The Real Cost of a Bad Hire in 2026 (and How Much of It Traces Back to AI)

Agencies have gotten reasonably good at pricing the obvious costs of a bad hire — the placement-fee refund, the re-run search, the client relationship damage. What's less priced in is a newer contributor to that number: a hire who got through the interview on answers that weren't entirely their own.

What "£180K" is actually made of

The commonly cited figure — roughly £180K lost per agency, per year, to bad hires — isn't one line item. It's the sum of several:

• The direct refund or re-placement cost owed to the client under most contingency and retained agreements
• The sunk recruiter time across sourcing, screening, and multiple interview rounds that produced a hire who didn't work out
• The client relationship cost — a bad placement is exactly the kind of thing that makes a client quietly start talking to a competing agency
• The re-run cost of doing the entire search again, usually under more time pressure the second time

Most of this was already true before AI-assisted interviewing existed. What's changed is one of the inputs.

The AI-assisted layer on top of it

With roughly 1 in 3 interviews now including some form of AI-assisted answer, a meaningful share of "bad fit" hires aren't actually fit problems — they're screening problems. A candidate who leaned on a stealth copilot to sound more technically fluent than they actually are doesn't just risk underperforming; they were never accurately screened in the first place, which means every downstream cost above traces back to an interview that should have been flagged and wasn't.

Why this is easy to miss after the fact

A bad hire rarely gets a clean postmortem that traces back to "the interview answers were suspiciously polished." It gets chalked up to culture fit, changed priorities, or bad luck — because without evidence from the actual interview, there's nothing to point to. This is exactly why evidence matters more than instinct here: a flag tied to the specific line in the transcript that triggered it is something a hiring manager can actually act on in the moment, not a vague unease that only makes sense in hindsight.

The threshold that actually matters

Not every AI-assisted answer means the same thing. Under roughly 90% confidence, it's typically closer to good preparation than live reading, and worth a follow-up question rather than a rejection. Above that threshold, it's a high-confidence match against a live AI copilot — close to what that £180K bad-hire cost would have been, caught before the offer went out instead of after.

Pricing the fix against the cost

Catching one bad hire before an offer goes out is worth roughly what the average bad hire already costs — which means a detection layer that costs nothing extra to run pays for itself the first time it catches something real, and every flag after that is pure upside. That's the actual math agencies are running when they add AI Detectability to a process that previously had no way to catch this at all.

Candit's AI Detectability runs in the background of every interview, flags the moment an answer looks AI-assisted, and shows the exact evidence behind it — real-time, per speaker, with zero extra setup. It also suggests a follow-up to ask the moment something is flagged (Counter AI), and after every call the AI use report shows where and how AI help showed, ready for the debrief before an offer goes out.

None of this requires a separate step, either. Flags and scores sync straight into Lever or Ashby, and the whole record is reachable from Claude, ChatGPT, or any other MCP-compatible tool — so the evidence that could have prevented a bad hire is actually where the people making the decision already work.

First session free. No card required.