Quo vadis, consulting

Consultants sell the future. They tell boards what is coming, how fast, and what has to be done now. That is what they write for. McKinsey, BCG, Bain, Deloitte, PwC, EY, KPMG, Accenture, Capgemini, Roland Berger and many others publish studies, outlooks and articles every week.

I wanted to know what those texts say. Not one of them, all of them. So I collected them: 41,594 texts from 28 firms, from 2019 to October 2026, from their public pages and the web archive. I counted without a language model, using fixed word lists. Where I used a model to pull out quotes, I checked every quote against the original text.

The result is not a scandal. It is a pattern. And it says more about the future of consulting than any study by the consultancies themselves.

What they write about

The first number is the simplest. In 2022, 15 per cent of the texts were about AI. In 2026, it is 48 per cent. No other topic has grown so large so quickly.

The others made room. Sustainability fell from 26 to 12 per cent, the pandemic from 22 to 2. Logistics, energy, retail, cars: all smaller. The firms are becoming more alike in what they write about.

For comparison: in 2025, one in five EU companies with ten or more employees used AI, and in Germany, according to Eurostat, just over one in four. So the texts run ahead of reality, by a factor of about one and a half to two. That is their job. Consultants write about what they want to sell, not about what is already there.

What is interesting is when they drop a topic. Sustainability held steady at just over 20 per cent for four years. Then it fell, in exactly the quarter in which the EU postponed its reporting rules. Geopolitics peaked not with the war in Ukraine but with the US tariffs. And on inflation and gas prices, the texts react rather than anticipate: inflation moves at the same time, gas two quarters behind.

So the topics do not follow the world. They follow whatever clients currently have a budget for.

Every wave gets a name

Within AI, the words change. First came the metaverse, in just over eight per cent of texts in early 2023, almost gone today. Then ChatGPT. Then “generative”, in a quarter of all texts at the end of 2024. Now the wave is called “agentic”.

Agentic stands at 22 per cent and has not yet peaked. Accenture used the word first, at the end of 2024. Capgemini, Deloitte and McKinsey followed a quarter later. Roland Berger arrived almost two years after that.

Every wave promises the same thing, only bigger. And every wave needs new consultants to explain what the new name means.

What they do at home

Now the cross-check. I searched for every text in which the firms write about their own industry and AI at the same time. There are 50 out of 10,579. I read all of them.

Most are author boxes, awards and promotional blocks. In some, the firms advise their clients to bring in consultants. Only three texts talk about how AI is changing the firm's own work. None of them questions the business model. One of them, by Bain in August 2026, says AI is not replacing consultants, and that the firm is even hiring more.

Outside the articles, it sounds different. In interviews, press briefings and quarterly results, the firms talk about their overhaul quite openly:

• McKinsey has around 40,000 people and 25,000 AI agents, according to managing partner Bob Sternfels. Client-facing roles grew by a quarter, according to reports of his remarks, and the others shrank by about a quarter. Roughly a quarter of McKinsey's fees are now based on outcomes rather than time, according to its UK head.
• Accenture had around 12,000 fewer employees over three months in the summer of 2025, down from 791,000 to 779,000. By the end of 2025 the restructuring programme had cost about 920 million dollars, mostly severance. Chief executive Julie Sweet said the firm was parting with people for whom reskilling was “not a viable path”. A year later, Accenture is back at around 814,000 employees, 110,000 of them AI and data specialists.
• The Big Four in the UK cut graduate places from 2023 to 2024: KPMG by 29 per cent, Deloitte by 18, EY by 11, PwC by 6. Deloitte Germany hired around 1,500 graduates in 2024/25, down from around 2,000 two years earlier.
• Deloitte Australia refunded part of a government contract worth around 440,000 Australian dollars. The report contained invented quotes and sources. The revised version names Azure OpenAI as a tool.

All of this is public. It just isn't where the firms advise their clients about the future of work.

This is not an accusation that they are sleeping through the change. They are not. The finding is a different one: what they preach to clients as a duty, they negotiate for themselves in press briefings.

What the market says

In Germany, the Federal Association of German Management Consultants, the BDU, asks its members every spring what they expect for the current year. A year later, the result comes in. You can put the two side by side.

For 2024, consultancies expected growth of 9.8 per cent. They got 4.3. For 2025, they expected 6.4 per cent. They got 0.5. After inflation, the market shrank in 2025, by about 1.7 per cent in real terms. For 2026, they expect 4.5 per cent again.

The experts in demand and workforce planning misjudged their own. Many firms hired heavily in 2021 and 2022 and then cut back, McKinsey for instance from over 45,000 to around 40,000 people.

What grows is technology. AI consulting grew by 18.8 per cent in 2025, and the BDU expects 22 per cent for 2026. IT consulting grew by 2.8 per cent in nominal terms, while organisation and process consulting shrank. Worldwide, Source Global Research puts spending on technology consulting in 2026 at over 400 billion dollars, growing seven to eight per cent.

Worldwide, too, things are slowing. According to Lünendonk, the 20 largest consultancies grew by only 3.0 per cent in 2025, after 7.3 the year before. At Deloitte, PwC, EY and KPMG, consulting most recently grew more slowly than the rest of the business. Accenture lost more than half its market value within twelve months, Capgemini almost 40 per cent.

So clients are not cutting consulting as a whole. They are cutting the part that works like temporary labour: analysts who build slides, gather data, write minutes. And they are buying more of what they cannot do themselves. It is a shift, not an exodus. The BDU's president, Iris Grewe, puts it this way: over the next few years, many consultancies will find out “whether they will be agents of transformation or its victims”.

Who they hire now

Look closely and you do not see a hiring freeze. You see a swap.

BCG has grown to 33,500 employees and names exactly whom it brought in: AI engineers, data scientists, IT architects, industry specialists. Almost 4,000 people there build AI workflows. By BCG's own definition, more than 40 per cent of revenue comes from AI and technology. At Accenture, the business of running systems for clients overtook classic consulting for the first time in fiscal 2026.

In Germany, job ads in consulting fell by almost ten per cent in the third quarter of 2025. Ads mentioning AI rose by almost 50 per cent over the same period. In the US, there were over 7,300 ads for “forward deployed engineers” in the first nine months of 2026, developers who sit with the client and build AI in. In 2023 there were fewer than 200.

So they are not looking for fewer people. They are looking for different ones. The graduate with good grades and great stamina who builds slides for three years is no longer the norm.

The new competition

And then come the people whose product the consultants sell.

OpenAI offers companies its own engineering teams that move in with the client. In May 2026, OpenAI set up a separate company for this with over four billion dollars in starting capital. Among the backers: McKinsey and Capgemini. Anthropic, the company behind Claude, has set up an AI services company for mid-sized firms with Blackstone, Hellman & Friedman and Goldman Sachs, with around 1.5 billion dollars in commitments. With the investor Hg, it is planning a consultancy with offices in London and Berlin.

The consultancies are partners and competitors at once. Accenture, BCG, Capgemini and McKinsey roll out OpenAI's agent platform at clients. Deloitte is giving Claude to 470,000 employees, KPMG to 276,000.

How quickly this reaches prices shows in a small case. KPMG International pushed down the fee of its own auditor by around 14 per cent. The reason given: AI.

Where the money is made

One detail stands out. In the industry, restructuring counts as a growth driver. The BDU reports 10.9 per cent growth in turnaround consulting. It names defence as a new source of revenue. In the firms' texts, restructuring appears in 2.1 per cent, and defence also in 2.1 per cent.

That is no secret either. It is just a different order. They write about what they would like to sell. They also earn from what they would rather not write about.

Numbers you pick

Now it gets more precise. Search the texts for numbers and you find what look like wild contradictions. How many companies get real value from AI? BCG writes in 2026: six per cent. KPMG writes in the same year: 76 per cent.

I read both numbers in their paragraphs. They do not contradict each other. They measure steps. BCG only counts companies with demonstrably more revenue or lower costs. KPMG asks executives whether they “see” value. At the same step, the two firms are close, about a factor of one and a half apart.

With AI agents it is the same, only steeper.

The number depends on the threshold. And which threshold a firm quotes is the message. In 2026, BCG almost always quotes the top: hardly anyone gets value out of it, you need an overhaul. KPMG quotes the broad step: most already see a benefit, keep going.

Sometimes the threshold moves within a firm. In November 2025, BCG reported that 35 per cent of companies were “exploring” agentic AI. In December, 35 per cent were “using” it. In September 2026, 35 per cent had it “in production”. The same number from the same study, read three different ways.

McKinsey lowers its own bar. In 2022 and 2023, an “AI high performer” was a company that attributed at least 20 per cent of its operating profit to AI. In 2025, five per cent is enough. Someone who counts as a leader today would not have made the cut three years ago.

Quietly corrected

Texts on the web can be changed. So for 1,141 articles with a verified number, I compared the oldest version in the web archive with today's.

Most changes are harmless, typos and footnotes. Some are not. McKinsey's well-known 2023 study on the economic potential of generative AI originally gave 0.2 to 3.3 percentage points a year for productivity. Today it says 0.5 to 3.4, under the same publication date. I found no note of the correction in the text.

At Deloitte, a number that was wrong by a factor of 500 stayed online for three weeks: 1.1 billion working hours a year became over 11 billion a week. Capgemini and Accenture deleted forecasts without replacement. In no case is there a correction note.

And some texts get younger. At Accenture, 37 of 106 checked articles carry a page date later than their first version in the archive. At BCG, not a single one does.

The consultants sound like their product

I did not look for the last pattern. I noticed it while counting.

Since spring 2026, certain words have been rising at the same time in almost every field. “Judgment”, “constraint”, “layer”, “durable”, “rarely”, “treated as”. Before, they appeared in about one per cent of texts. Now in five to 16 per cent, in banking just as in geopolitics.

These words share no topic. They share a tone. It is the tone you know from texts written or edited with language models. I cannot prove it. A word does not reveal who typed it.

But it fits the picture. The firms that sell AI to their clients increasingly write with it. They just don't say so.

Quo vadis

So where is consulting heading? The data shows no collapse. It shows three movements.

The bottom is thinning. The work through which young consultants used to learn their craft is increasingly done by agents: research, slides, data upkeep. The pyramid is losing its base. Whoever starts today gets fewer places and fewer hours of practice.

Payment changes. Whoever bills by the hour loses when a machine saves the hours. That is why the big firms increasingly bill by outcome. That sounds fair. But it shifts the risk. And it rewards projects whose outcome is easy to measure, such as cutting costs and cutting jobs.

At the top remains what needs a person. The signature under a recommendation. The conversation with a board that has to make a hard decision. The responsibility when it goes wrong. That cannot be handed to agents. And that is exactly what consultants were always brought in for, too: someone who stands behind it with their own name.

The question is whether the industry can still do that when nobody at the bottom learns how to get there.

I do have one request. Every week, the firms write about how other industries must change because of AI. I would like to read one text in which they write that about themselves. With numbers, with step and threshold, with a correction note.

How I counted

The corpus covers 41,594 texts from 28 consultancies, 37,764 of them dated, over 31 quarters up to October 2026. They were loaded from public sitemaps and from the web archive. McKinsey, PwC and a few others block direct requests. There I did not get around the block but used the latest archived version. From July 2026 the archive is thin, so the figures for the current half-year are less robust.

I count topics with fixed word families, without a language model. Every firm is weighted equally: first the share per firm and quarter, then the mean across all firms with at least ten texts. Boilerplate such as author boxes and promotional blocks was removed first. A text can have several topics.

For the numbers and quotes, a language model pulled out candidates. I only used quotes that appear verbatim in the text, with a number in the sentence and as the firm's own statement. I read the comparisons on value from AI and on agents by hand, paragraph by paragraph. Numbers only stand side by side when question, group and year are the same. Otherwise I show them as a ladder, with the wording of the threshold.

Later changes: for 1,141 articles, I compared the oldest archived version with status 200 against today's text, matched paragraphs and compared numbers in otherwise identical paragraphs. I read all 50 cases with a changed number by hand; about 35 were noise. The date of each change is narrowed down through the archived versions.

The figures on McKinsey, Accenture and the Big Four come from press briefings, results and media reports, not from my count. McKinsey's 40,000 employees are a rounded figure from its managing partner. Accenture's 12,000 is the fall in headcount over one quarter, not an official number of layoffs. The UK graduate figures go back to a single primary source, The Telegraph; later reports give slightly different values for Deloitte and PwC. The figures on outcome-based fees and roles at McKinsey come from media reports on press briefings, not from McKinsey itself. Market figures for Germany: BDU Facts & Figures; inflation: Destatis. World market: Lünendonk ranking 2026 and the firms' annual results. Job ads: Index Research via WirtschaftsWoche, Lightcast via CNBC.

Text and analysis: Jonas Höttler. Based on 41,594 publicly available texts from 28 consultancies, as of October 2026. The cover image and chapter pictures are concept images, AI-generated with Bria.