How to hire a Head of AI, or grow one internally

When you actually need the role

Most companies don't need a Head of AI on day one. Early adoption can be owned by a capable operations lead or a motivated department head alongside their day job. The role becomes necessary when the side-of-desk model breaks: several departments using AI in uncoordinated ways, real money going into tools and training, security and data questions with no clear owner, and no single person the board can ask what the spend is returning.

If those symptoms sound familiar, the question is not whether to fill the role but how. We've covered what the job actually involves, the remit, the reporting line, and the first ninety days, in what does a Head of AI do. This post is the buyer-side companion: how to get the right person into the seat.

One caution before you write the job spec. Creating the role too early can be as damaging as too late, because a Head of AI hired before there is real adoption to coordinate spends their tenure evangelising rather than delivering, and the title quietly loses credibility. The role should land on top of existing momentum, not substitute for it.

External hire or internal promotion?

The instinct is to hire externally, on the logic that AI expertise is the scarce ingredient. In our experience the logic runs the other way. The Head of AI's job is to change how your organisation works, and that depends on knowing your workflows, your systems, your politics, and your people. That domain knowledge takes years to acquire from outside. The AI capability, by contrast, can be trained into a strong internal operator in months.

External hires make sense in specific cases: when no credible internal candidate exists, when the role needs to build a technical team from scratch, or when the organisation needs an outsider's licence to challenge how things are done. But an external hire with impressive AI credentials and no feel for your business will spend their first year learning what an internal candidate already knows, and many stall in that year. If you have someone internal with the operator instincts, the grow-internal path is usually faster and cheaper than the search.

What to look for: an operator, not a researcher

The single biggest hiring mistake with this role is optimising for technical depth. A Head of AI does not need to build models. They need to change behaviour across departments, run vendor decisions, set guardrails, and prove value to a board. That is an operator's job.

• Evidence of shipped change: they have taken a workflow inside a business from old way to new way, and can tell you what it saved.
• Comfort with measurement: they instinctively talk about baselines and before-and-after numbers, not visions.
• Cross-department credibility: they can hold a room of sceptical department heads without hiding behind jargon.
• Hands-on fluency: they use AI daily on their own work. Someone who delegates all actual usage cannot judge what to roll out.
• Pragmatism about risk: they can write a usage policy that protects the company without banning everything useful.

A research background is not a disqualifier, but treat it as neutral. The profile that fails most often is the impressive technologist who has never had to make a change stick inside an organisation that didn't want to change.

Interview questions that expose real capability

Generic questions get generic answers. These force specifics:

• "Walk me through one workflow you changed with AI. What did it look like before, what does it look like now, and what did you measure?" Operators answer with detail; talkers answer with strategy.
• "Tell me about an AI rollout that failed. What went wrong and what did you do?" Anyone who claims no failures hasn't shipped much.
• "How would you spend your first month here?" Good answers start with listening and a baseline, not with tools.
• "What's something AI is bad at that most executives think it's good at?" This tests honest, current, hands-on judgement.
• "Show me something you've built for yourself." A prompt library, an automation, a custom workflow. The artefact matters less than that one exists.

In every case, push past the first answer. The difference between real capability and good reading shows up in the second and third layer of detail.

The grow-internal path

If you promote from within, be honest that you are asking someone to grow into a role that didn't exist, and resource it properly. The pattern that works: give them the mandate formally (title, time, a budget line, a reporting line to the executive team), close the AI capability gap with structured training rather than leaving them to YouTube, and support the leadership step-up, because most internal candidates are strong operators who have never held an organisation-wide remit.

This is exactly the gap our Head of AI Bootcamp exists to close: the vendor judgement, the measurement discipline, the rollout playbooks, and the guardrail design that an internal promotion needs to be credible in the seat quickly. For the step-up in seniority itself, some organisations pair that with leadership coaching for the first two quarters. Done properly, an internal candidate can be performing the full role inside six months, with the domain knowledge no external hire could match.

One last test, whichever route you choose. Six months in, the Head of AI should be able to show the board a before-and-after on named workflows, not a slide of activity. If the person you are about to appoint could not describe how they would produce that evidence, keep looking, or keep training.