BCG billed $3.6 billion from AI-related work in 2025, approximately a quarter of its total revenue, Bain is expecting to have 50% of their book be AI within a few years, and McKinsey already has AI touching nearly 40% of client projects. These firms will tell you this proves they understand AI. What it actually proves though is that a rebranding operation can move fast when the client base is scared enough.
Two odd things are happening inside the big three simultaneously, and neither of them shows up in their pitch deck.
The first thing is a staffing story that comes down to a billing model.
Bloomberg found around 150 consultants from these three firms were employed to train AI models for the entire bottom half of the consulting pyramid model: the slides, the analysis, the synthesis, the 2am first drafts that someone got billed at $600 an hour. In late 2025, McKinsey laid off most of those doing those precise functions.
The intellectual capital that has historically underpinned consulting’s pyramid model is now being automated. The invoice has remained the same. The thing that clients had been paying for, access to a massive, hyper-agile, highly trained pool of human talent, has stopped existing. It’s the same product at the same price, but the production function has fundamentally changed.
If this were any other industry, someone would call bullshit. In the world of consultants, it’s operational efficiency and buried in the firm’s own marketing materials about AI adoption without any of that efficiency getting passed onto clients.
If we're being honest, the consulting pyramid model has always been a classic scheme in better suits and non-disclosure agreements. The difference now is that the bottom tier has been automated but the invoice structure remains intact, which is either the most honest thing a pyramid scheme has ever done or the most audacious.
The second thing is a physics problem which is significantly worse.
Prior consulting failures had edges. Revlon losing $64 million on a botched SAP implementation had a shape. There were decisions that could be pointed to, people involved, and technical debt to be blamed. A failure had a face.
This does not. AI has presented a cost structure that operates like a broken law of thermodynamics; the strategy itself decays even as it’s being written, with the initial assumptions expiring before the invoice can clear. Not only that, but this wasn’t even apparent when most strategies were being written, as this type of risk didn’t exist at enterprise levels until that point.
AI has achieved the seemingly impossible: token prices have plummeted while corporate AI bills have exploded. When intelligence gets cheaper, you use more. Then you build systems that require more. Then those systems generate edge cases that require even more. The ceiling disappears, and with it, any meaningful way to budget.
Uber provided its engineers with AI coding tools in December 2025 and burned through the company's entire annual AI budget by April. Every dollar by only one quarter into 2026.
The average enterprise OpenAI API spend hit $384,000 by that same month, having more than doubled year-on-year. That figure doesn't even take into account the 40-60% in additional costs coming from retry logic, context window management, retrieval augmentation, and feature creep that will end up being larger than most regional marketing budgets and not visible until they show up on a quarterly finance report.
In the most appalling case, an American company handed the corporate AmEx to a probabilistic engine with no natural stopping point and called it transformation. One month later half a billion dollars had evaporated into a roaring furnace of tokens and prompts.
These issues weren't identified in the roadmap because the market doesn’t respect the consulting calendar, especially when the consulting calendar is the product. A three-year AI roadmap carrying assumptions about model capabilities, deployment reliability, unknown innovations, and cost structures has an effective shelf life of roughly a quarter before large parts of it become historical fiction. Not wrong at the time, but irrelevant on arrival.
Consulting runs on fixed cycles: 18-month engagements, six-week discovery phases, governance models built for a world where software changed slower than procurement. The gap between problems and action is engineered into that billing model.
AI doesn’t respect that clock.
Nobody appears to be accountable for being overtaken by velocity, which means nobody fixes the model that created the problem. That’s how we end up with 88% of AI agent pilots never reaching production, 40% of agentic projects headed for cancellation, 75% of executives now admitting their AI strategy is "more for show" than internal guidance, and 48% describing AI adoption as a massive disappointment (up from 34% the year before). These numbers describe a client base that paid for something and got very expensive decorations.
The firms’ clean version of this story is that this is the client’s fault. Insufficient change management, cultural resistance to adoption, too little investment, and/or not enough executive sponsorship. Pick your finger pointing. The more uncomfortable version is that it’s a redirect of accountability away from the people who delivered a strategy on assumptions with a shelf life shorter than the engagement itself.
What serious organizations need to take into account is that they've outsourced first principles technology thinking for long enough that the internal muscle for it has atrophied below useful levels. A 90-day hypothesis test against real market signals and killed or scaled on contact with actual results is the only realistic path to understanding what AI could mean for business. Any strategy that can't be substantially reworked inside a quarter is a strategy built for an unrealistic level of stability.
The big three are selling a repackaged version of the same fear mongering and alibi economics that drove the consultancy frantic rush and overspend corrections with ERP in the 90s, Y2K prepping, the internet, SOX compliance, the decade long digital transformation. Except this time, the AI timeline is revealing the solution charade cracks in almost real time.
Sadly, most boards and leadership won't figure this out themselves: they’ll only come to realize it when the same consulting firm returns this year with an urgent second engagement proposal, revised deck, and fresh new batch of invoices.
*** Views expressed on LinkedIn are personal and shared in an individual capacity. They do not represent those of any current or former employer ***