When budgets tighten, cutting AI feels smart. The recession winners do the opposite.

The headlines are darkening, and inside a lot of companies the reflex is already forming. When the belt tightens, the "future" projects go first, and AI, still sitting in most budgets under the heading of innovation rather than operations, is near the top of the cut list. It feels prudent, even responsible. The trouble is that the evidence from every recent recession says it is precisely the move the companies that lose tend to make.

Start with the research, because it is unusually clear. A landmark Harvard Business Review study by Gulati and colleagues tracked 4,700 public companies across three global recessions and found that only about 9% came out of them stronger than they went in. Worse, roughly 80% of the companies that merely survived took more than three years just to claw back their pre-recession rate of growth. A downturn is not a pause you recover from quickly; for most, it is a hole that takes years to climb out of. McKinsey's tracking of the 2008 crisis found the same shape: a resilient minority, around 10%, pulled ahead during the downturn and held that lead for years afterward.

What actually separated the winners?

Not the size of their cuts, and not the boldness of their spending. The distinguishing move was doing both at once. The study sorts companies into those that played pure defense (cut everywhere), those that played pure offense (invested through it), and the "pragmatic" ones that combined disciplined cost control with selective investment in the capabilities that would matter when the cycle turned. It was the pragmatic combination that most reliably produced the winners. Cutting alone left firms lean but weak; spending alone left them exposed. The companies that roared out had done the harder thing of tightening and investing in the same breath.

This is exactly where AI is unusual, and why cutting it reflexively is a mistake. AI is one of the few levers that lives on both sides of that equation simultaneously. It is a defensive move, a direct way to take cost out, do more with fewer people, and protect margin. And it is an offensive one, a way to build capability and take share while competitors are busy retreating. To cut AI in a downturn is not merely to forgo some future upside; it is to throw away your best cost-out tool at the exact moment you most need to cut cost. It is the losers' reflex wearing the costume of prudence.

Doesn't caution make sense when money is tight?

Discipline does; reflexive cutting does not, and nor does spending through a downturn on faith. The same research is a warning to the pure-offense crowd too: the firms that only invested, without the matching cost discipline, also underperformed the pragmatic combiners. So this is not a licence to pour money into AI while the economy contracts. It is an argument for raising the bar rather than dropping the tool. In a downturn every AI pound has to show a fast, measurable return, and plenty of half-baked pilots deserve to be cut on exactly those grounds. The point is that the answer to "which AI survives the cuts" is not "none of it", it is "the part that pays, and quickly". This is the same discipline behind measuring AI by value rather than activity, only now with the stakes raised.

How do I apply this when the belt tightens?

By treating a downturn as a moment to separate from your competitors, not simply to shrink.

1. Reclassify AI from innovation to cost-out. Whatever is filed under discretionary future-gazing gets cut first; the identical work filed under margin protection survives. Move it before the budget conversation, not during it.
2. Raise the bar, don't drop the tool. Demand fast, provable payback from each AI use. Keep what clears the bar and cut what doesn't, rather than cutting the whole category.
3. Play both sides on purpose. Use AI to take cost out now and to build the capability that lets you take share once rivals have pulled back. That combination is what the winners did.
4. Invest selectively, not indiscriminately. Pure offense loses too. Concentrate on the few AI bets with a clear, short line to margin, and starve the rest.
5. Keep your own base light. The counter-cyclical move only works from a position resilient enough to make it. Low fixed costs are what let you invest when others cannot.

What does this change for me as a leader?

It reframes a downturn from a reason to stop into a moment to separate. When most of your competitors reflexively cut their AI as "future spend", the ground opens up for the disciplined minority who treat it as the cheapest way to protect margin and quietly build advantage while everyone else is retreating. The recession winners were never the boldest or the most frugal. They were the ones who managed to be both at once, and AI is a tool built for precisely that dual job.

Two honest caveats keep this sober. This is not a prediction that a downturn is imminent, and it is not a call to spend recklessly. In a genuinely severe freeze, even provable spending pauses for a while because no one is authorised to decide, so the counter-cyclical play depends on keeping your costs light enough to wait it out. But whenever the cycle does turn, the playbook is the same one that has worked for forty years, now with a sharper tool in it: tighten with discipline, invest with precision, and let the reflexive majority cut the very thing that would have carried them out. It is the quiet logic underneath the end of business as usual, applied to the one part of the cycle most leaders get wrong.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.