From PMP to PMI-CPMAI, what changes when the project is AI

On 16 October PMI put out an interview on its blog with a program manager who holds the PMP and later earned the PMI-CPMAI. Most of what follows comes from there, because he is one of the few who can compare the two from the inside.

The problem he describes is familiar to anyone. You hold the PMP and from one day to the next the work you get handed has artificial intelligence written into the brief. It is not a software project with a model inside, it is a project where nobody knows yet whether the thing will work, and that unsettles anyone who learned to plan with the scope closed before starting. PMI has a certification for that gap and it is called PMI-CPMAI.

Let’s start with what its own page says

The official page presents it as a bundle that includes the certification together with an exam prep course. The exam has 120 questions and runs for 160 minutes, and the header of the page says it requires no prior experience, which is a big difference from the years of experience the PMP does ask for. Keeping it takes 30 professional development units every three years, and both the course and the exam are available in English, among other languages.

The course runs 21 hours and grants 21 PDUs, split into 7 of Business Acumen, 11 of Ways of Working and 3 of Power Skills. If you already hold the PMP, the page itself says those 21 PDUs cover more than a third of what you need to renew it, and that changes the arithmetic, because the time comes out of the study budget you had already committed.

The page does not publish the price, so you will not find it here either.

What the program manager said

It is worth reading because it does not sell the certification with adjectives, it explains it through the difference he found at work.

The sentence that best captures the relationship is his. The PMP provides a broad project delivery foundation, and the PMI-CPMAI adapts and extends that foundation for the iterative, data-driven nature of AI initiatives. The gaps he names are three, data readiness, not knowing whether a model works until you try it, and the obligation to build in iterations.

There is also a line that makes anyone who memorised the PMP uncomfortable. It says it is fine to go backward in the process and that you do not have to move to the next step just because the PMP said so. It is not a criticism of the PMP, it is a warning that the reflex of advancing in order works against you when the outcome is not known yet. There is the question underneath it all. How much of what you already know how to do still holds when you cannot promise the outcome?

The six phases and where the PMP already took you halfway

The names of the six phases are those of the methodology published by PMI on the PMI-CPMAI page. The mark on the right is our reading, it does not come from the source.

The first phase asks you to match the AI solution to a real business need, assess feasibility, define the return and set the scope. You recognise that, it is the business case and the scope of always. The fourth asks you to iterate development and delivery, and the PMP outline already asks you to recommend a predictive, adaptive or hybrid approach. The fifth asks you to test and evaluate, which is quality under another name. The sixth asks you to put the system into operation, and there the PMP outline already asks you to validate that the transition to operations is ready.

The two phases without a mark are the second and the third, identifying the data needs and managing their preparation. Think of it as sowing. You can be impeccable with the plough and with the calendar, and if the seed is bad there is no harvest. That ground the PMP does not cover, and not by oversight, but because in an ordinary project data is one more input and here data is the project.

Data stops being a deliverable

The program manager puts it in a way that sticks. In the projects of a lifetime data is a deliverable. In AI projects data is the lifeblood of the project, and understanding where it comes from and why it has value becomes part of the project manager’s job, not something handed down to the technical team.

Think about the last project with models that passed near you. Who decided the data was ready, and on what grounds?

Look also at how you decide a project is over. With the PMP the milestone is delivery and then closure follows. With a model in production delivery closes nothing, because the model degrades and somebody has to answer for that six months from now. Who takes ownership once the original sponsor has already moved roles is the question the blog post leaves hanging.

Back to your case, when it makes sense and when it does not

It makes sense if you are already inside AI initiatives and you hear about the decisions instead of taking part in them, and it makes sense if your PMP renewal is near, because the PDUs from the course count all the same.

It makes no sense as a shortcut, because it does not replace the PMP or make it old and nobody at PMI has framed it that way. If you are preparing the PMP, finish that first, which is the base the other one leans on. And if somebody sells it to you as mandatory in order to keep working, be suspicious, because the official source presents it as a specialisation and not as a requirement.

To place it on the map of what PMI is putting out, you have what we already covered about its artificial intelligence standard, the advanced certification it announced and the road to the PMO, which is the other classic fork after the PMP.

The hardest part of this certification is not what it adds, it is what it asks you to let go of. The habit of promising the outcome before having seen it. Unlearning that costs more than learning the new part.

Join the waitlist

At The PM Architect we write the PMBOK study guides in five languages, with real classroom cases and exercises that come with an explanation. They will be free. Leave your email on the home page and we’ll let you know the moment they’re ready.