There is a particular kind of quiet that settles over a team after a restructuring. The people who are left keep their heads down. They do the work. But something has changed in the room, and everyone can feel it. The unspoken question is whether they are next.
Now imagine that same team, a year later, watching the company quietly try to rehire the roles it just eliminated. The savings never fully materialized. The work that was supposed to disappear came back, messier than before. And the people being asked to pick up the pieces are the ones who survived the first cut.
That is the whipsaw. It is becoming one of the most expensive patterns in business right now, and most of the leaders living through it did not see it coming when they made the original call.
The numbers are starting to confirm what a lot of operators already suspected. Gartner has predicted that half of the companies that cut customer service staff because of AI will reverse course and rehire by 2027. Forrester found that a majority of employers who restructured around AI now regret the decision. Read those two findings together and you see something that has nothing to do with whether AI works, because it clearly does. What went wrong is more human than that. A lot of leaders misjudged what those jobs actually were.
It is worth being precise about what AI actually replaces. It is the most capable information system ever built, and it can hold everything your team knows. Holding what a team knows and doing what a team does are two different things, and the gap between them is where these decisions go wrong.
Soren Kaplan put it well in a recent piece for Inc.: hiring AI to do a judgment-heavy job is like hiring a surgeon who has only read the textbooks. The information is complete. The reading was thorough. But the surgeon has never once stood over a patient and made a call in the moment, with incomplete information and real stakes. You would never let that person operate. Yet companies made the equivalent decision when they swapped out workers whose value came from having done the job, under pressure, thousands of times.
Klarna is the case everyone points to, because Klarna said the quiet part out loud. In 2023, the company claimed its AI assistant was doing the work of 700 customer service agents and projected enormous savings. By 2025, it was walking that back, acknowledging publicly that it had prioritized cost over quality and that the result was not sustainable. The chatbot could handle volume. What it could not handle was the fraud dispute that did not fit the pattern, the billing problem with three things wrong at once, and the customer who was upset for reasons the script never anticipated. Those moments need someone who has seen the shape of the problem before. That is judgment, and judgment is not retrievable.
The instinct is to treat this as a hiring mistake you can simply undo. Let people go; discover you needed them; hire them back. Annoying, but recoverable.
It is not that clean, and the reason is cultural. When you cut a team and then scramble to rebuild it, the people who remained were watching the whole time, and they learned something about how decisions get made around here. The institutional memory you let walk out the door, the relationships, and the pattern recognition built over years, those do not come back when you repost the job. What you are trying to do is repurchase something that took years to accumulate, and the rebuilt version is rarely as good as the original.
Experienced operators carry a kind of tacit knowledge that does not live in any document. It is the reason a seasoned ops lead can look at a process and know, before the data confirms it, that something is about to break. That kind of intuition works like infrastructure, and you tend to find out how much you were leaning on it only after it is gone.
The alternative is not to adopt AI more slowly. It is to sequence it more carefully.
The companies avoiding the whipsaw are doing something that sounds obvious and is harder than it sounds: they put AI on the routine work and protect the people whose value comes from having been there before. The starting question is not how many people a tool can replace, but which decisions on the team actually require human judgment. Then they build around those decisions instead of betting they will disappear.
That means being honest in the audit. Some roles really are task execution, and automating them is the right call. Some roles look like task execution from the org chart but are actually held together by someone's accumulated judgment, and those are the ones that get expensive to lose. McKinsey's research on generative AI keeps landing on the same conclusion: the value shows up when AI and people are designed to work together, not when one is swapped wholesale for the other. The early evidence on how people actually use these tools points the same direction. The human stays in the loop precisely where real trade-offs are involved.
The sequencing matters too. You earn the right to automate a workflow by understanding it deeply first, while experienced people are still running it. There is a reason the most durable version of this looks a lot like a well-run managed operation: the people running the process document how the judgment calls actually get made before anything gets handed to a tool. Then you deploy the agent inside a process you already understand, with a human owner who can catch it when it drifts. That is the opposite of cutting first and hoping the tool fills the hole. It is slower at the start and dramatically cheaper across the year because you never have to run the rehire in reverse.
There is also a route that gets overlooked in the rush to automate in-house, which is to bring in a team that has already done this design work. A good managed team is not a stack of cheaper bodies, and it is not where you send judgment-heavy work to quietly disappear. The version worth paying for already carries its own accumulated judgment, with AI embedded where it belongs and a named human owning the calls that matter. The point is that someone else has already done the hard design work and pressure-tested it across a lot of clients, so you import a working human-and-AI model instead of running the experiment on your own team and eating the whipsaw when it goes sideways. The work still gets done by people who have seen the problem before, and the automation sits where it earns its keep.
If you are sitting on a restructuring decision right now, the useful question is not "Can AI do this?" AI can do a stunning amount. The question is narrower and more uncomfortable: of the things this team does, which ones depend on someone having made the call before, under pressure, when the answer was not in any document?
Protect those, automate around them, and keep a human owner on anything where a wrong judgment gets expensive. That is how you get the leverage AI promises without the rehire showing up in your budget eighteen months from now. It is a less satisfying answer than a clean headcount cut, and it holds up a lot better.
___________________________________________________________________________________________________
|
|
About the Author: Jen Spencer is the Chief Growth Officer at Booth. She spent nearly a decade scaling a 300-person agency on the GTM side before stepping in as CEO and now advises and invests in early-stage SaaS companies. She's the author of Lead Anyway, a permission-slip book for leaders ready to stop performing and start leading on their own terms.
|