A health insurer wants to summarise clinical notes automatically to speed up claims adjudication. They are not training anything. They are licensing a vendor's pre-trained model and sending the notes to it through an API. The vendor's published benchmarks are strong. The delivery lead argues that Phases II and III do not apply here, because there is no training dataset to source and no features to engineer. Two questions. Q1. Is the delivery lead right? If not, what work do Phases II and III still carry when you own none of the training data? Q2. The vendor benchmark is not the evidence you need. Name one condition that makes this a no-go even if the model performs well in the pilot, and say which gate should have caught it. Answer in the comments with one sentence of reasoning. That sentence matters more than the phase number. I will post the CPMAI-aligned answer here in 24 hours. Try it before you read the other replies.

AI projects have three people who sound similar but do very different jobs. 1. The data engineer builds and maintains the pipelines. They move data from source systems into a place where it can actually be used. The data scientist explores that data, builds and tests the models, and looks for patterns that answer the business question. The AI project manager owns the CPMAI lifecycle end to end. They keep Phase I connected to Phase VI and make sure the business question never gets lost along the way. None of these three roles reports to the others by default. They depend on each other, and the gap between them is usually where AI projects stall. Small teams often mix these roles into one person. Large teams split them across departments. The job does not disappear either way. On the exam, expect questions that test whether you know which role owns which phase, not just what each role does day to day. You do not need to do all three jobs. You need to know how they connect. Which of these three roles is missing or weakest on your current AI projects?