PMI • PMI-CPMAI
PMI Certified Professional in Managing AI (PMI-CPMAI) Practice Test
Validates expertise in managing AI, machine learning, and cognitive technology projects using the CPMAI methodology. Covers the full AI project lifecycle from strategy and data management to responsible AI implementation.
Exam Details
Questions
843
Duration
160 minutes
Passing Score
Pass/Fail
Difficulty
ProfessionalLast Updated
Feb 2026
Topics Covered
AI FundamentalsCPMAI MethodologyMachine Learning ConceptsData for AIManaging AI ProjectsTrustworthy and Responsible AI
Exam Domain Breakdown
Support Responsible and Trustworthy AI Efforts15%
Identify Business Needs and Solutions26%
Identify Data Needs26%
Manage AI Model Development and Evaluation16%
Operationalize AI Solution17%
Exam Overview
The PMI Certified Professional in Managing AI (PMI-CPMAI)™ is PMI's flagship certification for professionals who manage, oversee, and deliver AI, machine learning, and cognitive technology projects. Launched by PMI in 2024 as the evolution of the CPMAI v7 credential, it establishes a globally recognized standard for applying the six-phase CPMAI methodology to the unique challenges of iterative, data-driven AI initiatives. The certification validates competency across the full AI project lifecycle — from identifying business needs and defining data requirements, to overseeing model development, deployment, and ongoing governance.
Unlike traditional project management certifications, PMI-CPMAI addresses the distinctive complexities of AI initiatives: managing evolving datasets, aligning data scientists with business stakeholders, navigating model uncertainty and bias, and implementing responsible AI governance in accordance with frameworks such as the EU AI Act. The credential is tool-agnostic and methodology-driven, confirming holders can bridge technical AI execution with strategic organizational impact in any industry context.
Who Should Take This Exam
PMI-CPMAI is designed for project managers, program managers, product owners, business analysts, data professionals, and technology consultants who are involved in planning or delivering AI and machine learning projects. It is equally relevant for those transitioning into AI-focused roles from traditional project management backgrounds, as well as technologists and data practitioners who want a structured management framework to complement their technical skills.
The certification suits professionals across industries — including financial services, healthcare, manufacturing, government, and consulting — who are tasked with leading digital transformation initiatives involving intelligent automation, predictive analytics, natural language processing, or other AI/ML technologies. No prior AI or project management experience is required to enroll, making it accessible to a wide range of career stages.
Prerequisites
PMI-CPMAI has no formal educational or experience prerequisites — no prior project management certifications, technical AI knowledge, or work experience is required to enroll or sit for the exam. This makes it one of the most accessible professional-level AI credentials available.
However, completion of the official PMI-CPMAI Exam Prep Course is mandatory before scheduling the exam. The course is a 21-hour, self-paced online program organized around the six CPMAI methodology phases, using scenario-based exercises, case studies, and a downloadable workbook. Professionals with a background in project management, data science, or business analysis will find the content more immediately applicable, but the course is designed to build the required knowledge from the ground up.
Exam Format
The PMI-CPMAI exam consists of 120 total questions, of which 100 are scored and 20 are unscored pre-test (pilot) questions used to validate future exam content — candidates cannot distinguish which questions are pre-test. The exam is 160 minutes long and is delivered via Pearson VUE, either at an authorized testing center or through online proctoring. Questions are scenario-based multiple-choice in single-best-answer format.
The exam is scored on a pass/fail basis with no numerical score or domain-level performance feedback provided. It is preceded by an optional tutorial and followed by a survey, each up to 15 minutes, which do not count against exam time. Candidates may attempt the exam up to three times within a 12-month eligibility window; PMI recommends a minimum 30-day preparation period between retake attempts. The exam is currently offered in English, with additional languages (Arabic, Brazilian Portuguese, French, German, Japanese, Korean, Simplified and Traditional Chinese, and Spanish) planned for January 2026. The exam fee is $699 for PMI members and $899 for non-members.
Skills Measured
- 1.Domain 1: Support Responsible and Trustworthy AI Efforts (15%) — Covers overseeing privacy and security plans, managing transparency in AI/ML systems, conducting bias evaluations, ensuring regulatory compliance (including frameworks such as the EU AI Act), and maintaining audit documentation for AI governance.
- 2.Domain 2: Identify Business Needs and Solutions (26%) — The largest domain; encompasses AI problem identification, feasibility and risk assessment, scope definition, ROI determination, adoption change management, solution design, success metric development, and stakeholder resource identification.
- 3.Domain 3: Identify Data Needs (26%) — Covers defining data requirements for AI/ML projects, identifying subject matter experts, locating and evaluating data sources, coordinating data infrastructure, gathering and verifying data compliance, evaluating datasets for quality and bias, and communicating data findings to leadership.
- 4.Domain 4: Manage AI Model Development and Evaluation (16%) — Addresses selection of appropriate AI/ML model techniques, quality assurance processes, model training oversight, data transformation, quality verification, and readiness assessments prior to deployment.
- 5.Domain 5: Operationalize AI Solution (17%) — Focuses on deployment planning and execution, model governance post-launch, ongoing performance monitoring, transition and contingency planning, and capturing and documenting lessons learned from AI project delivery.
Study Tips
- Complete the mandatory 21-hour PMI-CPMAI Exam Prep Course before scheduling the exam — it is structured around the five exam domains and is the primary official resource aligned to the Exam Content Outline (ECO).
- Download and study the official PMI-CPMAI Exam Content Outline (ECO) from pmi.org. It lists all tasks within each domain and is the authoritative blueprint for what is tested; use it to identify knowledge gaps.
- Use the official PMI-CPMAI Practice Exam (available at pmi.org/shop) — it is the only PMI-sanctioned practice test, built by subject matter experts and aligned to the ECO, and each question includes a detailed explanation.
- Focus significant study time on Domains 2 and 3 (Identify Business Needs and Identify Data Needs), which together account for 52% of the exam. Practice applying the CPMAI methodology phases to scenario-based AI project situations.
- Study responsible AI governance frameworks alongside the CPMAI methodology — Domain 1 questions test practical application of bias mitigation, transparency, privacy, and regulatory compliance in real AI project scenarios.
- Access the free PMI introduction course ('Free Introduction to Cognitive Project Management in AI') on pmi.org to preview CPMAI concepts before investing in the full prep course.
- Join PMI community forums and study groups focused on PMI-CPMAI to exchange scenario-based practice questions and reinforce application of CPMAI phases to real-world AI project challenges.
Career Benefits
PMI-CPMAI positions holders for roles at the intersection of AI strategy and project delivery, including AI Project Manager, AI Program Manager, Digital Transformation Lead, and AI Governance Consultant. As organizations accelerate AI adoption — with global AI spending projected to reach $632 billion by 2030 and over 19 million AI-related jobs expected — certified professionals who can structure and govern AI initiatives are in high demand. PMI describes the credential as the only professional certification focused specifically on project management of AI transformation, differentiating it from broader data science or general PM credentials such as PMP or CAPM.
Salary data for AI-focused project managers ranges from approximately $95,000 to $150,000 annually in the United States, with those in senior or consulting roles frequently exceeding this range. Independent analyses cite a 20–30% salary premium for AI-proficient project managers over traditional counterparts. The certification also earns holders 21 PDUs applicable toward maintaining other PMI certifications, and is maintained with 30 PDUs every three years — a relatively low renewal burden. Global demand is strongest in the United States, Canada, United Kingdom, Germany, Singapore, and India.
Sample Questions
Preview — answers shown5 sample questions with correct answers and explanations. Start a practice session to test yourself across all 843 questions.
1. A CPMAI Phase VI Operationalization team implements a canary deployment strategy to release a new recommendation model version. The canary receives 5% of production traffic while 95% continues using the existing model. After 48 hours, the canary model shows 12% lower click-through rate and 8% lower conversion rate compared to the existing model. What should the team do according to MLOps best practices? (Select one!)
APerform automated rollback to remove the canary model and investigate the performance degradation
BImmediately promote the canary to full production because 48 hours provides sufficient statistical evidence
CAdjust the traffic split to 50/50 to accelerate data collection and reach conclusions faster
DContinue the canary deployment for an additional week to collect more data before making decisions
Explanation
Automated rollback should be performed immediately when the canary demonstrates clear performance degradation with 12% lower click-through rate and 8% lower conversion rate. Canary deployments exist specifically to detect problems before full rollout, and these metrics indicate the new model performs worse than the existing model. Promoting to production would harm business metrics. Continuing the canary prolongs negative business impact without justification. Increasing traffic to 50/50 would amplify the negative impact. The team should rollback, investigate root causes, and address issues before attempting redeployment.
2. A CPMAI project has deployed a recommendation model to production. Three months later, the team observes that model accuracy on new data has declined from 89 percent to 76 percent, while input feature distributions remain similar to training data. Which type of drift is most likely occurring? (Select one!)
AData drift where input distributions have changed
BConcept drift where the relationship between inputs and outputs has changed
CPrediction drift where model outputs have shifted
DNo drift, only normal variance in performance
Explanation
Concept drift occurs when the underlying relationship between input features and the target variable changes over time, causing accuracy degradation even when input distributions remain stable. This commonly happens when real-world conditions change but the patterns learned during training no longer hold. Examples include changing customer preferences, evolving fraud tactics, or shifting market dynamics. Data drift involves changes in input feature distributions, which the scenario explicitly states are similar to training data. Prediction drift describes changes in model output distributions without necessarily affecting accuracy. The 13 percentage point accuracy decline is too large to attribute to normal variance and indicates systematic concept shift requiring model retraining.
3. A manufacturing company implements MLOps maturity progression from Google's Level 0 to Level 1. At Level 0, the data science team manually trains models using Jupyter notebooks, exports model files, and the engineering team manually deploys them monthly. To achieve Level 1 ML pipeline automation, which three components must be implemented? (Select three!)
Multiple correct answersAContinuous training pipeline with automated model validation
BCI/CD pipeline automation for multi-environment deployment
CFeature store with centralized feature engineering and versioning
DMetadata management tracking experiments, lineage, and artifacts
EAutomated testing framework for data schemas and model quality
FProduction metrics-driven retraining with performance monitoring
Explanation
Google MLOps Level 1 focuses on ML pipeline automation, not full CI/CD automation which is Level 2. Level 1 characteristics include continuous training pipelines with automated validation, feature stores for consistent feature engineering, and metadata management for tracking experiments and lineage. These enable automated model development while deployment may still be manual. CI/CD pipeline automation is the defining feature of Level 2. Automated testing frameworks are Level 2 capabilities. Production metrics-driven retraining is a Level 2 characteristic representing full automation. Level 1 automates the model development and training process but not necessarily deployment and monitoring automation.
4. A CPMAI project team is selecting evaluation metrics for a medical diagnosis model where failing to detect disease (false negative) has much higher cost than incorrectly flagging healthy patients (false positive). Which classification metric should the team prioritize? (Select one!)
ARecall to minimize false negative rate and catch all disease cases
BF1 Score to balance precision and recall equally
CAccuracy to optimize overall correct predictions
DPrecision to minimize false positive rate
Explanation
Recall (sensitivity or true positive rate) measures the proportion of actual positive cases correctly identified, minimizing false negatives. In medical diagnosis where missing disease has severe consequences, maximizing recall ensures the model catches the maximum number of disease cases even if it generates more false positives. Precision minimizes false positives but allows more false negatives, which is inappropriate when false negatives have high cost. Accuracy can be misleading in imbalanced datasets and does not prioritize false negative reduction. F1 Score balances precision and recall equally, which does not reflect the asymmetric cost structure where false negatives are far more costly.
5. A retail analytics team implements a customer lifetime value prediction model using historical purchase data. During Phase IV Model Development, the team splits data into 70% training, 15% validation, and 15% test sets. The model achieves 92% R-squared on training data, 89% on validation data, and 88% on test data. According to machine learning best practices, what does this performance pattern indicate? (Select one!)
ASevere overfitting requiring regularization because training accuracy significantly exceeds test accuracy
BUnderfitting requiring a more complex model because all accuracy metrics are too similar
CData leakage from test set into training requiring new data splits and retraining
DGood generalization with acceptable performance degradation from training to test sets
Explanation
The model demonstrates good generalization with training R-squared of 92% and test R-squared of 88%, showing only 4 percentage point degradation. This small, consistent decline from training through validation to test sets indicates the model learned generalizable patterns rather than memorizing training data. Severe overfitting would show much larger gaps such as 92% training and 60% test. Underfitting would show poor performance across all sets. Data leakage would typically show unrealistically high test performance exceeding training performance or test performance matching training too closely.
$7
One-time access to this exam
Full access to all 843 questions
Or $15/mo for all 201 exams
Detailed explanations
Free preview stays available