Your first AI hire: who to hire, and when

This is the one article in this series that touches our recruitment side, and it belongs here because the question genuinely arises from the capability side. A company gets serious about AI, concludes it needs someone, and starts writing a vacancy. That sequence is common and usually one step too fast.

We place AI professionals. Recommending that you delay is against our short-term interest, which is exactly why it is worth writing down.

Why the first AI hire usually comes too early

A vacancy is a commitment to a shape of problem. When you post "data scientist" you have decided, often without noticing, that your bottleneck is modelling. In most mid-sized companies it is not. The CBS figures for 2025 show that the most cited barrier to AI use is a lack of relevant in-house expertise, at every company size, which sounds like an argument for hiring immediately. It is actually an argument for knowing which expertise before you buy it.

Hiring before you know is expensive twice. You pay a senior salary, and you get a capable person with nothing well-defined to do, which is the fastest way to lose a capable person.

A hiring conversation. Hire for the problems you have, not the ones a job title implies.

Three signals that you are ready

One: something is already running that a person maintains. Not a pilot, something in daily use that a named person keeps alive. That is proof there is real work, and it tells you what kind.

Two: you can describe the first six months of the role in specific tasks. If the job description reads "explore opportunities for AI within the organisation," you are hiring someone to do the work you should have done before hiring them.

Three: there is someone senior who will make decisions with this person. An AI hire reporting to nobody in particular, with no route to a decision, will spend a year producing interesting analyses that change nothing and then leave.

If you have all three, hire. If you have none, the honest first step is to define the work, which costs a fraction of a salary.

Which role, actually

The titles get used loosely, and picking the wrong one is the most common expensive mistake in this area.

In companies of fifty to three hundred and fifty people, the first genuine need is far more often the first or the last row than the second. The data scientist is the role people ask for and the least often the one that unblocks anything, because a model is rarely the constraint.

The seniority question

One senior beats two juniors for a first hire, and it is not close. A first AI hire has no existing team to learn from, no established practice to follow, and will be making judgment calls with nobody senior to check them. That is a bad environment for someone early in their career and an unfair thing to do to them.

It is also why the personality side of the assessment matters more than usual here. A first hire needs someone who can operate without an established function around them, explain their work to people who do not share their vocabulary, and say no to a director. Those are not technical properties, and they are what the APAC framework is built to look at: adaptability, personality, awareness and connection, rather than only what someone knows.

What to do instead, if you are not ready

The useful sequence when the three signals are absent: run one thing small and see what actually breaks, per AI-roadmap maken. Buy technical judgment by the day rather than by the year while you find out, as described in AI-strategie bepalen zonder eigen CTO. Then write the vacancy from what broke.

A vacancy written from a real bottleneck attracts a different and better candidate than one written from ambition, because good people can tell the difference at a glance.

About this page

Radical places AI professionals, so recommending delay runs against our short-term commercial interest; that is stated at the top rather than buried. The adoption-barrier figure comes from CBS StatLine 86119NED for 2025. The APAC framework is Radical's own selection model and is unrelated to Asia-Pacific. Written by Radical's own team; no client data was used.