A head of AI role needs a clear decision boundary across product, data, research, and engineering. The title alone does not explain whether the person will lead a team, set technical direction, run experiments, or ship application features.
Start with the business and product outcomes that need accountable leadership.
Build the decision map
| Decision | Clarification |
|---|---|
| Product use case | Who decides what user problem AI should address? |
| Model approach | Who chooses between existing capabilities and new development? |
| Data readiness | Who owns access, quality, and appropriate use? |
| Evaluation | Who defines acceptable behavior and release evidence? |
| Infrastructure | Who owns deployment, performance, and operation? |
| Research | Which uncertainties justify sustained experimentation? |
| Team | Who reports to the leader and what expertise is missing? |
Several decisions can be shared, but shared work still needs an escalation path.
Choose the center of gravity
A research-led mandate differs from an application-delivery mandate. A leader strong in one may need a complementary partner in the other.
Describe the hands-on contribution expected. If the team needs someone to build the first product integration, say so. If the main need is leading established specialists, assess that leadership directly.
Test prioritization with a fictional portfolio
Present several fictional initiatives: a promising prototype, an unreliable deployed workflow, and an ambitious research idea. Provide limited team capacity.
Ask the candidate to choose priorities, identify missing evidence, and explain what would change the decision. Look for a clear connection between technical work and user outcomes.
Do not reward the most elaborate AI roadmap by default.
Inspect evaluation ownership
Ask how the leader would distinguish model improvement from product improvement and how they would handle a release that looks strong in a demo but weak in important cases.
They should identify the people who need to participate in the decision, including relevant domain and security expertise.
Make the mandate executable
Specify budget and staffing authority, collaboration with engineering and product leaders, and the initial scope. Avoid making one person accountable for every AI-related activity without control over priorities.
Use the AI infrastructure versus application guide and research engineer review. Discuss an AI leadership search with Refery.