$10 Billion in AI Investment. Same Pyramid. Same Slides.
Every firm has an AI platform. Almost none have a new business model.
Between 2023 and 2025, the Big 4 and MBB collectively poured billions into artificial intelligence. By our count, the public commitments alone exceed $10 billion. PwC committed $1 billion across its US operations and became OpenAI’s largest enterprise customer, handing ChatGPT Enterprise licenses to over 100,000 employees. KPMG entered a multi-year alliance with Microsoft widely reported at $2 billion. Deloitte launched a $2 billion technology and AI program it branded “Industry Advantage.” EY invested $1.4 billion in its EY.ai platform, building an internal large language model called EYQ on top of Microsoft Azure and Azure OpenAI. McKinsey deployed an internal AI platform called Lilli, which the firm says is used by over 75% of its employees and has saved more than 1.5 million hours of search and synthesis work.
These are serious numbers. Billions of dollars, hundreds of thousands of licensed seats, executive pronouncements about reinvention.
And after all of it, the consulting business model has barely changed.
The pyramid is narrowing at the base, but nobody redesigned the structure. The billing model has started to crack in places, but three-quarters of the industry still charges by the hour. Partners still show up for kickoffs and final presentations. Juniors still build the slides. The only difference is that the juniors now use a chatbot to draft them faster, while being told this makes the firm “AI-native.”
We want to be specific about what those billions bought and what they did not, because the gap between the press releases and the operating reality is where the diagnosis lives.
What the money bought
Start with the internal tools. Every major firm now has one. McKinsey has Lilli, which indexes more than 100,000 internal documents and 40-plus knowledge sources. BCG has Deckster for slide editing. Bain has Sage, powered by OpenAI. PwC built ChatPwC on GPT-4 before upgrading to the full ChatGPT Enterprise rollout. EY built EYQ on Microsoft’s Azure stack.
These tools work. We know people at these firms who use them daily. McKinsey’s own case studies report roughly 30% time savings on information gathering and synthesis. PwC has reported substantial productivity gains from its generative AI tools. Senior consultants told us they can now produce a first-pass competitive analysis in two hours that would have taken three junior staff two weeks. The technology compresses what used to be the most labour-intensive part of engagement delivery.
None of that is in dispute. The technology is real, and in certain applications, it is good.
The question is what the firms did with those productivity gains. And the answer, when you look at the numbers, is: they kept the gains, cut the juniors, and left most of the billing model untouched.
The pyramid with fewer people at the bottom
In the UK, graduate-level accountancy job adverts fell about 44% year-on-year, with Big 4 firms cutting their intakes sharply. EY delayed graduate start dates for the third consecutive year, with some 2025 hires not starting until 2026.
At firm level, the cuts went deeper. According to executive search firm Patrick Morgan, KPMG reduced its UK workforce by 7% in 2024, PwC by 5%, Deloitte by 5%, and EY by 3%. The reductions continued into 2025, with KPMG and PwC each shrinking UK headcounts by another 4%.
McKinsey is the most candid about what is happening. CEO Bob Sternfels said in early 2026 that the firm now operates with roughly 25,000 AI agents alongside about 40,000 human employees, and expects the ratio to approach parity. McKinsey saved 1.5 million hours of search-and-synthesis work through AI in the past year. The firm’s headcount has dropped from roughly 45,000 to around 40,000 over 18 months, with the reductions concentrated in junior research, back-office, and non-client-facing roles.
Kate Smaje, McKinsey’s global leader of technology and AI, told reporters that the firm no longer needs “armies of business analysts creating PowerPoints” because the technology can handle much of that work.
So the base of the pyramid is being compressed. Fewer graduates, fewer analysts, fewer of the people who used to do the 60% of engagement work that could be described as “structured information processing.” The firms are producing comparable output with smaller teams.
Now ask the follow-up that none of the press releases address: if a team of eight can now deliver what a team of twelve used to, is the client paying for eight or twelve?
The billing model almost nobody wants to change
The answer varies by engagement, but the direction is clear. Engagement fees have not dropped to match the productivity gains. In many cases, they have increased, because the firms are repositioning themselves as “AI-enabled” and charging a premium for the label.
The logic goes like this: AI makes our teams more productive, which means the output per hour is higher, which means the value per hour is higher, which justifies the existing fee structure. This is a defensible argument on paper. It is also the exact argument that a taxi company would have made about GPS navigation in 2010: the technology makes our service better, so we should charge more per ride. Uber made that argument irrelevant.
Credit where it is due: McKinsey has moved further than most. The firm disclosed in late 2025 that about a quarter of its global fees now come from outcome-based pricing rather than hourly billing. That is a real shift, and it is worth watching. But McKinsey is the exception, not the pattern. Three-quarters of its own revenue is still traditional, and the Big 4 have barely begun to experiment with alternative pricing. The industry’s centre of gravity remains the billable hour.
The risk for the firms is that clients are starting to do the same math. A senior procurement officer at a FTSE 100 company told us last month that her team now asks a specific question during every consulting RFP: “What percentage of this engagement will be delivered using AI tools, and how is that reflected in the fee?” Two years ago, nobody asked that. Today, she said, it is a standard line item in the evaluation criteria.
KPMG may have accelerated this shift in the worst possible way. The Financial Times reported in February that KPMG International pushed Grant Thornton, the firm that audits KPMG’s own books, to cut its fees on the basis that AI should be making the audit faster and cheaper. KPMG threatened to switch auditors if Grant Thornton did not agree. Grant Thornton agreed. The audit fee dropped 14%, from $416,000 to $357,000.
The irony writes itself. KPMG acted as the aggressive procurement client, demanding an “AI discount” from its own service provider. Every one of KPMG’s consulting clients now has the playbook. If KPMG believes AI justifies lower fees when it is the buyer, why should the logic be any different when KPMG is the seller?
BCG’s numbers tell the story
Among the major firms, BCG is the most transparent about where revenue growth is actually coming from, and the picture should worry anyone who believes traditional strategy consulting has a stable future.
BCG reported $14.4 billion in revenue for 2025, up 7% from $13.5 billion the prior year. That sounds healthy. Then you look at the composition.
AI- and tech-focused services now represent over 40% of BCG’s total revenue. AI services specifically grew 25% year-over-year. The firm hired AI engineers, data scientists, IT architects, and industry specialists. It launched the BCG X AI Science Institute.
BCG does not break out the growth rate of its non-AI business, but the arithmetic is suggestive. When a firm grows 7% overall and its fastest-growing segment (already 40% of revenue) is compounding at 25%, the remaining business is not growing much. It may be shrinking in real terms. The traditional strategy and organizational work that BCG was built on, the practice that charges $500,000 for a four-week engagement to tell a CEO what three of their direct reports already know, is no longer the growth engine.
BCG’s own press release calls this “22 consecutive years of growth.” Technically accurate. But the firm is becoming a technology implementation business with a strategy brand bolted on, and the revenue composition makes that trajectory visible.
McKinsey’s pattern is similar, though the firm discloses less. Bain reports that AI- and tech-enabled work is already roughly 30% of its consulting business, projected to reach 50%. We are not saying this is wrong. Firms should follow revenue growth. (We spent decades telling clients exactly that.) What we are saying is that the press coverage treats these firms as if they are successfully integrating AI into a traditional consulting model when the reality is closer to the opposite: the technology business is cannibalizing the consulting business, and the firms are relabeling the result as intentional.
The credibility gap
These firms sell AI transformation to clients for fees that often start at seven figures. They produce reports on AI strategy. They advise boards on AI governance. They staff entire practices around helping organizations deploy AI responsibly. That is a significant revenue stream, and the firms market it aggressively.
During this same period, three of the four Big 4 firms have been caught publishing reports with AI-fabricated citations. (It’s only a matter of days before it’s four out of four - you read it here first).
Deloitte Australia delivered a 237-page government welfare report that contained fabricated references, including a quote from a Federal Court judgment that cited a judge whose name was misspelled. The report had been produced using Azure OpenAI GPT-4o. Deloitte refunded a portion of the A$440,000 contract.
Deloitte Canada submitted a 526-page health workforce plan to the Government of Newfoundland and Labrador, costing C$1.6 million, that contained at least four citations to academic papers that do not exist. Deloitte stood by its recommendations.
EY Canada published a cybersecurity marketing report in which GPTZero, the AI detection firm that has become the de facto auditor of the auditors, found that 16 of 27 citations were fabricated, misattributed, or pointed to dead links. Sixty percent. The report cited a “McKinsey & Company: Loyalty Economics Report (2022)” that does not exist. GPTZero traced this to what it calls a “secondhand hallucination”: a fabricated reference that originated in a blog post and was then laundered into the report as if it were a real source.
KPMG International published a report on AI itself, titled “Total Experience: Redefining Excellence in the Age of Agentic AI.” GPTZero found that 40 of 45 citations were fabricated, mangled, or misattributed. UBS, the NHS, Swiss Federal Railways, and Transport for London told the Financial Times that the report’s claims about their AI usage were untrue or misleading.
An AI report about AI, with AI-generated citations to AI projects that do not exist, published by a firm charging clients millions to implement AI responsibly. Read that sequence again and try to maintain the position that these firms have their own house in order.
The consistent response from each firm has been some variation of: “The substance and recommendations are unaffected.” The defence concedes more than it intends. The fabricated evidence does not undermine the conclusions, which is another way of saying the conclusions were never based on evidence in the first place. If the recommendations survive the removal of the sources that supposedly supported them, those sources were decorative. GPTZero coined a term for this: “vibe citing.” Generating references that feel plausible without checking whether they exist. As a description of the consulting industry’s relationship with evidence, it is uncomfortably precise.
As of mid-2026, PwC is the only Big 4 firm not publicly implicated in an AI-fabrication scandal of this kind.
Partner politics and the gravity well
Inside the firms, the people who understand the technology best are the most frustrated by what is happening with it. We have spoken to partners and directors at four different firms who describe the same dynamic, often using almost the same words.
One, a technology partner at a Big 4 firm, described it this way: the firm invested heavily in AI tools, trained thousands of people to use them, and then told the practice leaders to “integrate AI into delivery.” The practice leaders heard: make the existing model more efficient. They did not hear: rethink the model. Why would they? The existing model generates their compensation. A practice leader whose team of twelve is reduced to eight has just lost 33% of the billable headcount that feeds the partner’s revenue attribution. Nobody volunteers for that.
Another, a former BCG director, put it more bluntly: the partners who control the P&L of traditional strategy engagements have every incentive to use AI to make the old model faster and no incentive to replace the old model with something that might be better for clients but worse for the partner’s economics. The technology team proposes new delivery models. The practice leadership nods, agrees in principle, and then staffs the next engagement exactly the way they staffed the last one.
This is the pattern we observed with every attempted reform inside these firms. EY spent over $600 million on Project Everest trying to separate audit and consulting, and the US partners killed it in part because splitting the tax practice would have altered their economics. The same gravity applies to AI. Any change that threatens the current partnership structure gets slowed, diluted, or quietly abandoned once the press coverage fades.
What would genuine transformation look like?
If these firms were serious about AI changing their business model, you would see at least some of the following.
Engagement pricing would shift from hourly billing to fixed-fee or outcome-based models. McKinsey is experimenting with this; roughly 25% of its fees are now outcome-based. That is worth noting and worth watching. But for the remaining three-quarters of McKinsey’s business, and for the vast majority of Big 4 consulting, the billable hour persists. If AI compresses the hours, and the firms claim the value per hour has increased, then they should be willing to price on value rather than time. Most are not.
The leverage ratio would change visibly. If you need fewer juniors per partner, the pyramid should be narrowing at the base and broadening in the middle, with delivery redesigned around smaller, more senior teams augmented by AI. Some of this is happening through hiring cuts, but it is happening as a cost measure, not as a delivery redesign. The teams are smaller. The engagement structure is identical.
Clients would see the AI. Today, most clients on a Big 4 or MBB engagement have no visibility into which parts of their deliverable were produced by AI and which were produced by a person. There is no disclosure standard, no transparency norm. When we asked a partner at one firm why clients are not told, the answer was candid: “If the client knew how much of this was AI-generated, they would ask why they are paying partner rates.”
The training pipeline would be redesigned. The traditional model trained graduates through repetition: you learned to be a consultant by building slide decks, running analyses, and sitting in client meetings for two to four years until the pattern-matching became instinctive. If AI is now doing the analytical repetition, what replaces it as a training mechanism? Nobody has a good answer. UK accountancy graduate job adverts fell 44% in a single year, and the firms have not announced what replaces the on-the-job apprenticeship those roles used to provide.
If AI were genuinely transforming these firms, the operating model would look different: the fees, the staffing, the disclosure practices, the partnership economics. Instead, what has mainly changed is the cost base. AI made it cheaper to deliver the same product with fewer people. The savings went to partner compensation, not to clients and not to reinvention.
Clients are doing the math
The firms have bought themselves time, perhaps two to five years, by being early AI adopters in their own operations. But the clock is running. Procurement teams are getting better at distinguishing between “we use AI” as a capability claim and “we use AI” as a marketing claim. Boutique firms that were never built on the pyramid model can adopt AI without the partnership politics that choke change at the incumbents. Companies are building internal strategy teams and discovering that a senior hire with AI tools can replicate a large portion of what they used to buy from McKinsey.
BCG’s revenue composition is the leading indicator. The technology business is growing. The traditional business is stalling. The firms that figure out how to price, staff, and deliver for an AI-enabled world will survive in something resembling their current form. The firms that use AI to make the old model slightly cheaper, while cutting the junior staff who were supposed to be their future, will discover that they optimized themselves into irrelevance.
We spent our careers inside these firms. We know the partner politics, the compensation incentives, the gravitational pull of this-quarter’s-numbers on every strategic decision. And what we see, looking at billions in AI investment, is an industry that bought the technology and has been slow to change the business.
The pyramid is narrower at the base. The slides are faster. The business model, for the most part, is the same.
Has anyone here been on an engagement where AI genuinely changed the delivery model, not just sped up slide production? We are interested in examples where the staffing, the pricing, or the deliverable format was fundamentally different because of AI. If those engagements exist, we want to hear about them. If they don’t, that absence is the point.