Good Transformer
Before you cut a role because of AI, run this test
Over half of leaders who cut jobs for AI now admit they got it wrong, and rehiring has begun. Here are four questions to answer before your firm cuts a role.
Employers that cut jobs because AI could do the work have started hiring people back. More than half of the leaders who made AI redundancies now admit the decisions were wrong, and around a third of the US hiring managers who cut a role for AI have since refilled it. Before your firm makes any headcount decision that leans on AI, there is a test worth running: establish whether the AI has done the whole job or only the routine part of it, and cost what happens if you are wrong. This piece sets out that test as four questions.
The reversals, and what they share
CNBC pulled the reversals together this week. IBM's AI assistant came to handle about 94 per cent of routine HR requests, and the company found the remaining share still needed people, even as it kept hiring in judgement-heavy areas. Commonwealth Bank of Australia reversed call-centre cuts after volumes surged. Ford is rehiring engineers for quality work its automation could not cover. And the pattern is wider than the famous names: Robert Half found that 32 per cent of US hiring managers who eliminated a role primarily because of AI later rehired for the same or a similar position. In finance the figure is 44 per cent, the highest of any sector, which should give a professional-services leader pause.
Look at what these stories share. Each firm made the cut on the strength of what AI did with the routine majority of a job, then discovered that the job is priced on the awkward minority. A role that looks 94 per cent automated sounds like a business case. The missing six per cent is the client who is upset, the request that does not fit the form, the figure that a person with experience would have paused at. That six per cent was what the salary was really buying, and it does not show up in a pilot built on clean examples.
The numbers, honestly scoped
Two published figures put a size on the regret, and both deserve honest framing. Orgvue's annual workforce research found that 39 per cent of business leaders had made employees redundant as a result of deploying AI, and that 55 per cent of those admit wrong decisions were made. The sample is 1,163 senior leaders at large organisations, so this is big-company data rather than a survey of firms like yours. Gartner predicts that by 2027 half of the companies that attributed customer-service headcount cuts to AI will rehire for similar work, often under different job titles. That prediction is about customer service specifically, and it comes with a telling detail: only 20 per cent of the service leaders Gartner surveyed had actually cut staff because of AI at all.
So treat the percentages as a warning rather than a forecast for your firm. The mechanism, though, reads straight across, and a smaller firm carries it worse. When a twelve-person practice loses the person who handled the awkward cases, it has not lost six per cent of a job. It has lost the only person who could do that part.
The test: four questions before any AI headcount decision
Run these questions before the decision is made, and write the answers down. A real case survives being written down; a hopeful one does not.
1. Has the AI done the whole job through a full cycle? A pilot on clean examples tells you about clean examples. The evidence that matters is a full cycle of the actual role with the AI in place: quarter-end, the complaints season, the week the regulator writes, the client who calls instead of filling in the form. Measure outcomes the role owns rather than tasks the tool completes. If the AI has only handled the routine core, you have evidence about the routine core and a guess about everything else.
2. Who now catches what that role used to catch? Robert Half's diagnosis of the rehiring wave, as reported by CNBC, is that gaps in quality, oversight and decision-making surfaced after the early efficiency gains. Most roles carry a layer of checking that was never written down. AI output reads finished whether or not it is right, and the person who can tell the difference is usually your most experienced one. If the answer to "who catches it now" is a person who already has a full job, or nobody, the case is not ready.
3. What does the reversal cost if you are wrong? Price the round trip before you bank the saving: recruitment fees, months of ramp-up, the institutional knowledge that left and is not coming back, and the quiet damage done when the person you let go is asked back because the judgement was wrong. The person on the other side of this decision lost a job to a forecast. If the forecast fails, your firm did that for nothing, and the people who stayed will remember it when you next ask them to adopt a tool. There is also a slower cost: routine work is where juniors learn the trade, and cutting it removes the training ground your next senior people were going to grow up on.
4. Is headcount even the right unit? In most professional firms the honest saving from AI is hours rather than roles, at least at first. Freed hours have obvious places to go: the backlog, faster turnaround, business development that never happens, the checking work from question two. Whether those hours are actually worth money is its own check, and we published a short method for testing whether one AI use pays for itself that pairs with this one. If the savings case only works by converting hours into a person, question one decides it, and question one takes a full cycle to answer.
If the case still stands
Some roles genuinely do shrink or change, and a firm that pretends otherwise is not being honest either. If the four questions come back clean, run the change as a trial with the person still in place: route the work through the AI, have the person handle what it cannot, and record what they end up doing for a quarter. That record tells you what the role actually was, which is knowledge worth having whatever you decide. It also treats the person as a professional whose work is being understood, rather than a line in a projection. And none of this replaces the legal side: redundancy in the UK carries its own consultation and fairness obligations, so take proper advice before any process starts. Working out which jobs in your firm AI genuinely changes, and which it only appears to, is the ground our AI lessons for leaders cover, built around the work your firm actually does.
Questions leaders ask
Are firms really rehiring people they cut for AI?
Yes. Robert Half found 32 per cent of US hiring managers who eliminated a role primarily because of AI later rehired for the same or a similar position, and Gartner expects half of the companies that attributed customer-service cuts to AI to rehire for similar work by 2027. IBM, Commonwealth Bank of Australia and Ford have all publicly walked back parts of AI-linked workforce decisions.
How does a firm know whether AI can actually replace a role?
Only by testing the whole role, end to end, over a full working cycle with the edge cases included. A percentage of requests handled is evidence about the routine share of the work. The decision needs evidence about the rest: who handles it, how often it appears, and what it costs when it is missed.
Should a smaller firm make AI redundancies at all?
Rarely as a first move. A small firm feels the loss of an edge-case handler faster than a large one, and pays proportionally more to reverse the decision. Take the saving in hours first, redeploy them somewhere visible, and let a full cycle of evidence tell you whether a role has genuinely changed.
The reversals in this week's news were avoidable, and the firms making them are paying twice: once for the redundancy and once for the rehire. If a headcount decision is on your table this quarter, run the four questions with the people who actually do the work, before anything is announced. If you want a second pair of eyes on the answers, book a discovery call.