirm or override high-impact recommendations before they take effect on a person or process.
    2. BLogging the model's inputs and outputs along with the reviewer's final decision, so that each decision can be audited later.
    3. CLetting the model adjust its own decision thresholds automatically whenever its performance drifts away from the agreed baseline.
    4. DRemoving the reviewer step once the model's accuracy passes an agreed threshold, so that decisions can be made faster in operations.
    Show answer and explanation
    Correct answer: A, B

    Trained reviewers who can confirm or override decisions, combined with audit logs of inputs, outputs and final decisions, keep humans meaningfully in control. Options C and D remove human control, either by letting the model change its own thresholds or by dropping the reviewer step.

  5. Question 25

    Which statement best describes the role of a responsible AI or AI ethics board?

    1. AIt writes and maintains the production code for AI models and manages the release pipeline for new versions of each model.
    2. BIt negotiates vendor pricing and manages software license renewals for the organization's portfolio of AI tools and services.
    3. CIt sets and oversees policies on acceptable AI use, fairness, privacy and transparency, and reviews higher-risk use cases.
    4. DIt replaces the legal department in deciding whether a particular regulation applies to a product or to a specific use case.
    Show answer and explanation
    Correct answer: C

    A responsible AI board sets and oversees policies on acceptable use, fairness, privacy and transparency, and reviews higher-risk cases. Options A and B describe engineering and procurement work, and option D describes legal interpretation, which is not the board's role.

Realize and Sustain Value (5 questions)

  1. Question 26

    Eighteen months after an AI forecasting tool went live, forecast accuracy has improved, but planners have started returning to their spreadsheets. The sponsor asks how to sustain the benefits. What is the most appropriate action?

    1. ARetire the spreadsheets immediately and announce that all planners must use the forecasting tool from now on, with no further review of their needs or concerns.
    2. BReview adoption data to find out why planners are reverting, then reinforce the new process through coaching, updated procedures and recognition of good use.
    3. CReduce the support team to fund a new AI project, since the forecasting tool is considered to be operating without problems at this point in its life.
    4. DClose the project, since the accuracy target has been reached and the benefits are considered to be fully realized and therefore no longer need attention.
    Show answer and explanation
    Correct answer: B

    Reversion is a signal that adoption is slipping, so the right response is to diagnose the cause and reinforce the new way of working. Options A and C remove support or force compliance without diagnosis, and option D treats a one-time accuracy target as proof that benefits will continue.

  2. Question 27

    Which statement best describes benefits realization in an AI initiative?

    1. ACompleting the technical deployment, measured by the number of systems that are installed and configured successfully in the environment.
    2. BTracking whether the expected business benefits are achieved, and sustained, after the solution has been put into use by the business.
    3. CRetraining the model for the final time before it is retired from service and its data is archived securely under the retention policy.
    4. DA financial audit that confirms the project was delivered within the approved budget and on the schedule set at the start of the program.
    Show answer and explanation
    Correct answer: B

    Benefits realization is the tracking of expected business benefits after use, and sustaining them over time. Options A and D measure delivery rather than benefits, and option C describes model retirement, which is a different activity.

  3. Question 28

    A team disbands after an AI tool's launch, and no one is assigned to track the savings the tool was expected to deliver. A year later, the savings have quietly faded. What would have prevented this?

    1. ANaming a business owner for each expected benefit, with benefit metrics reported in routine business reviews after the project has closed.
    2. BKeeping the project team in place indefinitely, so that the team can continue to maintain the tool and fix any issues that arise over time.
    3. CMaking sure the vendor's support contract runs for the longest possible term, so that the vendor is accountable for the tool's ongoing results.
    4. DPublishing the expected savings in the annual report before launch, so that the figures become a public commitment that cannot be changed later.
    Show answer and explanation
    Correct answer: A

    Benefits fade when no one owns them after the project closes, so a named business owner and routine reporting keep them visible. Options B and C rely on the project team or the vendor, which does not own business benefits, and option D makes a public promise without any mechanism for tracking it.

  4. Question 29

    What is the main purpose of a post-implementation review?

    1. ATo evaluate whether the solution delivered its expected outcomes, capture lessons learned, and recommend follow-up actions for the business.
    2. BTo decide which vendor will be contracted for the next phase of the program, based on the performance of the current vendor's work.
    3. CTo retrain the model on new data so that its accuracy matches the level that was reported during the initial validation of the solution.
    4. DTo close the project's budget by approving all remaining invoices and releasing the unused contingency funds back to the finance function.
    Show answer and explanation
    Correct answer: A

    A post-implementation review checks outcomes against expectations, records lessons learned, and sets follow-up actions. Options B, C and D cover vendor selection, retraining and budget closure, which may follow from a review but are not its main purpose.

  5. Question 30

    Six months after launch, an AI scheduling tool shows lower overtime costs. However, a seasonal drop in demand also occurred in the same period. The sponsor wants to report the full overtime reduction as the tool's benefit. What should the benefits lead do?

    1. AReport the full reduction, since the tool was in use during the period and the sponsor's approval depends on showing the savings clearly in the report.
    2. BSeparate the tool's effect from the seasonal change, using a comparison group or an adjusted baseline, before reporting the benefit to stakeholders.
    3. CRemove the overtime measure from the benefits register, since external factors make any credible attribution to the tool impossible to establish at all.
    4. DWait two years before reporting any benefit, so that seasonal patterns have had enough time to average out in the organization's data.
    Show answer and explanation
    Correct answer: B

    Attribution requires separating the tool's effect from seasonal change, for example with a comparison group or an adjusted baseline. Option A overstates the benefit, option C gives up on measurement when it is feasible, and option D delays reporting without solving the attribution question.