Certified AI Product Development Specialist
For those taking AI product from hypothesis to production: feasibility and approach, data and privacy, quality evaluation and inference economics.
- Questions
- 50
- Duration
- 60 min
- Passing score
- 70%
- Investment
- $100
The pass mark is fixed: you pass with 70% correct answers, with no curve and no comparison with other candidates.
- Changed your mind? Ask for a full refund within 14 days of purchase, if the voucher has not been used.
- Your voucher never expires. Take the certification exam when you are ready.
- If you do not pass, your next attempt is half price for 90 days.
- The credential never expires and has no annual fee. You pay for the certification exam once.
Who this is for
You have already shipped an AI feature
And you want a verifiable credential for what you learned deciding scope, cost and risk in practice.
You decide on AI products
And you need to hold your ground on feasibility, inference budget and risk exposure without relying only on the technical team.
You are moving into AI products
And you want a clear reference for what to master, instead of a tool list that changes every quarter.
What you take with you
A permanent verification code
Anyone confirms your credential at pmcert.org/verify, without having to ask you for anything.
Approval that does not expire
A credential you earned does not lapse, asks for no annual renewal and charges no subscription.
LinkedIn and a public profile
Add it to your licenses and certifications section, or share the link to your profile.
A certificate with your score by domain
In PDF, showing how you performed in each area assessed.
What you need to know
Eight domains, drawn at random from a bank written for this certification exam. Read each one and see where you stand.
Discovery and opportunity framing
6 questions · 12% of the exam
When AI genuinely solves the problem and when it only adds cost. Telling what the user needs apart from what the model can deliver.
Feasibility and approach selection
7 questions · 14% of the exam
Whether the error is reversible, and what that changes in the scope. When to use retrieval, when rules are enough, and when to use no model at all.
Data, context and privacy
6 questions · 12% of the exam
What can come from the knowledge inside the model and what has to come from your own base. Access permission, data deletion and the contract with whoever produces the data.
Quality evaluation
9 questions · 18% of the exam
How to build an evaluation set that does not lie. Why 95% accuracy can be a useless number, and what to measure instead.
Experience under uncertainty
7 questions · 14% of the exam
What the interface does when the model gets it wrong. Confirmation, undo, and why user acceptance does not follow measured accuracy.
Inference economics
6 questions · 12% of the exam
Cost per call, a margin that does not improve with scale the way the rest of software does, and the point where a feature stops paying for itself.
Launch and monitoring
5 questions · 10% of the exam
How to release without breaking the trust of people already using the product. A switch to turn one feature off on its own, and what support needs to know beforehand.
Governance and responsible use
4 questions · 8% of the exam
Who approves what, according to the risk of the decision. And the path to contest it, for anyone affected by an automated call.
Test your knowledge with the Open Assessment
The Open Assessment has twenty questions at the level of the certification exam, with the result on the spot and a score for each domain. It is the honest way to decide whether you are ready.
Test your knowledgeSee what the questions are like
Two questions from the free open assessment. The certification exam uses different ones in the same format: a work situation and a decision.
A marketplace receives 40,000 new products per month. In 94% of cases the supplier sends a category code that maps directly to the internal taxonomy. The remaining 6% arrive without a code. The team wants to automate categorization. Which approach fits best?
- ATrain a classifier on the full product history and apply it to 100% of the input, guaranteeing a single categorization criterion.
- BUse a language model on the entire input, since it handles both the cases with a code and the cases without one.
- CApply the deterministic mapping to the 94% and use a model only on the 6% that arrive without a supplier code.
- DPostpone automation until suppliers send a code in 100% of cases, avoiding the maintenance of two distinct paths.
Show answer and explanation
Answer: C
The first question in AI product development is not which model to use, but which part of the problem already has a deterministic solution. When a reliable rule exists, it is cheaper, faster, easier to debug and introduces no variability between runs. The model should be reserved for the residual part the rule does not cover. Running a model over the 94% already solved by mapping adds inference cost, error risk and a source of inconsistency, with no corresponding gain. This pattern, solving the deterministic part with rules and the residual with a model, appears in most AI products that sustain production operations.
A company wants an internal assistant that answers questions about HR policies. Policies change every month, and every answer must indicate which policy supports the information. Which technical approach best meets these requirements?
- ARetrieval augmentation over the base of current policies, with the retrieved passage cited alongside the answer.
- BMonthly fine tuning of the model on the updated policy text, redoing the training on every change cycle.
- CA fixed prompt containing the full text of every policy, updated manually whenever there is a change.
- DA proprietary model trained from scratch on the company document corpus, guaranteeing full command of internal vocabulary.
Show answer and explanation
Answer: A
Two requirements decide this choice: the content changes often and the answer must be traceable to its source. Retrieval augmentation meets both because the knowledge lives outside the model, in a base that updates without retraining, and the retrieved passage can be shown as a citation. Fine tuning burns behavior and style into the model, not volatile facts, and retraining every month is expensive and slow. A fixed prompt with all the content hits context limits and cost per call as soon as the base grows. A proprietary model is the most expensive and least rewarding option in almost every corporate case. The practical rule: knowledge that changes goes to the retrieval base, behavior that repeats goes to fine tuning.
Practise the topic with a free tool
No sign-up tools on subjects of this certification, to apply at work what the certification exam assesses.
No course is required
The certification exam is the same for everyone, and how you prepare is your call: a course, a book, or the experience you already carry from the job.
- Changed your mind? Ask for a full refund within 14 days of purchase, if the voucher has not been used.
- Your voucher never expires. Take the certification exam when you are ready.
- If you do not pass, your next attempt is half price for 90 days.
- The credential never expires and has no annual fee. You pay for the certification exam once.
Your access to the certification exam arrives by email as soon as the payment clears, and it does not expire. Payment is handled by Stripe.
See who has already been certifiedSample credential
Certified AI Product Development Specialist
- Holder
- Emily Katherine Thompson
CAPD-S3M7-K42P-W
Frequently asked questions about the certification exam
›What is the certification exam like?
It is an online multiple-choice certification exam, taken in your own browser, with no scheduling and no exam centre. Questions describe work situations and ask for a decision, rather than testing memorised definitions. Some ask for more than one option, and the screen says how many to select. The number of questions, the duration and the passing score for each exam are on the certification page.
›What is the passing score?
The passing score varies by certification and is published on each certification exam page, alongside the number of questions and the duration. There is no curve and no comparison with other candidates: your result depends only on your answers. There is no partial credit on multiple-response questions either, where the whole set has to be right.
›Can I retake the certification exam? Do I have to wait?
You can retake it, and there is no waiting period and no limit on attempts. What each new attempt requires is a new voucher, because each voucher is worth one attempt. Questions are drawn afresh each time you open a certification exam, so a retake is not a repeat of the previous paper. If you do not pass, you receive a coupon to buy the new attempt at the same certification for half the price, valid for 90 days.
›Does the voucher expire?
A voucher bought on the site does not expire. Vouchers issued to an institution in bulk may carry a deadline set at issuance, and in that case the date is shown with the code. If you lost the email, the code is also in your account, under Certification exams, once your address is confirmed.
›Can I cancel my purchase and get a refund?
Yes, within 14 days of purchase, as long as the voucher has not been used. Write to [email protected] from the same email used in the purchase, saying you want the refund. We return the amount through the same payment method and cancel the voucher in the same act, so the code stops working. A voucher already used is not refundable, even within the 14 days, because the attempt is consumed when the certification exam is opened.
›Does the certification expire? Is there renewal or an annual fee?
No and no. A PMCert credential does not expire: there is no expiry date, no periodic retesting, no renewal fee and no annual dues to keep what has already been earned. You pay for the certification exam once. The only way a credential stops being valid is revocation, which is an administrative act with a recorded reason, not an automatic expiry.
Questions about the certification exam →All frequently asked questions →