Over the last eighteen months we have run eleven searches for a company’s first Head of AI, eight in New York and three in London. Nine have closed. In almost every one, the first conversation was about pay, and in almost every one the pay question turned out to be the wrong place to start.
The market for this title is noisy. Candidates who have held it for a year are asking for packages that would have hired an experienced Chief Data Officer in 2022. Some are worth it. Many are being paid for the word in their title rather than the work behind it.
Start with the problem, not the title
A first Head of AI is hired to do one of three quite different jobs, and the brief should say which.
Build a product capability. Shipping AI features to customers, with the engineering, evaluation and reliability work that implies.
Change how the business operates. Finding and delivering efficiency in operations, finance, service or sales.
Set policy and manage risk. Governance, vendor selection, model risk and regulatory exposure, usually in a regulated business.
These need different people. A strong product builder is often frustrated by an operating brief, and a risk-minded leader from a bank will struggle in a company trying to ship weekly. Of the nine searches we closed, the two that needed a second attempt both began with a brief that tried to cover all three.
Where the hype premium comes from
In New York this year, the median base for a first Head of AI at a business with 500 to 5,000 employees was $335,000 in our data. For a Head of Machine Learning or Head of Data Science at similar companies, with a comparable team, it was $270,000. That is a 24% gap for roles that, on inspection, often involve much the same work.
Some of that premium is fair. Leaders who have taken generative models into production, with real users and real failure modes, are genuinely scarce. But the premium is being paid to a much wider group. In our screening this year, roughly a third of candidates with “AI” in their current title had no production deployment they could describe in detail.
The way to avoid the premium is not to underpay. It is to make sure you are paying for evidence.
What to test for
We now run a structured assessment on every first Head of AI search, built around four questions. None needs a technical panel to ask, though one helps.
Have they shipped, and what broke?
Ask for one system they took from prototype to production. Then ask what failed in the first three months after launch, and how they found out. People who have done the work answer quickly and in detail. People who have run pilots tend to talk about the pilot.
How do they measure quality?
A credible leader will describe how they judge a model’s output against a business outcome, not just a benchmark. Listen for test sets, human review, cost per task, and how they decide something is good enough to release.
What would they not build?
The strongest candidates we met this year were as clear about what they would buy from vendors as what they would build. A first Head of AI who wants to build everything will spend your budget on infrastructure before delivering anything a customer notices.
Can they explain it to the board?
A first Head of AI will be in front of the board within six months. We ask candidates to explain one past project, including its risks, to a non-technical panel member in five minutes. It is a revealing exercise.
Pricing the offer on evidence
With evidence in hand, the offer becomes easier to calibrate. Our guidance for a New York hire at this level:
Base of $290,000 to $360,000 for a candidate with at least one production deployment at scale
Above $380,000 only for a proven builder who has led a team of 20 or more and shipped to customers
Equity weighted towards performance, with a portion tied to measurable delivery in the first two years
In London, equivalent bases are typically £150,000 to £190,000.
The first ninety days
A first Head of AI succeeds or fails on whether the business gives them a clear first problem. The placements that are going well a year on all started with a single named use case, a sponsor on the executive team, and a budget line that existed before the new hire arrived.
Hiring the person first and finding the problem later is the most expensive way to do this.
