CertPrepNow
PMIPMI-CPMAI71 concepts

PMI-CPMAI Cheat Sheet

Quick reference for the PMI Certified Professional in Managing AI exam.

CPMAI Six Phases (the Methodology Backbone)

1. Business Understanding
Clarify the problem, stakeholder needs, and business question before deciding whether AI is the right approach. Maps primarily to Domain II.
2. Data Understanding
Inventory, explore, and validate what data exists and whether it can support the proposed solution. Maps primarily to Domain III.
3. Data Preparation
Transform raw data into model-ready inputs: cleaning, feature engineering, normalization, augmentation. Maps primarily to Domain IV.
4. Modeling
Select algorithms, train models, tune hyperparameters, and manage experiment tracking. Maps primarily to Domain IV.
5. Evaluation
Test model performance against success criteria and make the model-readiness go/no-go decision. Maps primarily to Domain IV.
6. Operationalization
Deploy, govern, monitor, and maintain the AI solution in production. Maps primarily to Domain V. Replaces CRISP-DM's 'Deployment' phase name.
CPMAI is built on CRISP-DM
CRISP-DM's six phases (Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation, Deployment) map almost one-to-one to CPMAI, with 'Operationalization' replacing 'Deployment' and PM governance layered on top.
Phases are iterative, not linear
AI projects revisit and refine earlier phases as data shifts, goals evolve, or models underperform — a key difference from traditional waterfall project management.

Seven Patterns of AI

Hyperpersonalization
Uses ML to build an evolving individual profile for personalized content, product recommendations, or experiences that adapt over time.
Recognition
Identifies and classifies data — images, speech, text, objects — into meaningful categories (e.g., facial recognition, OCR).
Conversation / Human Interaction
Enables systems to interact with users through natural language — chatbots, virtual assistants, voice interfaces.
Predictive Analytics and Decision Support
Forecasts future outcomes and supports decision-making using historical and real-time data patterns.
Autonomous Systems
AI systems that operate and make decisions independently with minimal human intervention (e.g., robotics, self-driving systems).
Patterns and Anomalies
Detects normal patterns and flags deviations — fraud detection, network intrusion detection, predictive maintenance.
Goal-Driven Systems
AI designed to achieve specific organizational goals through optimization and planning, often combining multiple other patterns.
Why patterns matter (Domain II)
Every AI initiative should map to one or more of these seven patterns during problem identification — it's the vocabulary used to translate a business problem into an AI approach before scoping begins.

Domain I — Support Responsible and Trustworthy AI (15%)

Oversee privacy and security plan
Data governance for PII, encryption/access controls for training data, privacy impact assessments (PIAs) BEFORE deployment, GDPR/CCPA compliance, secure handling across the AI lifecycle.
Manage AI/ML transparency
Document model/data selection rationale, transparent reporting on data sources and preprocessing, explainability requirements, audit trails, interpretability tools.
Conduct bias checks (3 sources: data, algorithm, model)
Analyze training data for representation imbalance, perform fairness testing across population groups, implement bias detection metrics, review model outputs for discriminatory patterns, apply mitigation techniques.
Monitor regulatory and policy compliance
Track evolving AI regulations (GDPR, CCPA, EU AI Act, sector-specific rules), coordinate with legal/compliance teams, implement monitoring/reporting, maintain audit documentation.
Manage accountability documentation and audit trail
Comprehensive records of model development decisions, version control for models/data/training, stakeholder approval and go/no-go records, chain of custody for training/test data.
Transparency vs Explainability
Transparency = documenting WHAT was done and WHY (process). Explainability = making a SPECIFIC decision interpretable to a stakeholder (outcome). Frequently confused on the exam.

Domain II — Identify Business Needs and Solutions (26%)

Identify problem to be solved
Stakeholder interviews, process analysis for automation opportunities, define personas/use cases, map problems to the Seven Patterns of AI, validate with SMEs.
Evaluate initial AI feasibility
Assess technical viability, data availability/quality, compute resource requirements, organizational readiness, compare AI vs traditional solution alternatives.
Conduct risk assessment(s)
Security, safety, and ethics risks: failure modes, cybersecurity vulnerabilities, ethical implications, reputational/continuity risk, mitigation strategies. Separate task from feasibility.
Develop AI project scope statement
Project boundaries/deliverables, success criteria, in/out-of-scope functionality, assumptions/constraints, alignment with business objectives.
Determine ROI (include TCO)
Calculate expected benefits, estimate Total Cost of Ownership (infrastructure + maintenance, not just dev cost), business case, ROI metrics, cost-benefit analysis.
Manage adoption/integration risks
Organizational change management, user resistance/adoption barriers, integration with existing systems, training/communication strategy, adoption metrics monitoring.
Draft AI solution
High-level architecture, data flow/processing requirements, algorithmic approach, integration points, deployment/operational considerations. NOT the actual technical build — that's Domain IV.
Define success criteria (KPIs)
Measurable performance indicators, business impact metrics/thresholds, technical benchmarks, user satisfaction criteria, alignment with organizational objectives.
Support business case creation
Gather financial data/projected benefits, collaborate with finance, develop executive narratives, provide technical validation input.
Identify project resources
Skill requirements for team composition, hardware/infrastructure needs, contractor/consultant gaps, resource allocation timeline, procurement coordination.

Domain III — Identify Data Needs (26%)

Define required data
Data types/formats, volume requirements/sampling strategy, temporal/granularity requirements, quality standards/acceptance criteria, mapping to business objectives.
Identify data SMEs
Domain experts, business users who understand data context, data stewards/governance teams, technical experts — not just IT staff.
Identify data sources and locations
Internal databases/warehouses, external/third-party sources, cloud/distributed repositories, legacy systems/archives, document ownership/access permissions.
Coordinate AI workspace and infrastructure
Provision compute resources, secure development environments, storage/backup systems, collaboration/version-control tools, security/governance compliance.
Gather required data
Execute extraction, coordinate transfers/migrations, implement ongoing collection processes, validate completeness/accuracy, establish refresh procedures.
Check data privacy, compliance, and access
Verify usage rights/licensing, ensure regulatory compliance, implement access controls, conduct data-usage PIAs, document data lineage for audit.
Oversee data evaluation
Quality dimensions (accuracy, completeness, consistency), distribution/bias/gap analysis, freshness/relevance, schema compatibility, exploratory data analysis (EDA).
Determine if data meets solution needs (go/no-go)
Compare against requirements, assess sufficiency for robust training, identify gaps, validate REPRESENTATIVENESS (not just volume), make go/no-go decision.
Convey data understanding to leadership
Executive summaries, visualizations/reports, readiness status and recommendations, translate technical concepts to business language, progress updates.

Domain IV — Manage AI Model Development and Evaluation (16%)

Oversee AI/ML model technique(s)
Research/evaluate algorithms, guide supervised/unsupervised/reinforcement learning selection, assess complexity-performance-interpretability trade-offs, coordinate with data scientists.
Oversee AI/ML model QA/QC
Testing protocols, configuration management for model versions/parameters, performance monitoring during dev/test, peer reviews, coding standards. Pre-deployment activity.
Manage AI/ML model training
Training schedule/resource planning, monitor progress/compute utilization, coordinate hyperparameter tuning, oversee cross-validation, manage training data versioning/experiment tracking.
Manage data transformation (data preparation)
Oversee cleaning/preprocessing, coordinate feature engineering, manage normalization/standardization, supervise data augmentation, ensure transformation reproducibility.
Verify data quality for go/no-go
Final data quality checks BEFORE training, validate preprocessing results, assess representativeness/bias, data-readiness decision. Distinct from the model-readiness gate below.
Verify model ready for operationalization (go/no-go)
Performance vs success criteria, robustness/generalization, deployment readiness/infrastructure, documentation validation, final approval decision. Happens AFTER training/evaluation.
PM oversees, does not build
The AI project manager coordinates and governs technical work in this domain — the data scientists write the code and train the models.

Domain V — Operationalize AI Solution (17%)

Manage creation of AI solution deployment plan
Deployment strategy/timeline, infrastructure/resource planning, IT coordination, ROLLBACK procedures and contingency plans, deployment checklists.
Manage AI solution deployment
Coordinate deployment across technical teams, monitor progress/resolve issues, validate production functionality, manage user access/security config, post-deployment verification.
Oversee model governance
Model lifecycle management, versioning/change control, performance and DRIFT DETECTION monitoring, retraining coordination, governance policy compliance. Ongoing, post-deployment.
Oversee AI solution metrics
Monitoring dashboards for business/technical metrics, KPI tracking, performance trend/degradation analysis, stakeholder reports, alerting for threshold breaches.
Prepare final report/lessons learned
Document outcomes vs objectives, capture lessons learned/best practices, analyze what worked and what needs improvement, knowledge transfer docs, present results to stakeholders.
Manage AI solution transition plan
Plan handoff from project team to operational support, coordinate knowledge transfer, establish ongoing maintenance procedures, define roles/responsibilities, handover documentation.
Oversee AI solution contingency plan
Incident response procedures for AI failures, backup/disaster recovery, escalation procedures, business continuity plans, regularly test/validate contingency procedures.
Deployment vs Transition vs Contingency plans
Three SEPARATE Domain V deliverables: deployment plan = go-live strategy; transition plan = handoff to ops; contingency plan = incident/disaster response. Do not merge them.

Key Distinctions and Exam Traps

Feasibility vs Risk Assessment
Feasibility (Domain II Task 2) = CAN we build it (technical viability, data, compute). Risk assessment (Task 3) = what could go WRONG (security, safety, ethics). Separate tasks.
Data-readiness go/no-go vs Model-readiness go/no-go
Data readiness (Domain III Task 8 / Domain IV Task 5) gates entry into training. Model readiness (Domain IV Task 6) gates entry into operationalization. Two separate gates.
Model QA/QC vs Model Governance
QA/QC (Domain IV, pre-deployment) = testing and configuration management during development. Governance (Domain V, post-deployment) = ongoing lifecycle management, drift monitoring, retraining.
ROI vs TCO
ROI = benefit relative to investment. TCO = full cost including infrastructure, maintenance, retraining. A business case using only dev cost for ROI is incomplete.
Data Drift vs Concept Drift
Data drift = input feature distribution changes over time. Concept drift = the relationship between features and target changes. Both degrade deployed model performance and fall under Domain V governance.
CPMAI vs Traditional/Waterfall PM
Traditional PM is predictive and scope-driven. CPMAI is iterative and data-driven — phases loop back as data quality, goals, or model behavior change. AI projects fail more often under rigid waterfall management.
Bias check scope: data, algorithm, model
A bias check limited to training-data representation is incomplete. Fair data does not guarantee a fair algorithm or model output — all three must be checked (Domain I Task 3).

Exam Logistics and Program Facts

120 total questions
100 scored + 20 unscored pretest questions, randomly mixed throughout the exam and indistinguishable from each other.
160 minutes, no scheduled breaks
2 hours 40 minutes total. A tutorial (before) and survey (after) can each take up to 15 minutes but are not counted against testing time.
Pass/fail scoring, no published passing score
PMI does not disclose a numeric passing threshold for PMI-CPMAI. Only the 100 scored questions count toward your result.
Mandatory Exam Prep Course
Completion of the PMI-CPMAI Exam Prep Course (21-30 hours) is REQUIRED before you can schedule the exam. No experience or prior certification prerequisites otherwise.
$699 members / $899 non-members
Bundled price for the mandatory Exam Prep Course plus one exam attempt. PMI membership is not required to obtain the certification.
Pearson VUE delivery
Take the exam in-person at a Pearson VUE test center (PMI's recommended option) or online via OnVue remote proctoring.
Up to 3 attempts within 1 year
You have 12 months from purchase to earn the certification, with up to 3 total attempts. PMI suggests 30 days of additional study between retakes. Each retake requires its own fee.
30 PDUs every 3 years to maintain
PMI-CPMAI falls under the Continuing Certification Requirements (CCR) Program — earn 30 Professional Development Units every 3 years to keep the certification active.

Ready to test yourself?

Start a timed PMI-CPMAI mock exam or review practice questions by domain.