Most AI projects that fail do not fail because the model was not accurate enough. They fail because nobody defined what business outcome the model was meant to move, or because the data turned out to be unusable three months in, or because the thing was scoped as a single delivery when it needed to be six iterations, or because it was built and then never operationalised. Those are project management failures. The interesting claim behind CPMAI is that AI delivery needs its own methodology rather than borrowing agile wholesale, and having watched a reasonable number of these projects, I think the claim is correct.
PMI acquired Cognilytica in 2024 and with it the CPMAI methodology, which has clear lineage from CRISP-DM, the data mining process model that has quietly outlived most of its contemporaries. The methodology runs in phases covering business understanding, data understanding, data preparation, model development, evaluation and operationalisation, and the crucial structural point is that it is iterative rather than a waterfall with new labels. Each phase can send you backwards, and the framework expects it. The other crucial point is that data understanding and data preparation come before model development and are treated as first class phases rather than a task somebody does in the background.
Anyone who has watched a project discover in month four that the historical data has a different schema before 2022 will recognise why that ordering matters. So the intellectual content is sound. My problems are almost entirely commercial. The one that bothers me most is that you cannot sit the exam without first completing PMI's own exam prep course.
There are no other prerequisites at all, no experience requirement, no prior certification, nothing. So the only gate on this credential is a purchase. That is the wrong gate. It means the certification does not attest to any prior competence, only to having paid for training and then passed a test about that training, and it means PMI has arranged things so that everyone who wants the credential must buy two products.
I would have far more respect for a version of this that required three years of delivery experience and let you prepare however you liked. The cost follows from that structure. The exam sits somewhere in the $600 to $800 range depending on membership status, and the mandatory prep course ranges from several hundred to well over a thousand depending on format and provider. Realistically you are looking at four figures, and at the upper end you are approaching the cost of a decent conference plus a year of membership.
For an employer funded certification in an organisation with a serious AI delivery portfolio, that is defensible. For an individual paying out of pocket to improve their prospects, it is a lot of money for a credential that is not yet widely asked for by name. The PDU arithmetic is the part that changes the calculation for a specific group. If you hold a PMP, the prep course yields 21 PDUs, which is over a third of what you need for a renewal cycle.
Framed that way, a chunk of the cost is money you were going to spend on continuing education anyway, and you get a second credential out of it rather than a certificate of attendance. If you are a PMP holder in an organisation doing AI work, that reframing is worth doing properly with real numbers before you dismiss the price. The exam itself is 120 questions in 160 minutes, which is 80 seconds a question and comfortable enough. The questions are scenario oriented, which suits people who have actually run delivery and disadvantages people who have memorised the phase names.
Two retakes are permitted. Nothing about the format is objectionable. Maintenance is 30 PDUs every three years, which stacks with whatever your existing credentials demand, and it is worth being clear eyed that certifications are a subscription rather than a purchase. The transition from Cognilytica to PMI is still visible.
Materials in places read as if they were written for a different audience under a different brand, and the integration into PMI's broader ecosystem is not yet seamless. That will improve. It does mean early holders are paying full price for a product that is not finished. My 3.6 splits the difference between a methodology I genuinely rate and a commercial structure I do not.
If you want the substance without the certificate, the CPMAI phases are documented publicly and a competent delivery lead can read about them in an afternoon and apply them on Monday, and that path costs nothing. The certificate is worth buying when somebody else is paying, when you need the PDUs regardless, or when your organisation is standardising on PMI vocabulary and you want to be fluent in it. Outside those cases, take the ideas and keep the money.