Scope the role before you scope the person
The hardest part of a first AI hire is not finding candidates. It is that the role is usually defined as "someone to do AI", which is not a job — it is a department. The first hire needs one problem, described concretely enough that you could tell whether it had been solved.
A good test: can you describe, in one sentence, what will be different in six months if this hire goes well? If not, the role is not ready to advertise.
Decide which shape you need first
Almost every first hire should be applied rather than research-leaning — someone who builds a product around existing models rather than training new ones. It is the broader pool, the lower rate, and much more likely to match the problem you actually have. The three role shapes guide covers the distinction, and it matters more for a first hire than for a fifth.
Interviewing for work you cannot assess
You do not need to evaluate someone's mathematics to make a good first hire. You do need to evaluate judgement, and that is assessable by anyone who has shipped software.
-
1
Ask them to scope your actual problem
Describe the thing you want built and ask how they would approach it. Strong candidates ask what the data looks like, what "good enough" means, and what happens when it is wrong. Weak ones go straight to model choice.
-
2
Ask what they would not build
The most valuable answer a first hire gives is that part of your problem does not need machine learning at all. Candidates who cannot say that will build whatever they are asked to.
-
3
Borrow a specialist for one stage
One conversation with someone who does this work — an advisor, an investor's technical contact, a friendly contractor — will tell you more than three internal rounds. It is a favour worth asking for.
-
4
Check how they explain things
Your first hire will spend a lot of time explaining trade-offs to people without the background. If they cannot do it in an interview, they will not do it on the job.
When a contractor is the better first move
If the problem is genuinely unclear, hiring permanently is an expensive way to find out what you needed. A contractor for a defined piece of work will often tell you within a couple of months whether the problem needs a specialist at all — and if it does, you will be able to write a far better advert afterwards.
The trade-off is that contractors do not build institutional knowledge, and a first hire is partly about that. Contract first when the problem is undefined; hire first when it is defined and ongoing.
Setting them up to stay
Before their first day
Start with the market
Before writing the advert, it is worth seeing what comparable teams are offering and who else is hiring for the same shortlist.