McKinsey Should Hire McKinsey to Fix McKinsey

As it enters its second century, the world's most famous consulting firm is shrinking, losing ground to BCG, and counting AI agents as part of its workforce.

In late October 2025, McKinsey gathered thousands of its people in Chicago to kick off the firm’s centennial celebrations. Bob Sternfels, the global managing partner, delivered a rallying speech about the second century. There were, one imagines, balloon arches. Possibly a cake shaped like a 2×2 matrix. Away from the festivities, managers in some non-client-facing functions were being told to prepare for a leaner organisation.

The firm that taught corporate America how to restructure, re-engineer, and rightsize entered its centennial year applying the playbook to itself. The workforce had already fallen from more than 45,000 at the end of 2023 to about 40,000. McKinsey named 224 new partners in November 2025, roughly 44% fewer than the class of 2022. Bloomberg reported discussions about reducing headcount in non-client-facing departments by a further 10% over the following 18 to 24 months.

McKinsey attributed much of the reduction to attrition and tougher performance reviews. I’ve sat through enough partner meetings to recognise the language. When a client told us their 10% headcount reduction was “natural attrition,” we’d nod politely and then put a slide in the deck about organisational honesty. (The slide was titled “The Attrition Myth: A Framework for Candid Workforce Planning.” It had a little iceberg graphic. The client loved it. They then fired 800 people and called it a “talent refresh.”)

The numbers McKinsey doesn’t volunteer

The Economist recently asked whether McKinsey has “fallen below fighting weight.” Their business editor, Tom Lee Devlin, laid out the basics: revenue growth of about 2% in 2024, BCG growing at roughly 10%, the gap closing fast. He was being diplomatic.

In 2012, McKinsey was roughly twice BCG’s size. McKinsey reported $16 billion in revenue for 2023; more recent reporting has placed its annual revenue at roughly $15 billion to $16 billion. BCG, by contrast, reported $13.5 billion in 2024 and $14.4 billion in 2025. Whatever the exact private-company figure, the gap has narrowed sharply.

For most companies, being the largest in a market by a shrinking margin is manageable. For McKinsey, it is an identity problem. The firm’s value proposition to clients, to recruits, and to the partnership itself rests on being the default choice when the stakes are highest. “We hired McKinsey” carries weight in a boardroom precisely because McKinsey is understood to be number one. That sentence loses force when BCG is growing three to five times faster and the crossover is a matter of arithmetic. (A simple extrapolation of the 2024 growth rates implied BCG could overtake McKinsey around 2027, though BCG’s growth slowed to 7% in 2025. The gap is still closing; the timeline is less certain.)

The revenue stagnation follows a decade of expansion that has now clearly overshot. McKinsey went from 17,000 employees in 2012 to 45,000 by late 2023, acquiring a succession of analytics, design, cloud, and digital-product businesses and pushing hard into implementation work. The firm was trying to compete with Accenture and the Big Four on delivery, which meant hiring thousands of specialists in software engineering and cloud implementation where McKinsey had no traditional strength.

When the growth slowed, those specialists went first. In 2023, the firm cut roughly 1,400 back-office roles. In 2024, McKinsey cut approximately 400 more in areas such as data engineering and software.

The firm cut its data engineers and software developers while intensifying its pitch as an AI-first firm.

BCG did what McKinsey tells its clients to do

What makes the comparison painful is that BCG succeeded by following the advice McKinsey sells: invest in capability before you need it, retain your specialists, commit to the transformation rather than hedge. BCG essentially walked into McKinsey’s own library, checked out The McKinsey Way, and actually did the homework. McKinsey wrote the book, framed the cover, and then fired the people who could implement chapter four.

BCG played a similar expansion game, building out its digital and technology offerings and standing up BCG X with more than 3,000 technologists. The difference, according to the Economist and multiple other reports, is that BCG did a better job retaining the specialists who make technical advisory work credible. BCG says AI- and technology-focused services represented more than 40% of total revenue in 2025, with AI services growing 25% year over year.

McKinsey’s equivalent is QuantumBlack, with approximately 1,700 people. McKinsey describes a broader technology community of roughly 6,000, plus another 4,000 to 5,000 consultants leading technical work from the business side. Those numbers matter, because McKinsey’s own AI leadership has said that the 4,000-to-5,000 group does not necessarily possess deep technical expertise. That admission is more revealing than any outside criticism could be. It’s like a restaurant announcing, “We have 12 chefs, plus 40 people who have watched a lot of cooking shows and can describe what a béarnaise sauce should taste like.”

QuantumBlack is not a vanity project. Its engineers build data pipelines, deploy models into production, and integrate systems into client operations. The firm has published credible deployed work in pharma, telecommunications, and manufacturing. But the gap between QuantumBlack’s genuine capability and how the broader partnership sells “AI transformation” to clients is wide enough to notice from the outside. From the inside, it is impossible to miss. You can smell it. It smells like a partner who learned the word “vector database” on the flight to the client meeting.

The talent problem nobody in the partnership will say out loud

The Economist covered the financials well. What it cannot cover, because its reporters don’t sit in the partner meetings, is the talent problem underneath them.

The best AI talent in the world is not going to consulting firms. The competition for people who can build, deploy, and scale AI systems is the most intense talent market in a generation. ManpowerGroup’s 2026 survey of 39,000 employers across 41 countries found that AI skills are now the single hardest capability for employers to find, topping every other skill category for the first time. The people who have those skills can work at OpenAI, Anthropic, Google DeepMind, Palantir, or any of hundreds of well-funded startups. They can start their own companies. They can name their compensation.

Why would they go to McKinsey?

The consulting model pays well at the junior level but is not competitive with what top AI engineers earn at technology companies. At the top end of the US AI market, senior individual contributors receive total compensation above $500,000 without entering a management track. More importantly, the work is different. A Palantir forward-deployed engineer builds production systems embedded in client operations. A McKinsey engagement, even one with QuantumBlack involvement, typically produces recommendations, architectures, and pilot implementations that the client’s own teams (or a technology vendor) must then scale. The career paths diverge further up: in tech, a senior ML engineer can stay technical indefinitely. In consulting, the path to partner-level compensation requires becoming a salesperson, managing client relationships, and surviving partnership politics. Your reward for being the best engineer on the team is that you never get to engineer anything again. Your new job is to play golf with a man named Greg and pretend to care about his boat.

The result is a firm that has genuine AI capability in QuantumBlack and the broader tech community, but whose AI-labelled work across the wider partnership relies heavily on generalist consultants whose technical depth does not match how the work is sold. I have seen this configuration firsthand. Partners who were selling “digital transformation” in 2019, “ESG transformation” in 2021, and “AI transformation” in 2024, with the same skill set and a new cover page on the same deck.

I want to be specific about this because it goes to the core of the AI credibility problem. The consulting partnership model produces people who are exceptionally good at learning a domain’s vocabulary in two weeks and presenting confidently on it in three. That skill is valuable for many types of advisory work. It is inadequate for AI, because AI clients are increasingly technical buyers who can tell within fifteen minutes whether the person across the table has ever trained a model, debugged a data pipeline, or shipped a feature. The partner who pivoted from ESG to AI cannot pass that test. They know it. The engagement manager staffing the team knows it. The only person who might not know it is the CEO who approved the statement of work, and even that gap is closing as in-house technical leadership gets involved in vendor selection. (The CEO will find out eventually. Usually at 11 p.m. on a Sunday, when the CTO sends a Slack message that begins, “So I just got off the call with McKinsey, and I have questions.”)

McKinsey’s own AI leadership has said that QuantumBlack “drives” all the firm’s AI initiatives and that AI-related work accounts for roughly 40% of what McKinsey does. If that is true, the question becomes whether QuantumBlack’s specialists are consistently attached to the engagements sold under the AI banner, or whether the firm’s commercial machine is selling AI work faster than the specialist bench can staff it. The public headcount arithmetic suggests the latter. This is a restaurant selling 400 steaks a night with a kitchen staffed for 80. At some point, someone is getting a very creative interpretation of “medium rare.”

Sternfels gave the game away in early 2026 when he described McKinsey’s workforce as “roughly 60,000,” including about 25,000 AI agents alongside roughly 40,000 humans. The headline agent is Lilli, an internal generative chatbot that lets consultants search the firm’s knowledge base and generate slides from prompts. McKinsey says about 72% of employees use Lilli and that it saves up to 30% of time spent searching and synthesising knowledge. That may well be true. But an internal chatbot that helps your own people produce slides faster is a productivity tool, not evidence that your workforce includes 25,000 additional “employees.” Every major consulting firm has deployed similar tools. Every client has access to the same foundation models. Describing your agent count as headcount is the kind of framing McKinsey would tell a client never to use publicly.

The competition McKinsey can’t acquire its way out of

The competitive picture is more complicated than a simple “builders versus advisors” split, but the direction of travel is clear.

Palantir grew its revenue by 48% year over year in Q2 2025, and by 85% in Q1 2026. Its forward-deployed engineers embed with clients, build production systems, and iterate on working software. A CFO who has paid McKinsey $3 million for an AI strategy that recommends hiring specialists to do the actual implementation has paid for a very expensive shopping list. A beautiful shopping list, with a proprietary framework and a memorable acronym. But a shopping list.

OpenAI launched its Deployment Company in 2026, staffing it with forward-deployed engineers to embed in enterprises. Interestingly, OpenAI also named McKinsey and BCG as Frontier Alliance partners for strategy, operating-model design, and change management. That alliance complicates the binary: McKinsey is not simply being displaced by technology companies; it is being repositioned as one layer in a stack where the technology partner owns the engineering. Whether that layer remains premium-priced is the central question. It is a little like being told you are still the most important person at the restaurant, but your new title is “the man who suggests which wine goes with the meal someone else cooked.”

Accenture announced its acquisition of Faculty, a British AI company, in January 2026, adding more than 400 AI professionals. Bain expanded its Palantir partnership in March 2026 rather than try to build AI delivery capability from scratch. (Bain had first partnered with Palantir in May 2025; the expansion itself tells you something about how the build option looked after six months of examining it.)

Enterprise AI delivery is becoming an alliance market. Frontier-model companies, platform vendors, engineering specialists, systems integrators, and management consultancies are combining capabilities. McKinsey’s challenge is to prove that its part of the stack produces enough measurable value to remain premium-priced, and that its genuine technical capability reaches the engagements sold under the AI banner.

The Economist’s Devlin made the useful observation that strategy projects bundle together two things: deep cogitation and a huge amount of grunt work (crunching numbers, preparing slides) that is usually done by the most junior people. AI compresses the grunt work, which is where the junior leverage margin lives. McKinsey is reportedly moving about a quarter of its engagements to outcome-based pricing, which suggests some recognition that the hours-for-dollars model is under pressure. Whether that transition is fast enough, or genuine enough, is the kind of question McKinsey would ask its clients. It has been less willing to answer it about itself.

The recommendation McKinsey would give itself

If McKinsey were its own client, the engagement would write itself. Any competent strategy team (including McKinsey’s own) would identify the same issues within the first two weeks of diagnostic interviews.

The firm expanded aggressively into implementation and delivery work, then cut the technical specialists it had hired for that expansion while intensifying its AI positioning. Its primary competitor invested more effectively in the same capabilities and has been growing three to five times faster. Technology companies are entering its market with delivery capabilities it cannot match independently, and McKinsey is being absorbed into alliance models where it occupies the advisory layer rather than the engineering layer.

The reputational bill is substantial. McKinsey had agreed to $1.6 billion in opioid-related settlements and resolutions before an additional $125 million Purdue agreement was approved in April 2026. McKinsey Africa separately agreed to a $122.85 million criminal penalty over bribery in South Africa. The opioid fallout damaged McKinsey’s federal position, freezing it out of FDA work and contributing to a sharp decline in prime-contract revenue. Saudi Arabia, which Bloomberg-derived reporting described as paying McKinsey at least $500 million annually in the decade leading up to 2024, has reportedly pared back consulting payments. Republican lawmakers have alleged that McKinsey failed to disclose work for Chinese government and state-controlled entities while holding more than $480 million in Defence Department contracts; McKinsey has said it does not work for the Chinese Communist Party or China’s central government. Each of these carries cost, reputational drag, or both, and they accumulated during the same period that commercial growth stalled.

And the internal incentive structure rewards partners for selling rather than building, which prevents the firm from making the capability investments its own analysis would recommend.

The recommendation would be obvious. Stop rebranding existing partners as AI experts and invest in hiring and retaining people with genuine technical depth, even if that means a different compensation structure. Accept that the leverage model (billing clients for junior hours while the partner shows up for the kickoff and the final presentation) needs rethinking when AI can do what the juniors used to do. Engage with the reputational drag directly rather than waiting for it to fade, because it is not fading. And restructure the partnership incentives so that building long-term capability is rewarded as highly as short-term revenue origination.

McKinsey’s partners would nod along to every one of those recommendations. They would commission a beautiful deck summarising them. They would agree in principle and then table the implementation discussion until the next partner meeting, where it would be tabled again, because every one of those recommendations threatens someone’s current compensation or status. The politics that block change at McKinsey’s clients are the same politics that block change at McKinsey. The deck would be gorgeous, though. The font would be impeccable. The recommendations would be numbered, prioritised, and colour-coded by urgency. And then the deck would go into a folder called “Final_v7_ACTUAL_FINAL(2).pptx” and never be opened again.

I spent nearly two decades in large consulting firms telling clients they needed to transform or die. The irony of watching these institutions resist the same advice has a flavour you don’t forget. It tastes like a $400-per-night hotel bar, a cold club sandwich, and the slow dawning realisation that you are billing someone for the privilege of watching them not listen to you.

Happy birthday. Maybe call a consultant.

McKinsey entering its second century is a firm that can still produce a brilliant report on why every other company needs to change. The partners are, individually, among the sharpest people in professional services. The institution’s problem is the one McKinsey diagnoses in other institutions every week: the incentive structure prevents the organisation from acting on what it knows.

McKinsey’s problem is not that it failed to acquire AI capability. QuantumBlack has real builders doing real work. The problem is that the institution may be unable to let that capability remake the partnership at the speed its own advice demands. The firm kicked off its centennial celebrations while quietly planning the largest headcount reduction in its history, counted its AI agents as employees, cut its data engineers while repositioning as AI-first, and watched its closest competitor close the revenue gap using the strategy McKinsey would have recommended to a paying client.

Somewhere, in a McKinsey office, there is a partner working on a case study about a century-old institution that failed to adapt to technological disruption despite having all the information it needed. I hope someone tells them to check the mirror.

---

Current or former McKinsey people: when did you first notice the gap between how the firm positions its AI capability and how that capability is staffed on actual engagements? And for clients who’ve hired McKinsey for AI work: was QuantumBlack on the team, or was it generalist consultants with a new vocabulary and a very confident handshake?