Why most AI hires fail before the search starts
Companies blame the talent market when AI hires go wrong. More often, the role was wrong from the start. Here's how to scope an AI role so the right person can succeed.
When an AI hire doesn’t work out, the post-mortem usually blames the candidate or the market: the person didn’t have the skills, or good people are impossible to find. Sometimes that’s true. But after years on the hiring side of the table, we’ve come to believe most AI hires fail earlier, before a single résumé arrives. They fail because the role was wrong.
The three ways an AI role goes wrong
The role is really two roles. A job description asks for someone to build data pipelines, train models, deploy them to production, design the product experience and present to the board. Each of those is a real job. Ask for all of them and you’ll either find nobody or hire someone who is strong in one area and quietly struggling in the rest.
The title doesn’t match the work. “Data scientist” gets used for everything from dashboard building to research. “AI engineer” can mean someone who fine-tunes models or someone who wires a language model into a product (what we’d call an LLM engineer). When the title and the work don’t match, you attract the wrong candidates and screen out the right ones.
Nobody has defined success. If you can’t say what this person should have built or changed in their first 90 days, you can’t interview for it, and they can’t deliver it. Vague roles produce vague hires.
What good scoping looks like
Before any search starts, a well-scoped AI role answers five questions:
- What problem are we trying to solve, and by when? Not “we need AI”, but “we need to cut invoice processing time” or “we need our support assistant to stop giving wrong answers”.
- What will this person build or own in their first 90 days? Three concrete outcomes, written down.
- What stack, data and tools will they work with? The skills follow from the work, not the other way round.
- Is this a builder role, a leader role, or both? Most failed hires sit in the gap between the two.
- What are the governance, security or regulatory requirements? These shape the role as much as the technology does.
Then add the practical questions: contract or permanent, the realistic rate for this skill set, and who decides.
Why the rate question matters early
A common reason AI searches stall is a budget set for a different role. A team budgets for a mid-level developer and writes a job description for a senior engineer who can own production AI. The search runs for months, the strong candidates decline, and the role gets re-opened at a higher rate anyway.
Naming a realistic rate during scoping isn’t about pushing budgets up. Sometimes the honest answer is that the role should be smaller, or split, or started on contract while the team learns what it needs.
The honest answer is sometimes “this is the wrong role”
The most useful thing a recruiter can say during scoping is sometimes uncomfortable: the role as described won’t work. Maybe it should be two roles. Maybe a contractor should prove the use case before you hire permanently. Maybe the problem is a data problem, and you need a data engineer before you need a machine learning engineer.
That’s why we start every engagement with a free 30-minute scoping call and send a written role brief within 48 hours, whether or not you hire through us. If the brief says the role is wrong, it says so and recommends the right one.
A simple test for your next AI role
Before you post the job, try writing down the three things this person must have achieved by day 90. If you can’t, the search isn’t ready. If you can, you’ve done the hardest part of hiring well, and every interview gets easier from there.
If you’d like a second pair of eyes on a role you’re about to open, book a scoping call. You’ll leave with a sharper brief either way.