A recruiter and an AI tool walk into a hiring process. Twenty minutes later, the AI has drafted the advert, screened four hundred CVs, and scheduled the interviews. The recruiter is still deciding whether the candidate on paper is actually the right call. Guess which one the client thinks they’re paying for.
That’s not a joke so much as a diagnosis. We’ve spent years getting better and better at measuring the wrong things – and technology just handed us the invoice.
Time-to-hire. Cost-per-hire. Offer acceptance rate. Candidate NPS. Source effectiveness. Interview-to-offer ratio. We’ve built entire TA strategies around shaving these numbers down, like a profession-wide obsession with personal bests in a sport nobody asked us to compete in.
None of them answer the only question a business leader actually cares about: did this hire make the organisation better? Not faster. Not cheaper. Better.
Scarcity used to be the whole business model
For most of recruitment’s history, doing the process well was the value, because the process was hard. Finding people was hard. Market intelligence was hard. Screening a pile of CVs took actual human hours you’d rather have spent on literally anything else. Recruiters who could push through that friction had a real edge. Fair enough – that’s what we built the scorecards around.
AI is now quietly taking a sledgehammer to that whole arrangement. It drafts the advert. It builds the sourcing strategy. It screens the CVs. It summarises the interview. None of this is a forecast, it’s Tuesday. Once a capability is everywhere, being good at it stops being a personality trait. Nobody hires an accountant because they own a calculator. Soon nobody’s picking a recruiter because they knocked five days off a hiring process but that’s table stakes now, not talent.
Efficient isn’t the same as good – it just feels similar
If everyone’s got roughly the same AI tools, time-to-hire compresses for everyone. Cost-per-hire drops for everyone. The playing field flattens. None of that makes you exceptional. It makes you current, a different thing, and a much less impressive one.
So what’s left when you take the stopwatch away? Judgement. Context. The ability to actually read a person. Which, awkwardly, is the thing recruitment always claimed to be about while quietly measuring everything else instead.
AI can’t fix this problem, mostly because it was trained on it
Here’s the part that doesn’t get said enough: AI tools built for recruitment aren’t neutral. They’re optimised against the same metrics we’ve just established are broken, time-to-fill, keyword match, “successful placement” defined as didn’t leave in 90 days rather than actually mattered. There’s no universal definition of a good hire baked into any dataset, because there’s no universal definition of a good hire, full stop. What counts as success at one client is a rounding error at another. I won’t get into the biases baked into the algorithms themselves – if you want to go deeper on that, “Rage Inside the Machine” by Dr. Robert Elliott Smith is worth your time.
So when a tool learns from historical hiring data, it isn’t learning what a good hire looks like. It’s learning what got approved, which, given everything above, is a proxy for a proxy for a guess. Feed a model bad definitions of success and it doesn’t correct the definition. It just gets faster and more confident at reproducing it. That’s not a hiring algorithm improving over time, it’s a mistake compounding at scale, with better production values.
The problem: “judgement” doesn’t sell itself
Recruitment consultancies have been pitching “we’re strategic partners, not CV pushers” for at least fifteen years, with patchy results, because judgement is nearly impossible to demonstrate before you’ve already delivered it. A client can’t test-drive your instincts the way they can test-drive a faster shortlist.AI doesn’t fix that trust problem. If anything, it sharpens it: clients now have a cheap, fast, self-serve alternative sitting right next to you for comparison. “I understand your market better than an algorithm does” is a claim. It needs proof, not conviction.
Which means the real question isn’t “is judgement valuable?” It’s: how do you make judgement visible before the client has any reason to trust it?
Retention was never the answer, but it’s not just a philosophical problem either
Someone can sit in a seat for a decade and move nothing. Someone else can be there four years, rebuild the team, and leave the place stronger than they found it. Retention conflates the two. Fine, but “did they matter?” is a lovely question to ask at a dinner party and a useless one to ask a hiring manager on a Tuesday who needs a decision by Friday. If “did this hire create value” is going to mean anything operationally, it has to become something you can actually track during a search, not just muse about afterwards.
Mechanism not mood: the Recruitment Value Index
If efficiency metrics are becoming baseline and AI tools can’t quietly fix the definition problem underneath them, I don’t think the answer is to abandon measurement and retreat into vibes about “humanity.” It’s to build a metric that measures the thing that actually matters, and to make it as concrete as time-to-hire ever was.
Call it a Recruitment Value Index. Four components, scored per hire, built entirely from things no dataset contains:
1. Fit Accuracy. Before you source a single candidate, write down, with the client, in the room, what “this hire mattered” would actually look like in 12 months. Specific, falsifiable, theirs. Not “strong communicator,” but “reduced escalations to the ops director by X.” Twelve months later, you check it: met, partially met, or missed.
2. Time-to-Contribution. Not time-to-hire – time from start date to the point the hiring manager actually says “they’re adding value now.” Same instinct as the old metric, pointed at the right target.
3. Movement Integrity. For every candidate you put forward, you already write a paragraph on why they’d actually leave, not just why they’re qualified. Movement Integrity checks whether that read held up: did the reason you gave for why they’d move end up being the reason they took the role seriously and stayed engaged? If you said “growth, not money” and six months later that’s exactly why a counter-offer didn’t land, that’s a hit. It’s the one line in the Index no algorithm can touch, because it was never data. It was a conversation.
4. Downstream Impact. One question, asked of the hiring manager at 90 days and 12 months, logged every time: what’s different now that wouldn’t have happened otherwise? Converted into a scored outcome rather than left as an anecdote you half-remember at the next QBR.
Four numbers, one hire. Track it across every placement with a client and you’ve built something no competitor with a faster ATS can replicate overnight: an actual evidence base, specific to that client, that says this is what we predicted, and this is what happened.
Go back to that recruiter and the AI walking into a hiring process together. The AI will always win the speed contest, it doesn’t get tired, it doesn’t need coffee, it doesn’t have an instinct to second-guess. But nobody’s hired for the process to be quick. They’re hired for the outcome to be right. The Index is how you prove that’s still true and it’s the one part of this job the machine was never actually invited to do.
