CAIC Technology Project Management 2 — Questions and Answers
Question 1: Which AI project risk is best mitigated by establishing a model monitoring pipeline post-deployment?
- Scope creep during development
- Model performance drift over time (Correct answer)
- Inadequate stakeholder buy-in
- Underestimated initial training costs
Correct answer: Model performance drift over time
Model drift occurs when real-world data distributions shift, degrading accuracy, so continuous monitoring detects and triggers retraining.
Question 2: In an AI project, a 'proof of concept' (PoC) phase primarily serves to:
- Finalize production infrastructure
- Validate technical feasibility before full investment (Correct answer)
- Train the final production model
- Complete regulatory compliance documentation
Correct answer: Validate technical feasibility before full investment
A PoC tests whether the AI approach can solve the problem at small scale before committing full resources.
Question 3: A project manager notices that AI model training jobs are consuming 3x the estimated GPU hours. The BEST immediate action is to:
- Cancel the project and reassess scope
- Profile training bottlenecks and optimize hyperparameters or data pipelines (Correct answer)
- Switch to a simpler rule-based system
- Request unlimited budget from sponsors
Correct answer: Profile training bottlenecks and optimize hyperparameters or data pipelines
Profiling identifies the root cause of compute overrun so targeted optimizations can bring costs back in line.
Question 4: Which methodology is MOST commonly adapted for AI/ML projects due to its iterative experimentation cycles?
- Waterfall
- PRINCE2
- Agile/Scrum (Correct answer)
- Critical Path Method (CPM)
Correct answer: Agile/Scrum
Agile's short sprints align with the experimental, iterative nature of model development and evaluation.
Question 5: When managing a cross-functional AI team, which communication artifact best keeps data scientists, engineers, and business stakeholders aligned?
- A model card documenting model capabilities and limitations (Correct answer)
- A detailed mathematical derivation of the loss function
- A raw dataset schema dictionary
- A vendor hardware comparison spreadsheet
Correct answer: A model card documenting model capabilities and limitations
Model cards provide a standardized, accessible summary that bridges technical and business audiences.
Question 6: A stakeholder requests a new AI feature mid-sprint. Following agile best practices, the project manager should:
- Immediately halt the sprint and implement the feature
- Add it to the product backlog and prioritize it for a future sprint (Correct answer)
- Reject the request permanently as out of scope
- Implement it without telling the team to avoid disruption
Correct answer: Add it to the product backlog and prioritize it for a future sprint
Agile practice protects sprint goals while capturing new requests in the backlog for proper prioritization.
Question 7: Which metric is MOST useful for tracking whether an AI project's iterative model improvements are delivering business value?
- Number of commits to the code repository
- Model accuracy improvement vs. baseline business KPI change (Correct answer)
- Total lines of training code written
- Number of features in the feature store
Correct answer: Model accuracy improvement vs. baseline business KPI change
Linking model accuracy gains to business KPI movement confirms technical progress translates to real-world value.
CAIC Technology Project Management 2
Which AI project risk is best mitigated by establishing a model monitoring pipeline post-deployment?