The major consulting firms spent the past three years telling clients that AI would change their businesses, and in 2025 and 2026 the same technology started changing the firms. The evidence is now in the firms' own headcount and statements rather than in forecasts. McKinsey's global managing partner Bob Sternfels said in January 2026 that the firm's non-client-facing staff has shrunk by about a quarter even as its client-facing side grew by roughly the same share. In the same period KPMG cut about four hundred US advisory jobs, Accenture said on its September 2025 earnings call that it is exiting people for whom reskilling on AI is not a viable path, and PwC's annual report showed global headcount down 5,600 in a year, the firm's first contraction since 2010, as the Financial Times reported. These are four cost decisions at four large firms, and the common thread is a pricing model that AI has put under pressure.
What the firms actually sell
A consultancy's economics rest on the leverage model: a small number of partners sell engagements, and a large base of junior staff delivers the hours those engagements require. Revenue targets, utilization rates, and partner compensation are all calibrated to that pyramid of billable time. The product the client receives is analysis, a recommendation, and an implementation plan, and the work is priced as a function of how many qualified people spent how many hours producing it. When Accenture's chief executive Julie Sweet states that promotion now requires using AI, the instruction is aimed at the input side of that equation. The firm is asking its people to produce the same deliverable with fewer hours.
That instruction sits in direct tension with the revenue model, and the tension is arithmetic rather than cultural. If a research memo that took an analyst forty hours now takes eight, and the firm bills by the hour, the firm has cut the price of that memo by eighty percent without deciding to. Either it bills the old number of hours for work that no longer requires them, which clients eventually notice, or it bills the new number and the revenue per engagement falls. The junior roles that performed the forty hours are the roles most exposed. The pyramid was built to convert junior hours into partner profit, and the conversion rate has changed.
Why the pricing model resists the fix
The obvious response is to stop billing for hours and start billing for results, so that faster delivery does not mean a smaller invoice. The firms know this. McKinsey's managing partner for the UK, Ireland and Israel, Michael Birshan, said in November 2025 that the firm is doing more performance-based arrangements with its clients, and Business Insider reported from the same briefing that about a quarter of McKinsey's global fees now come from outcome-based pricing, with the rest billed traditionally. That figure is the center of the problem. A firm that has spent decades pricing the human hour cannot reprice three quarters of its book overnight, because outcome-based pricing requires the firm to carry delivery risk it has never had to carry, and to have a deliverable concrete enough that an outcome can be measured against it. A slide deck does not produce a measurable outcome on its own. Someone inside the client still has to act on it, which is the structural reason consulting recommendations so often fail to land. We have written before about why the billable hour and the outcome are at odds, and the consulting restructuring is that conflict appearing in public payroll data.
Where the money went
The firms did not ignore AI. PwC's US firm announced a $1 billion, three-year investment in generative AI in April 2023, EY launched its EY.ai platform on the back of $1.4 billion of investment, and KPMG committed to multibillion-dollar spending on Microsoft cloud and AI over five years. The spending is real, and so is the difficulty of recovering it through a pricing model that charges for the hours AI removes. Investment in tools that make delivery cheaper is straightforward to justify when you sell software, because cheaper delivery widens margin on a fixed price. It is harder to justify when you sell time, because the same efficiency shrinks the billable base it was meant to defend. The billions buy capability the firms cannot fully monetize without abandoning the way they charge, which is the bind the headcount numbers describe.
Scale stops being the advantage
For most of the industry's history, size was the decisive advantage. A large firm could field more people, cover more geographies, and absorb more risk than a boutique, and clients paid for that reach. AI weakens the part of the advantage that rested on having the most people, and the firms entering the market now are built on exactly that change. Unity Advisory, founded in 2025 by PwC UK's former managing partner Marissa Thomas and EY UK's former chair Steve Varley with an equity line of up to $300 million from Warburg Pincus, sells CFO advisory services and says AI will play a central role in how it works and serves clients from day one. When the marginal unit of work is a model call rather than an analyst, a ten-person firm and a ten-thousand-person firm face a more similar cost of production than they used to. This is the mechanism that lets a services firm built on software grow without growing its headcount in step, and it is why the AI-native entrant and the legacy incumbent are now competing on terms that did not exist five years ago.
A reasonable counter
A reasonable counter is that consulting demand is cyclical, that firms cut staff in every downturn, and that reading an AI repricing into a soft market confuses a temporary slowdown with a structural shift. There is real truth in that. Some of the 2025 and 2026 reductions are ordinary responses to slower demand, and headcount at large firms has fallen before without any structural cause. Two things distinguish this round. The cuts are concentrated in the junior and support roles whose work AI most directly substitutes, rather than spread evenly across seniority, and the firms themselves are naming AI as the reason in their own statements about retraining and promotion. A cyclical downturn does not usually come with the chief executive telling staff that their advancement depends on adopting the tool that is reducing the need for them. The model the major firms are circling, outcome-based fees on a software-delivered deliverable, is the model an AI roll-up or a services-as-software firm starts from rather than retrofits. The difference between a firm repricing against its own incentives and a firm that never carried those incentives is what the restructuring data is actually measuring.