The Layoff Mirage

I have built the deck that justifies a workforce reduction. More than one, at more than one institution — offshore transitions, shared-services consolidations, operating-model changes where the headline outcome was fewer people. I know what the cost page looks like because I assembled it. Salaries, benefits, real estate, severance, all modeled to the dollar, reviewed by finance, approved upstairs.

Here is what was never in the deck: the knowledge line. Nobody models what walks out the door with the people. The analyst who can tell a real anomaly from a data error in three seconds because she has seen a decade of normal ones. The guy who knows why the exception process works the strange way it does, because he was there the day it broke the other way. None of that has a line item, so none of it exists in the business case. The savings land in quarter two. The bill lands in year two, and by then it has a different name — a quality problem, unexpected attrition, a performance issue. It is never traced back to the cut. I have watched that bill arrive on programs I built. At the time, we called it attrition.

What the Data Actually Says

Hold that against the best data we have. Gartner surveyed 350 large enterprises this year. Around 80% of those deploying AI cut jobs. There was no correlation between the cuts and the returns. None. High-ROI firms and low-ROI firms cut at almost the same rate. The companies actually getting value from AI were mostly the ones using it to make their people more productive, not to remove them. Gartner's own framing is the cleanest version of the problem: cuts create budget room, they don't create return. Those are two different things, and a lot of boards are confusing them right now.

The confusion is structural, not stupid. Earnings season demands a number. AI is the story of the year and the market wants it translated into something concrete this quarter. You cannot show a redesigned operating model in 90 days. You can show a reduction. So the cut becomes the proof of the strategy, when in a well-run transformation it should be close to the last thing that happens. There is already a term for the dynamic — AI-washing. Reductions driven by the rate cycle or plain cost pressure, relabeled as AI efficiency, because that is the version the market pays for.

Headcount Is an Output, Not an Input

Sometimes the right operating model genuinely needs fewer people. Pretending otherwise is its own kind of dishonesty. The argument is about sequence. Headcount is an output of a redesigned operating model. It is what falls away after you have rebuilt how the work gets done — the process redrawn, the roles re-scoped, the freed capacity redeployed or genuinely eliminated. When the reduction is the output, it holds. When it is the input — when the program opens with a reduction target and works backward — you have not transformed anything. You have run a cost exercise under a more fashionable name.

The tell is easy once you have been inside both. A real transformation can tell you what changed about the work before it tells you how many people left. It can show you the process that no longer exists, the decision that moved, where the capacity went. The headcount number falls out of that, and it is explainable. A cost program in costume leads with the number and reverse-engineers the story. It can tell you exactly how many roles are going. Ask what about the work is different and you get nothing, because the answer is nothing. The same work, spread across fewer people, until quality erodes, the knowledge gaps surface, and the savings quietly reverse. Usually about two years out. Comfortably past the quarter that took the credit.

Why the Mirage Survives

The two-year lag is what makes the mirage durable. The executive who ran the cut was promoted on the savings before the bill arrived. The institution metabolizes the loss as ordinary friction. The next cut gets approved on the same flawed case, because the last one was never honestly accounted for. I watched this run on offshore transitions. I watched it run on shared services. AI is just the newest and most expensive cover story.

So when you are shown an AI transformation whose centerpiece is a headcount number, the question is not whether the savings are real. On paper, for a while, they usually are. The question is: what about the work is actually different? If there is a clear answer, the cut is an output and it will hold. If the answer is "the same things, with fewer people," you are looking at the mirage. The savings will be real for about two years. So will the loss. Only one of them will be on the slide.