Straight exam-prep advice today, for anyone with a date on the calendar. First. Study by domain weight, not by comfort. The published split (ECO September 2025, p.5): Domain I 15%, Domain II 26%, Domain III 26%, Domain IV 16%, Domain V 17%. Domains II and III together are 52% of the exam, decided before a model ever exists. If your remaining study hours are going mostly into model development, you are polishing the domain worth the least. Second. Every task title has a verb, and the verb is the trap. Oversee, verify, manage, convey, coordinate. Almost none of them are build or code. When an answer option has the AI project professional doing the technical work personally instead of overseeing the person who does, that option is usually wrong, however correct the technical content sounds. Third. Learn to name the artefact, not just the phase. For any situational item, ask what document or deliverable should already exist at this point in the story: a deployment plan, a documented fallback, a go/no-go decision recorded on evaluation evidence, a handover with named owners. If the scenario describes something missing, the task that should have produced that missing artefact is usually your answer. Fourth. Know where the six go/no-go gates sit, and which direction a No-Go sends you. A No-Go does not end the project. It returns you to the phase that owns the missing input. The exam tests whether you know which phase that is, not whether you know the concept exists. Fifth, exam-day mechanics, verified against the ECO. 120 questions, 100 scored. The other 20 are unscored pretest items placed randomly, and you cannot tell which is which, so treat every question the same. 160 minutes total, no scheduled breaks, roughly 80 seconds per question at a flat pace. The 15-minute tutorial and the closing survey do not count against that time. If a situational item is past 90 seconds and you are still rereading the stem, mark it and move on. Come back at the end with whatever time is left.

A logistics company deployed a route-optimisation AI eleven weeks ago. Four separate things land on the AI project professional's desk in the same week. Situation one. The monthly stakeholder report shows that on-time delivery, the business KPI agreed before the project started, has slipped for the second month running. Technical accuracy on the model dashboard has not moved. Situation two. The vendor has released a new version of the underlying model and the team wants to swap it in on Friday. Nobody can currently say which version produced the routes customers complained about last month. Situation three. The system was unavailable for six hours during a regional outage. Dispatchers had no documented way to fall back to manual routing, so they improvised, and two depots stopped dispatching entirely. Situation four. The operations team that now owns the system can read the dashboards but has never been shown how to trigger a retrain. The project team was released three weeks ago. Two questions. Q1. Domain V has seven tasks. Which task owns each of the four situations? Name the task and give one sentence of reasoning each, not just a number. Q2, part one. In the post-mortem the head of delivery says: "All four are Domain V failures, and the decision to deploy was Domain V too, since Domain V is the deployment domain." One half of that sentence is wrong. Which half, which domain and task actually owned the decision to deploy, and what evidence did that decision require? Q2, part two. Take situation one on its own. Accuracy has not moved and the business KPI has fallen. Which phase defined the number that is now failing, and why is "the model is performing as designed" not an acceptable answer? Answer in the comments with one sentence of reasoning per item. That sentence matters more than the task number. I will post the CPMAI-aligned answer here in 24 hours. Try it before you read the other replies.