Candidate handbook
CAPDS candidate handbook
Certified AI Product Development Specialist
Everything that applies to the CAPDS certification exam: what it covers, how the exam works, how it is scored and where to study.
Blueprint version 1, published on August 25, 2026. Study guide, edition 2026.1.
What the certification attests
For those taking AI product from hypothesis to production: feasibility and approach, data and privacy, quality evaluation and inference economics.
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.
Prerequisite
None. You can take the certification exam directly.
Certification exam format
- Questions
- 50
- Duration
- 60 min
- Pass mark
- 70%
- Languages
- English, Portuguese, Spanish
The pass mark is fixed: you pass with 70% correct answers, with no curve and no comparison with other candidates.
Domains and weights
Questions are drawn for every certification exam, always in this proportion. The weights are the same the exam engine uses.
| Domain | Questions | Weight | Where to study |
|---|---|---|---|
1. Discovery and opportunity framing When AI genuinely solves the problem and when it only adds cost. Telling what the user needs apart from what the model can deliver. | 6 | 12% | Guide CAPDS, chapter 4, Opportunity discovery and framing |
2. Feasibility and approach selection 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. | 7 | 14% | Guide CAPDS, chapter 5, Feasibility and choice of approach |
3. Data, context and privacy 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. | 6 | 12% | Guide CAPDS, chapter 6, Data, context and privacy |
4. Quality evaluation How to build an evaluation set that does not lie. Why 95% accuracy can be a useless number, and what to measure instead. | 9 | 18% | Guide CAPDS, chapter 7, Quality evaluation |
5. Experience under uncertainty What the interface does when the model gets it wrong. Confirmation, undo, and why user acceptance does not follow measured accuracy. | 7 | 14% | Guide CAPDS, chapter 8, Experience in the face of uncertainty |
6. Inference economics 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. | 6 | 12% | Guide CAPDS, chapter 9, Inference economics |
7. Launch and monitoring 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. | 5 | 10% | Guide CAPDS, chapter 10, Launch and monitoring |
8. Governance and responsible use Who approves what, according to the risk of the decision. And the path to contest it, for anyone affected by an automated call. | 4 | 8% | Guide CAPDS, chapter 11, Governance and responsible use |
Certification exam rules
- Format
- Online certification exam, in your browser, with no scheduling and no exam centre. Questions describe work situations and ask for a decision. Some ask for more than one option, and the screen tells you how many.
- Time
- The clock runs on the server and starts when you open the certification exam. It does not pause: reloading the page, switching devices or closing the browser does not give time back. When time runs out, the exam is submitted with what you have answered.
- Navigation
- You can go back, change answers and flag questions for review. Before you finish, a screen shows what is answered, flagged and blank.
- Scoring
- Your score is compared with the pass mark, with no curve. A blank question counts as wrong, and multiple-answer questions get no partial credit.
- Result
- Immediate, with your score for each domain. It is kept in your account and does not appear on the certificate or the public profile.
- Voucher
- Each voucher is worth one attempt and is used when the certification exam is opened. A voucher bought on the site never expires. Whoever types the code takes the exam, so treat the code like a ticket.
- New attempt
- There is no waiting period and no limit. Each attempt uses a voucher, and questions are drawn again. If you do not pass, you get a 50% coupon for the same certification, valid for 90 days.
- Refund
- Within 14 days of purchase, if the voucher has not been used, through the same payment method.
- Technical problem
- Come back through the same address: the certification exam resumes with the same questions and the answers already saved. If a platform failure compromises your exam, a new attempt is released at no cost.
- Credential
- Issued when you pass, with a public verification code. It never expires and has no renewal or annual fee.
- Revocation
- A credential can be revoked for fraud, disclosure of questions, false identity, misleading use or administrative error. The code remains searchable and shows the revocation.
- Integrity
- The certification exam is individual and closed book, including artificial intelligence assistants. There is no camera, microphone or screen recording. Signals such as time per question and loss of window focus lead to human review, and no signal fails anyone on its own.
- Accessibility
- The certification exam screen works with keyboard and screen reader. Extended time of 25% or 50% is requested at checkout or in the account, in Certification exams, before using the voucher. For another adjustment, write to [email protected].
Where to study
The official guide indicates 40 to 50 hours of study, in paths of 6 weeks (appendix D of the guide).
Short on time? Study by weight: the domains below are in the order of how much they count in the certification exam.
- Quality evaluation (18%)
- Feasibility and approach selection (14%)
- Experience under uncertainty (14%)
- Discovery and opportunity framing (12%)
- Data, context and privacy (12%)
- Inference economics (12%)
- Launch and monitoring (10%)
- Governance and responsible use (8%)