How long does hiring an AI specialist take, and why is it that long?

Where the weeks actually go

A properly run search for a genuine AI specialist takes 6 to 10 weeks from brief to signed contract. Roughly 2 weeks to scope the role and identify candidates. Another 2 to 3 weeks for technical assessment and interviews. A week for reference checks and contract terms. A further 1 to 2 weeks for notice periods and, where the engagement is inside IR35, for the determination itself.

Most of that time is worth spending. Some of it exists only because searches have always run that way. The stages worth protecting are the ones that tell you something true about the person in front of you. The stages worth cutting are the ones that tell you nothing new.

That timeline assumes a straightforward hire: one specialist, one discipline, a clear business need. Add ambiguity to the brief, or a hiring panel that cannot agree on the mandate, and the same process runs to 3 months without any single stage taking obviously longer. Time gets lost between stages far more than it gets lost inside them.

Scoping is not theatre

The single biggest cause of a slow, failed search is a brief written around a tool rather than a discipline. Asking for someone who knows a particular framework tells a specialist nothing about the problem you are solving, and it tells you nothing about who to shortlist.

A brief written around a discipline, LLM applications, MLOps and platform, data engineering, AI governance, forces a decision before the search starts rather than during it. That decision takes 1 to 2 days. It saves weeks later, because you stop interviewing people who are good at a different job to the one you actually have.

The cost of getting this wrong is not measured in delay alone. A brief scoped around a tool rather than a discipline routinely produces a shortlist of people who can use that tool but cannot own the outcome it is meant to produce. You then discover the mismatch during onboarding, which is the most expensive place to discover it.

Sourcing is slow because titles lie

Nothing at all is published for AI governance, AI risk and compliance, AI safety and evaluation, AI product, or AI training and enablement as job titles in the UK advertised market. There is no median rate to quote for any of them, because almost nobody advertises under those names.

That work gets bought under borrowed titles, most often machine learning engineer or data scientist. A role advertised honestly, like Machine Learning Engineer, does carry a published figure: a median contract day rate of £575 over the six months to August 2026, according to IT Jobs Watch. A search for someone doing governance or evaluation work under that same title takes longer, because you are reading past the label on every CV rather than filtering by it.

This is worth naming plainly to a specialist during the brief conversation rather than assuming they will read it off a job board. Ask directly which discipline they work in, using the discipline's own name, and watch whether the answer is immediate and specific or vague and reaching.

None of this is a reason to avoid AI governance or AI risk and compliance work altogether. It is a reason to write the requirement honestly and expect the search to take slightly longer than a more commonly advertised role.

Technical assessment: cut the generic test, keep the specific one

A take-home exercise that could apply to any backend role is theatre. It costs the candidate a weekend, costs you a week of waiting for a result, and mostly confirms something the CV already told you.

A short exercise built from your own data, or your own failure mode, tells you something real: how the candidate reasons about the actual problem you have. It takes less of everybody's time and it is the highest-value hour in the whole process.

There is a fairness argument here too. Asking a senior specialist, likely charging close to the median contract rate for their discipline, to spend an unpaid weekend on a generic exercise is a poor use of their time and yours. A 2-hour paid exercise, scoped tightly to your problem, respects both.

None of this requires exotic tooling. A spreadsheet of messy customer records, a folder of real support tickets, or a real API with sample traffic is enough. The specificity is what does the work, not the sophistication of the exercise.

One serious interview beats several polite ones

Multiple interview rounds with different panels asking similar questions are a hedge against making a decision, not a route to a better one. Each additional round adds a week and rarely adds information.

A single 90-minute conversation, run by someone who can actually assess the discipline in question, and covering a single piece of real past work in detail, tells you more than 4 separate half-hour rounds. Shrink the panel and go deeper instead of wider.

If a panel genuinely needs more than one perspective, run it as two people in the same room asking different questions, not two separate meetings a week apart. The information gained from a second panellist rarely justifies the calendar cost of a second appointment.

Reference checks are the stage most people skip, and shouldn't

Reference checks are the first thing cut when a search runs late, and they are the stage most likely to catch the gap between what a CV claims and what actually happened. A 30-minute call with the person who commissioned the work, not a colleague who watched it happen, closes questions no interview can.

Skipping it saves a few days. It does not save the cost of a hire that does not work out, which runs to a great deal more than a few days of someone's time.

A good reference question is specific: what did this person actually build, what went wrong, and how did they respond. A vague question like 'would you work with them again' produces a vague answer that protects the referee more than it informs you.

Contracts and IR35: slow, but not skippable

Since April 2021, for medium and large private-sector clients, it is the client who makes the IR35 determination, not the contractor. Getting that right, and agreeing rate, notice period and scope, takes a week that should not be compressed, because it is a legal position you may have to defend later, not paperwork to clear.

Where an engagement runs inside IR35 through an umbrella company, deductions on the contractor's side are substantial, though the exact amount depends on their arrangement. Worth remembering when a candidate negotiates on day rate: they are negotiating take-home pay, not the headline figure on the contract.

None of this needs to feel adversarial. A specialist who has done this before will usually have their own standard contract terms and a clear, quick answer on their IR35 position, because they have been through the determination many times. The slow contract stage is usually slow on the client side, not the candidate's.

Building a week of slack into the plan for this stage, rather than treating it as an afterthought once someone has accepted, keeps the whole timeline honest.

What to do about it

• Write the brief around a discipline, not a tool or a framework.
• Run one deep technical interview with a real assessor rather than several shallow ones.
• Build a short assessment from your own data instead of using a generic take-home test.
• Always call a reference who commissioned the work, not a peer who watched it.
• Do not compress the IR35 determination or the contract stage to save a few days.

Questions people also ask

Where the figures come from

Every rate and salary quoted in this article is a median or percentile of figures advertised in UK job postings over the six months to 8 August 2026. They are not rates paid, and the gap widens at the top of a range.

• IT Jobs Watch, UK contract rates, 6 months to 8 August 2026, read 2026-08-08.
• IT Jobs Watch, UK permanent salaries, 6 months to 8 August 2026, read 2026-08-08.

More for hirers

• What does a fractional AI leader actually cost in the UK?
• Fractional AI leader or AI consultancy: which do you actually need?
• How do you tell a good AI engineer from a confident one?
• What belongs in the contract when you hire a fractional AI lead?