The downturn playbook for AI-native firms: cut the fat, not the capability
If a downturn hits, the tempting move is to turn AI into a cost-cutting program. Across 4,700 companies and three recessions, firms that ran that play had the worst odds of leading afterward. The winning move: cut the good-times fat, protect service quality, and put AI-freed hours into retention.
Sometime in the next two years, revenue may dip. When it does, every owner of a service firm will hear the same pitch, from a partner, a CFO, or their own tired brain at 11pm: AI lets us do the same work with fewer people.
That play has been run before, without the AI part, and the results are on record. Harvard Business Review studied 4,700 public companies across three recessions. Firms that leaned hardest on cost-cutting had a 21% chance of leading their industry when the economy recovered. Firms that cut selectively while continuing to invest had a 37% chance. Only about 9% of all companies came out of a recession stronger than they went in.
The dip itself is a live possibility, not a thought experiment. The Philadelphia Fed's survey of professional forecasters puts the odds of a negative GDP quarter near 25% for each of the next three quarters, and the estimates have been rising all year.
So this piece is a playbook for that moment. The short version: cut the fat you built up in good times, protect customer experience and the technology that improves it, and spend the hours AI frees up on keeping and winning clients.
The cost-cutting pressure is building before any downturn
The reframing of AI as a headcount tool is already underway in a growing economy. Challenger, Gray & Christmas counted 101,743 US job-cut announcements citing AI in the first half of 2026, nearly double the 54,836 for all of 2025. In June 2026, AI was the most cited reason for layoffs in the country.
CFOs are moving the same direction. In Gartner's surveys, 42% of CFOs anticipate some AI-driven headcount reduction in support functions, and headcount growth targets have dropped from 6% in 2025 to 2% in 2026. The budget context explains the mood: 62% of finance leaders admit they have labeled a software purchase "AI" to get it approved, while two-thirds sit under orders to cut spend, vendors, or both. AI is the last easy line item in a tightening budget, and a recession would turn it into the designated savings program overnight.
What happened to cost-cutters in past recessions?
The HBR study's losers followed a script worth reading closely, because it will sound reasonable when someone proposes it. The authors call them prevention-focused firms. They cut R&D, marketing, and people across the board and defended margin while demand was weak. After the recession, their sales grew 6% a year and profits 4%, against 13% and 12% for the firms that kept investing.
Sony shows what that looks like in one company. In the 2000 downturn it cut its workforce by 11%, R&D by 12%, and capital spending by 23%. Margin improved from 8% to 12%. Sales growth fell from 11% a year before the recession to 1% after it. The cuts worked, in the narrow sense that the spreadsheet said they would, and the company stalled anyway.
Samsung ran the opposite play through 2009. Bain's Beyond the Downturn analysis describes it doubling down on R&D while competitors cut, filing four times as many US patents, and holding marketing spend through the worst of the crisis. The first Galaxy phone shipped in 2009, into the teeth of the recession. Samsung entered that period ranked No. 21 on Interbrand's global brand-value list and now sits at No. 6.
McKinsey asked the same question of the 2008 cohort and reached the same answer with different data: about 10% of 1,100 large companies came through materially stronger. At the trough in 2009 their earnings had risen 10% while peers lost nearly 15%, and by 2017 their cumulative shareholder-return lead had passed 150%.
One more finding explains why the damage lasts. NBER research on the 2008 crisis found that financially stressed firms cut basic research first, and that the cut persisted after the crisis ended, because rebuilding skilled teams costs more than keeping them. For a service firm in 2026, the equivalent of that research line is AI capability: the learning hours, the workflow redesign, the rework before the payoff. A downturn will tempt you to kill precisely that line, and the evidence says it is the one to protect.
AI cost-cutting is already failing where it was tried
There is no need to wait for a recession to grade the cost-first version. Klarna announced that its AI assistant was doing the work of 700 customer-service agents and shrank its support staffing to match, then started moving people back into support. CEO Sebastian Siemiatkowski's own postmortem: "We went too far." The cost focus, in his words, reduced the quality of the company's offering and eroded customer trust. His conclusion, from the man who made the original announcement: "in a world of AI nothing will be as valuable as humans."
The counterexample sits in the same Gartner survey that predicts more Klarnas. Gartner expects that by 2027, half of the companies that cut customer-service staff citing AI will rehire for similar roles under different titles. But the same survey found only 20% of service leaders had cut staff because of AI at all. The majority kept headcount flat while serving more customers with the same team. Those firms get no headlines, and they are quietly running the play this article recommends.
Even where cost was the whole point, the savings are coming in light. In Bain's 2026 survey of 951 companies, firms that targeted 11% to 20% cost reductions mostly landed between 0% and 10%, and 90% raised their AI budgets anyway. Missed targets, rehiring bills, and eroded service quality: that is the preview of the 21% path, delivered before the recession has even started.
The measured gains are a capacity story
Field studies of AI at work keep producing the same shape of result, and the shape matters more than any single number.
| Study | Work measured | Average gain | Biggest winners |
|---|---|---|---|
| NBER, 5,179 support agents | Issues resolved per hour | +14% | Novices: +34% |
| 758 BCG consultants with GPT-4 | Consulting tasks | +12.2% tasks, 25.1% faster, +40% quality | Lowest scorers: +43% |
| MIT writing experiment | Professional writing | Time down 0.8 SD, quality up 0.4 SD | Weakest writers; skill gap halved |
| GitHub Copilot trial | Coding task | 55% faster | Task completion up 78% vs 70% |
Four studies, four kinds of work, one pattern: the biggest gains go to the least experienced people. AI raises the floor of a delivery team, which means the same people can carry more work at higher quality. A firm that reads these numbers and concludes "fewer people" has kept the smaller half of the effect and discarded the larger one.
The consultant study also supplies the caveat that keeps this honest. On tasks beyond the model's competence, consultants using AI got worse, dropping from 84% accuracy to the 60s and 70s. Somebody has to know where that boundary sits for your services, and that somebody is a person on your payroll.
Why do firms see no results from AI? The saved hours evaporate
Between those study numbers and what most firms report sits a gap that needs explaining. Three of the largest surveys of AI adoption agree on its size.
| Survey | Share seeing real results |
|---|---|
| McKinsey State of AI, 2025 | 39% attribute any profit impact to AI, mostly under 5% of earnings |
| Deloitte State of AI, 2026 | 40% have cut costs, 20% have grown revenue, 74% still hoping |
| BCG value-gap study, 2025 | 60% see hardly any material value; the top 5% get five times the revenue gains |
My hypothesis for the gap is boring. Firms bolt AI onto existing functions, save hours, and lose them. Any owner who has watched a timesheet knows the mechanism: unassigned hours leak. Time freed by an automation dissolves into the workday exactly the way unbilled time dissolves between projects, a real cost that no report ever shows. Six months later leadership looks at a flat P&L and concludes AI does not work, when what failed was the absence of a plan for the hours.
The stakes run past any one firm, and I want to flag this as hypothesis rather than forecast. The OECD now lists lower-than-expected returns from AI investment among the triggers that could reprice markets and weaken demand. The St. Louis Fed calculates that AI-related categories supplied 39% of US GDP growth in the first nine months of 2025, a larger share than the equivalent categories held at the dot-com peak. Chain those together: evaporated hours show up as "no results," aggregated "no results" disappoint investors, and disappointed investors are a named candidate cause of the downturn that would then push firms into AI cost-cutting. The failure mode helps create the conditions for its own repeat.
For an owner the instruction is smaller and more useful than the macro story: have a plan for the saved hours before the savings exist.
The playbook: cut the fat, not the capability
If revenue drops, cut. The winners in the HBR data cut too. What separated them was what they cut and in what order, so here is the sequence I would run.
Go deep on the P&L before touching AI or people. We all carry fat from good times: tools nobody opens, vendor stacks that grew by accretion, low-margin work kept out of habit, processes that exist because they always have. A firm that has never audited its own cost to deliver has no business cutting capability first. The first 10% usually lives here.
Set the cutting order in advance and hold it. Customer experience, service quality, and the technologies that improve both go last. In the HBR data, firms that protected these areas grew sales at 13% after the recession against 6% for the deep cutters. Write the order down while conditions are calm, because orders drafted mid-panic have a way of inverting.
Double down on AI in delivery, deliberately. Well-integrated AI belongs to how the service gets produced, the same as your senior staff and your process documentation. Keep pushing for efficiency through automations, workflows, and agents on the routine load. Treat the spending as capability, and remember the NBER finding: capability cut in a squeeze stays cut, because rebuilding costs more than maintaining ever did. That capability is a place on the Delivery Model Ladder: this playbook is written for firms at Stage 3, AI-native, or climbing toward it, where revenue and quality have already come loose from headcount.
Assign the freed hours before they exist. Every automation that saves 20 hours a month should arrive with a decision about where those 20 hours go. In a downturn my answer is retention: customer experience, service-delivery improvement, repeat purchases, the unglamorous work of keeping clients who are under pressure themselves and being courted by your cheaper competitors. Share held in a bad year is bought at a discount.
Play offense. Downturns are when market position changes hands. In Bain's study of 700 US firms through the early-90s recession, more than a fifth of bottom-quartile companies jumped to the top quartile during the downturn itself, and over 70% of the firms that gained ground held it through the following boom. Dell grew unit sales 11% in 2001 while its industry shrank 12%. Competitors who chose the cost-cutting story are degrading their own service quality on a schedule you can almost set a watch by. Their clients will notice before they do. Be the firm those clients call.
Have a plan B. Offense needs a floor. Decide now which costs go if revenue falls another notch, so the worst case triggers a list instead of a debate. A firm with a pre-committed worst-case list never reaches the panicked meeting where someone proposes cutting the thing that makes clients stay.
The decision is being made now
A downturn will not change your AI strategy. It will reveal it. A firm that already treats AI as part of delivery will cut fat, hold service quality, and take share from competitors busy proving the 21% statistic again. A firm that treats AI as a cost story will cut people, watch quality slip, and rehire in 2028 under new job titles, right on Gartner's schedule.
The choice between those two firms gets made in a quarter like this one, while choosing is still cheap. Write the cutting order. Assign the saved hours. If the downturn never arrives, the cost of being wrong is a tighter firm with stickier clients, which is the kind of mistake worth planning for.
If you want these pieces as they publish, the newsletter is the easiest way to follow along.
