PMI-CPMAI COGNITIVE
PROJECT MANAGEMENT
FOR AI PRACTICE EXAM
COMPLETE 120
QUESTIONS
Question 1: An AI recommendation service must exchange data with an existing ERP system. What should be identi�ed during solution drafting?
Choices:
1) Integration points, interfaces, data �ows, and operational dependencies 2) Only the model's training loss 3) Only the project manager's certi�cation history 4) Only the cloud vendor's marketing roadmap Correct Answer: Integration points, interfaces, data �ows, and operational dependencies Explanation: A high-level AI solution should document how data �ows through the system and where the AI capability integrates with existing applications and operational processes.Page 1
Question 2: A warehouse-routing agent learns by receiving rewards for faster, safer routes and penalties for poor choices. Which approach best matches this training setup?
Choices:
1) Reinforcement learning 2) Unsupervised clustering 3) Ordinary linear regression only 4) Manual sorting
Correct Answer: Reinforcement learning
Explanation: Reinforcement learning trains an agent through feedback in the form of rewards or penalties tied to actions and outcomes.Question 3: Two model builds have the same name, but di�erent hyperparameters and preprocessing were used. Which QA/QC practice would best prevent confusion?
Choices:
1) Con�guration management and versioned experiment tracking 2) Deleting all intermediate runs 3) Using only verbal descriptions 4) Changing the project sponsor Correct Answer: Con�guration management and versioned experiment tracking Explanation: Con�guration management records the model, data, code, and parameter settings for each experiment so results can be reproduced and compared.Page 2
Question 4: During deployment, the new model service is healthy but the consuming application sends features in the wrong order. What should the team do?
Choices:
1) Pause or roll back as needed, correct the integration, and rerun production validation 2) Ignore the issue because the model endpoint is online 3) Retrain the model on the incorrect order 4) Remove all interface tests Correct Answer: Pause or roll back as needed, correct the integration, and rerun production validation Explanation: Deployment includes end-to-end integration validation. Incorrect feature mapping can invalidate predictions, so the team should correct the interface and verify behavior before continuing.
Question 5: A sponsor wants to measure success only by model accuracy, but the
business objective is to reduce claim-handling time. What should the project manager add?
Choices:
1) A business KPI tied to claim-handling time 2) A second accuracy metric with the same meaning 3) A larger training dataset regardless of need 4) A new programming language
Correct Answer: A business KPI tied to claim-handling time
Explanation: Technical model metrics should align with business success metrics. If the objective is faster claim handling, the project needs a KPI that measures the actual process outcome.Page 3