Skip to content

Strategy & Business

Quo Vadis, Consulting? What 41,594 Consulting Texts Reveal About Hype

Consultancies write about what sells. That makes their texts a good measuring instrument for hype. I counted 41,594 of them: metaverse, ESG, ChatGPT, now AI agents. Every wave rises steeply and falls again. And I looked at what the industry itself is doing right now, in the middle of the biggest wave.

Jonas Höttler27 min read

What matters most

  1. Consulting texts are a measuring instrument. Consultancies write about whatever companies have budget for right now. So if you count 41,594 of their texts, what you mainly see is what sells in business at the moment.
  2. Every wave rises and falls. New normal, metaverse, ESG, ChatGPT: most terms lost a third to more than half of their attention within the first year after their peak. AI now appears in almost one in two texts. Sustainability has halved.
  3. Numbers are part of the offer. Whether 4 or 89 percent of companies get value from AI depends on the rung being quoted. Some numbers were quietly changed. If you read studies, you should know how to check them.
  4. The industry is in the wave itself. The firms barely write about AI in their own work, 3 of 10,579 texts. Their financial results, on the other hand, show the biggest overhaul in a long time: less entry level, more engineers, more operations, fees by outcome.
  5. Where it's heading. The slide pyramid has no future. The consulting that stays runs systems or is liable for results. At the end are five dated forecasts I can be measured against.
Contents15 sections

Two weeks ago I wrote a polemic about classic management consulting. Pointed, with a trigger, with an opinion. This text is the counterpart: less opinion, more counting. And it is only partly about consultancies.

Consulting texts are a surprisingly good measuring instrument. Consultancies live off selling topics. So they write early and a lot about whatever companies are releasing budget for, and they stop when the budget is gone. Count their publications over the years and you see less of what the firms think than of what sells in business right now. A seismograph for hype.

Since October 17, 2026, I have collected everything 28 consultancies have published about the future since 2019: studies, outlooks, articles, white papers. McKinsey, BCG, Bain, Roland Berger, the Big Four, Accenture, Capgemini, Oliver Wyman and nineteen more. 41,594 texts in total. Everything was counted without a language model, with fixed word lists and each firm weighted equally, so that KPMG with its many short announcements doesn't set the average. Where a model helped find quotes, I then checked every quote against the original text. The details are at the end.

So, two questions. How does hype play out in business, measured by what the firms write about? And what does that mean for the future of management consulting itself, which this time stands in the middle of the wave it sells?

Half of It Is AI Now

Bar chart: share of consulting texts with an AI reference per year. 2019 18 percent, 2020 21, 2021 17, 2022 15, 2023 21, 2024 30, 2025 36, 2026 48 percent, the last bar orange. Above it, dashed, the share of EU enterprises using AI: 13.5 percent in 2024 and 20 percent in 2025.
The texts run 1.5 to 2 times ahead of actual use. The rise starts about three quarters after ChatGPT.

In 2022, one in seven of the firms' texts referred to AI. In 2026 it is almost one in two. For comparison: in 2025, one in five EU enterprises with ten or more employees used AI, in Germany a little more than one in four, according to Eurostat. So consulting writes far more about AI than its clients use it. That's its job. It is supposed to think ahead. What's interesting is what had to make room.

Comparison of all major topic areas, share of texts in 2022 and 2026 as lines. AI rises from 15 to 48 percent. Highlighted, sustainability falls from 26 to 12 and automotive from 8 to 3 percent, both more than halved. Regulation 15 to 22, supply chains 19 to 21, digitalization 20 to 18, industry 16 to 14, skilled labor 12 to 12, banking 10 to 10, inflation 8 to 7, geopolitics 3 to 6, M&A 6 to 6 percent.
Eleven topics stay roughly where they were, or decline. One explodes.

For four years, sustainability held steady at just over a fifth of all texts. Then it fell, precisely in the first quarter of 2025, when the EU postponed its CSRD reporting obligations. A topic that had counted as a strategic must for years lost half its attention in step with a deadline extension from Brussels. This shows how closely the topics track clients' budgets: when the obligation goes, demand goes, and the texts go with it.

The bigger the firm, the stronger the pull. At the large firms almost one in two texts has an AI reference, at the mid-sized specialist consultancies one in six. The specialists write about their specialty. The big ones write about the same thing. Put the insights pages of five big firms side by side in 2026 and you read the same idea five times in five corporate designs.

Photograph: a tall, slightly crooked stack of identical white reports without any label on a dark conference table. Only from the top one does an orange bookmark stick out.
Lots of paper, little difference. Image: generated with Bria.

Every Wave Gets a New Name

The best way to see how such a wave runs is over time. I counted 23 terms that were at some point the next big thing, quarter by quarter, from blockchain to 5G to ESG.

Curves over time, 2019 to 2026, share of texts per quarter. Four gray waves rise steeply and fall again: new normal peaking at 14.8 percent in 2020, metaverse 8.0 percent in late 2022, ChatGPT 6.2 percent in early 2024, ESG 12.1 percent in late 2024. On the far right an orange curve for agentic AI rises to 26.6 percent and keeps rising.
Inflated, then shelved. On the right, the wave that is running now.

The pattern is always the same. A term appears, rises steeply within a few quarters, reaches a peak and falls. In mid-2020, “new normal” appeared in almost one in seven texts. Today it appears in fewer than one in a hundred. The metaverse reached 8 percent in late 2022, today 0.4. ChatGPT as a word peaked in early 2024 and has since fallen to a third, even though the technology behind it keeps growing. It just goes by a different name now.

Hype graveyard: peak vs. today. Agentic AI 26.6 percent and rising. Below it new normal 14.8 to 0.8 percent, ESG 12.1 to 4.5, 5G 11.7 to 3.7, net zero 9.5 to 1.8, metaverse 8.0 to 0.4, blockchain 7.5 to 1.6, RPA 7.3 to 1.4, ChatGPT 6.2 to 2.1, Web3 4.0 to 0.2 percent.
What was once the next big thing, and what is left of it.

For most terms, a third to more than half of the attention was gone just one year after the peak. Only net zero held for a year before it fell. No firm has explained what became of its clients' metaverse strategies. None has to. The texts follow demand, and demand had moved on.

Now the wave is called “agentic.” A good one in four texts mentions agentic AI or AI agents, still rising. Accenture started in the fourth quarter of 2024. Capgemini, Deloitte and McKinsey followed in the first quarter of 2025, BCG in the third, Bain, Oliver Wyman and PwC in early 2026. Roland Berger arrived in the third quarter of 2026, almost two years after Accenture. Whoever brings the word first usually has the matching product on the shelf. That's exactly why the texts work as a measuring instrument.

What's telling is where the firms don't think ahead. I set the share of texts about inflation against the inflation rate in the euro area. The two move at the same time. The correlation is 0.63. With gas, the firms write the most two quarters after the price spike. Geopolitics peaked not at the invasion of Ukraine but at the US tariffs. The firms react, they don't foresee. For the measuring instrument, that means it reliably shows what is in demand right now. It doesn't show what comes next.

Consulting texts show less of what the firms think than of what sells right now.

Between 4 and 89 Percent

The nicest question you can ask consulting studies: how many companies actually get real value from AI? The answers from 2024 to 2026 range from 4 to 89 percent. That sounds like chaos, but it follows a ladder.

Eleven bars in three groups. Top, revenue or costs measurably changed: BCG 2024 4 percent, BCG 2025 5, McKinsey late 2025 6, BCG 2026 7.5, KPMG early 2026 11 percent. Broad, some value: BCG 2024 26, 2025 40, 2026 48.5 percent. Self-reported, see value: KPMG 64, 76 and 89 percent.
Not a contradiction, but rungs of the same question.

At the top are the companies whose revenue or costs have measurably moved: 4 to 11 percent, depending on firm and year. Below them, everyone reporting some value: 26 to almost 50 percent. At the bottom, the self-reports on whether executives “see” a benefit: 64 to 89 percent. On the same rung, by the way, the firms are close together, about a factor of 1.5. The data hardly contradict each other.

The message lies in the rung a firm quotes. In 2026, BCG quotes the top in eight out of ten articles with a global figure: almost nobody gets value, a rebuild is needed. In 2026, KPMG quotes only the broad rung: the majority sees benefits, now it's about scaling. Both are backed by their own data. And both fit strikingly well with what each firm offers in the next paragraph.

KPMG itself shows how elastic such a number is. In the Global AI Pulse for the third quarter of 2026, 89 percent of respondents see measurable value. The same report says only 12 percent calculate that value against the costs across the company. And the report for the first quarter has, on the same page, 59 percent who deploy agents and 9 percent who “currently already” use agents “in their workflows.”

McKinsey, in turn, has lowered its own bar. In 2022 and 2023, an “AI high performer” was anyone who attributed at least 20 percent of their operating profit to AI. In 2024 the threshold was more than 10 percent, at the end of 2025 it was 5 percent. Anyone who counts as a leader today would not even have made it into the analysis three years ago. This makes progress over the years hard to read, and none of the reports says so.

At BCG, a number simply wandered:

Three cards with arrows. November 18, 2025, press release: 35 percent are exploring agentic AI. December 10, 2025, article: 35 percent use it. September 17, 2026, article: 35 percent have it in production. Each time the same survey with 2,102 participants.
Same study, same number, three meanings. In the same half-year, Capgemini counts 2 percent with agents across the whole company.

In November 2025, 35 percent are exploring agentic AI. In December, 35 percent use it. In September 2026, 35 percent have it in production. Each time the same survey by BCG and MIT Sloan Management Review. In practice, between “we're looking into it” and “it's running in our operations” there are months to years and usually a six-figure budget. Here, what lay in between was an editorial change.

A number says little as long as it doesn't say what exactly was counted.

Quietly Corrected

Numbers change. That's normal. What's interesting is how. For 1,141 articles with a verified figure, I set the oldest version in the web archive against today's text. Most differences are harmless: footnotes, numbering, bios. About a dozen are not.

Changed without notice · Archive vs. todayold → new
McKinsey

“The economic potential of generative AI,” 2023: productivity gain of 0.2 to 3.3 percent per year. Today: 0.5 to 3.4. Publication date unchanged.

Lower bound more than doubled.
Deloitte

GenAI in Asia Pacific, 2024: “1.1 billion hours per year.” Three weeks later: “over 11 billion hours per week.”

Factor of about 500.
Deloitte

AI insurance, 2024: AI could add “US$12.5 billion” to global GDP by 2032. Later: “US$12.5 trillion.”

Factor of 1,000.
Accenture

Gen AI talent study, 2024: 86 percent of executives consider gen AI transformative. Later: 97 percent.

Two different questions, according to its own infographic.
Capgemini

AI in robotics, April 2026: 65 percent expect scaling within five years. Early May: deleted.

Forecast removed.
No correction notice

In none of these cases does the page carry a note such as “updated” or “corrected.” Errors happen wherever things get published fast, in newspapers just as much as here. The difference is the missing notice. To be fair: most changes came shortly after publication. They are corrections of errors, not reinterpretations after the fact. But an error by a factor of 500 that stays online for three weeks and then quietly disappears was quoted during those three weeks. Anyone who picked it up will never learn it was wrong.

Then there's the date. At Accenture, 37 of 106 articles now carry a publication date well after their first appearance in the archive. At EY 21 of 72, at Deloitte 11 of 224. At BCG, Bain and McKinsey not a single one. A text about “ambient computing” from Deloitte's Tech Trends 2015 is now dated December 10, 2025. Some of this may be genuine revision. To the reader, an eleven-year-old outlook looks just like a fresh one.

And the forecasts themselves? In 472 texts, BCG made 2,231 statements about the future, each checked word for word. Among them is exactly one forecast of its own with a number and a target year up to 2026. Lots of future, little you can check. The checkable ones that do exist miss by factors on AI. In 2019, McKinsey forecast $67 billion in revenue for AI semiconductors in 2025. What came, according to Gartner, was over $200 billion for AI processors alone. To be fair to McKinsey: in 2019, ChatGPT was not foreseeable. The lesson applies to every number with a target year on AI. They miss by factors.

The industry also misjudges its own market, by the way. The German industry association BDU expected growth of 9.8 percent for 2024. It came in at 4.3. For 2025 it expected 6.4 percent. It came in at 0.5. For 2026 it expects 4.5 percent. The expectation comes from a survey of the consultancies themselves.

Paired bars on the German consulting market: 2024 expected 9.8 percent growth, result 4.3. 2025 expected 6.4, result 0.5. 2026 expected 4.5 percent, result due in March 2027.
Clearly off two years in a row. The check-back for 2026 comes in March 2027.

The Mirror Test

Photograph: an empty dark blue suit on a hanger in front of a tall standing mirror. The mirror does not show the suit but a server rack with small orange lights.
What hangs in front, and what shows in the mirror. Image: generated with Bria.

Now the real question. For three years the firms have been telling their clients that AI is upending business models, organizations and job profiles. What do they write about their own?

In all texts with an AI reference, I searched for passages where the consulting industry, the consulting business or management consulting appears near an AI term. 50 hits in 10,579 texts, all read by hand. Most are author boxes, awards and promotional blocks. Six recommend that the reader bring in consultants. Three texts talk about how AI is changing their own work. None questions its own business model. The clearest is a Bain podcast from August 2026, titled “AI isn't replacing consultants.”

On the left, a funnel through the expert texts: 10,579 texts with an AI reference, 50 mention consulting and AI together, 3 talk about AI in their own work, 0 question the business model. On the right, four statements from press interviews: McKinsey with 40,000 people and 25,000 AI agents, a quarter of fees tied to outcomes, PwC Germany: I have destroyed our business model, Accenture: parting ways where reskilling is not a viable path.
Left, what the texts say. Right, what the same firms say elsewhere.

Outside the expert texts it sounds very different. In press interviews, on stages and in quarterly results, the firms talk openly about their own overhaul:

  • McKinsey. About 40,000 people and 25,000 AI agents, CEO Bob Sternfels said in January 2026. A spokesperson confirmed the figure. A year and a half earlier it was about 3,000 agents. Client-facing roles grow by a quarter, the others shrink by a quarter. And about a quarter of global fees now depend on outcomes, according to UK head Michael Birshan.
  • PwC Germany. In June 2026, partner Jan Herrmann opened a talk with the sentence: “I have destroyed our business model!” („Ich habe unser Geschäftsmodell zerstört!“) A project that used to take six to eight weeks and cost €150,000 to €200,000 would in future be possible in 24 hours. “The pyramid is now turning into more of a diamond.” („Die Pyramide wird nun eher zum Diamanten.“)
  • Accenture. The workforce fell in one quarter of 2025 from 791,000 to 779,000. CEO Julie Sweet: the firm is parting ways “on a compressed timeline” with people for whom reskilling is “not a viable path.”
  • BDU. The association's vice president: “Anyone who still shows up at a client without AI tools today looks like someone arriving by horse-drawn carriage.”

The firms didn't sleep through it. They just discuss it elsewhere: at conferences, in press interviews, in quarterly results. That's understandable. Expert texts are sales, and you don't sell anything with your own overhaul. For readers, this means: if you want to know how seriously a consultancy takes AI for itself, read its financial results rather than its studies.

Expert texts are sales. If you want to know how seriously a firm takes AI for itself, read its financial results.

Who Comes, Who Goes: Headcount and Revenue 2026

What the expert texts don't tell, the annual reports show. The results for fiscal 2026 have been out for a few weeks.

Fiscal year 2026: growth and headcount
FirmRevenueGrowthWhat stands out
Accenture$74.2bn+5%Consulting +3%, operations (managed services) +6%. About 814,000 employees, about 110,000 of them for AI and data.
Deloitte$74.5bn+3.8%Technology and transformation only +2.5%.
PwC$60.2bn+3.5%Advisory +3.0%, tax and legal +6.3%.
EY$57.0bn+4.7%Headcount Americas −4.9%, delivery centers +12.3%. Revenue from AI services +49%.
BCG$14.4bn (2025)+7%AI and tech services over 40% of revenue. Headcount +500 to 33,500.
McKinseynot disclosedHeadcount from about 45,000 (2022) to about 40,000.

Growth in local currency, BCG in dollars. Sources: the firms' annual reports and press releases, McKinsey per Bloomberg and Business Insider.

Three things stand out. First, at Deloitte, PwC and KPMG (there fiscal 2025), it is consulting of all things that grows the slowest. Tax and legal do better. Second, at Accenture, operations grows faster than consulting. There is more money in running systems than in saying which ones to build. Third, workforces are growing again, but elsewhere. In one year, EY has almost 4,500 fewer positions in North and South America and a good 10,000 more in its global delivery centers.

Paired bars on growth in fiscal 2026: Accenture consulting 3 percent vs. operations 6, PwC advisory 3.0 vs. tax and legal 6.3, KPMG advisory 2.9 vs. total 5.1 in fiscal 2025, Deloitte technology and transformation 2.5 vs. total 3.8 percent.
Where there are figures by segment, consulting trails. BCG and EY-Parthenon are the counterexamples.

At entry level it becomes obvious. Job postings from Deloitte, PwC, EY, KPMG, McKinsey, BCG and Bain fell by about a quarter in the year to May 2026, according to an analysis by Draup for Business Insider. In the same period, open positions for so-called forward deployed engineers, engineers who sit with the client and build systems, rose from zero to 1,404. In the UK, the Big Four cut entry-level positions between 2023 and 2025: KPMG by 29 percent, Deloitte by 18, EY by 11, PwC by 6. In Germany, consultancies' job postings fell by almost 10 percent in the third quarter of 2025, while postings with an AI reference grew by half, according to Index Research as reported by WirtschaftsWoche.

Bars around a zero line: EY headcount Americas minus 4.9 percent, global delivery centers plus 12.3. Job postings of the Big Four and the three big strategy consultancies minus 24 percent. Entry-level positions in the UK 2023 to 2025: KPMG minus 29, Deloitte minus 18, EY minus 11, PwC minus 6 percent.
Less entry level, fewer staff in expensive countries, more in the delivery centers.

At the same time, the tech side grows. McKinsey's AI unit QuantumBlack has, by its own account, “well above 5,000” AI specialists. BCG X has almost 3,000 technologists. Bain says it is hiring “more, not less,” just different people.

Photograph: an empty open-plan office in a high-rise at night. Rows of white desks with dark screens, only one monitor in the back corner glows orange. In a glass room hangs a whiteboard with a faded pyramid.
Fewer seats for tomorrow's analysts. Image: generated with Bria.

One thing matters for context. The firms themselves mostly don't blame cuts on AI, but on low attrition, weak demand and costs. KPMG UK doesn't mention AI at all in its announcement of more than 500 eliminated roles. That may be true. It may also be convenient. Either way, the pyramid gets narrower at the bottom and wider at the top, and PwC has already supplied the word for it: diamond.

What nobody answers: if fewer analysts start today, who will be partner in 2035? In Lünendonk's survey, 58 percent of German management consultancies name the generational change among partners as the biggest hurdle. Training young talent was always the invisible part of the business model. Clients paid for it through the juniors' day rates. If that goes away, the middle will be missing in ten years.

What They Shed, What They Buy

If you want to know where a company is going, don't read its strategy papers. Read its purchase and sale agreements.

Divested and acquired, 2025 to 2026
OutIn
McKinsey sells its in-house asset manager MIO Partners to Neuberger Berman.Accenture buys 17 companies for $4.9bn in fiscal 2026. For 2027, about $5bn plus cyber acquisitions are planned.
PwC parts ways with member firms in nine French-speaking African countries, and according to the FT also in Zimbabwe, Malawi and Fiji.Capgemini buys WNS for $3.3bn: business processes run on contract, with agents.
Porsche sells its IT consultancy MHP to Tata Consultancy Services (closing pending).Roland Berger founds a company that places restructuring chiefs in firms, and as sole investor funds Jonas Andrulis's AI start-up.
PwC Australia hands its restructuring unit to Teneo.McKinsey and Capgemini take stakes in the OpenAI Deployment Company.
Deloitte is cutting up to 40 jobs in sustainability assurance in Germany, about 30 percent of the unit, according to Handelsblatt.
Accenture and IBM no longer report their AI bookings separately.

The last row is the most interesting. Until the end of 2025, Accenture proudly reported every quarter how much revenue and bookings came from generative AI. Since early 2026, it no longer does, because AI is now in almost everything. IBM has also stopped publishing its figure for the GenAI business. That may be reasonable. But it also means that the metric disappears at exactly the moment you could have checked whether the AI business makes up for shrinking classic consulting. At the same time, in its 2026 annual report Accenture warns that results could suffer if clients “do not accept new pricing or commercial models.”

Sustainability shows up again too, this time as cuts. It fell from 26 to 12 percent in the firms' texts, and now the teams are shrinking. What the industry described for years as a megatrend was, for the industry, mostly a trend in its order book.

The pattern is clear: away from side businesses and slides, toward operations, software, restructuring and stakes in exactly the companies competing with them.

The New Competition Sits With the Model Makers

In May 2026, OpenAI and Anthropic founded their own service companies. The OpenAI Deployment Company is meant to help businesses with adoption, with about $4 billion in starting capital. OpenAI holds the majority, and the backers include TPG, Goldman Sachs, SoftBank, Capgemini and McKinsey. Anthropic founded a standalone firm with Blackstone, Hellman & Friedman and Goldman Sachs whose engineers work with Anthropic staff to build AI solutions in mid-sized companies, such as community banks, mid-sized manufacturers and regional health systems. Both do what the consultancies want to sell right now: engineers at the client who build systems.

The consultancies react in two ways. They take stakes, like McKinsey and Capgemini at OpenAI. Or they become sales partners. Accenture has its own Anthropic business unit with about 30,000 trained people, and KPMG is giving Claude to more than 276,000 employees. The firms that tell their clients to limit their dependence on single vendors are building their own business on partnerships with one or two model makers.

And the clients? The Financial Times asked in August 2026. Bristol Myers Squibb now has AI handle its IT security monitoring, “and those contracts are going away.” SAP says its changes will cut the cost of external consultants by up to 50 percent. Bayer is using 30 AI agents for coding and testing in its SAP overhaul. UniCredit spent 24 percent less on external consultants in the first half, Société Générale 9 percent less. According to Source Global, 55 percent of clients want to use the Big Four more next year. A year earlier it was 80 percent. Only one in three clients says an IT project led by consultants was fully successful.

The stock market has already priced it in. At the end of August, Accenture's share was 27 percent below its level at the start of the year, Capgemini's 31 percent.

To keep it fair, the other side of the ledger: overall the business is growing. Source Global estimates global spending on technology consulting in 2026 at $420 billion, 8 percent more than the year before. Clients are reallocating: less labor for hire, more technology and implementation.

Germany: Standstill, As Announced

In 2025, the German market grew nominally by 0.5 percent to about €49 billion. With 2.2 percent inflation, that's a decline in real terms. Firms with more than €50 million in revenue stagnated at zero. Headcount still grew by 1.5 percent, according to the industry news service consulting.de, so faster than revenue.

Lünendonk titled its 2026 ranking “Consolidation instead of growth” („Konsolidierung statt Wachstum“). Six of the twenty largest German management consultancies lost revenue, including Porsche Consulting (−2.2 percent), Horváth (−1.0) and zeb (−4.0). Revenue per head fell, even though AI supposedly makes everything more efficient. Roland Berger, on the other hand, reported the best year in its history, with just over €1 billion. The IT consultancies grew organically by only 1.5 percent, for the first time in a long while slower than the IT market as a whole.

On pricing, AI has moved little so far. According to Lünendonk, 43 percent of management consultancies see no effect on their prices, and 16 percent bill AI services separately. BDU president Iris Grewe put it this way in March: “For many consultancies, the next few years will decide whether they become an agent of transformation or its victim.”

And what is making money in Germany right now? AI consulting, which grew by 18.8 percent in 2025 according to the BDU. Turnaround and restructuring, which led the BDU's business climate index at the start of the year. Public sector contracts and defense. The strange part: restructuring appears in 2.1 percent of the firms' texts, defense also in 2.1 percent. What they earn from, they barely write about. What they write about, they earn little from so far.

What consultancies earn from, they barely write about. What they write about, they earn little from so far.

They Sound Like Their Product

One last finding, and it's the strangest. I counted which words grew most in the texts from April to October 2026 compared with 2024 and 2025, separately by topic area.

At the very top are everyday words: “judgment,” “constraint,” “layer,” “durable,” “rarely,” “treated as.” “Judgment” rose in AI texts from 1 to 8 percent, in texts about work from 1 to 10. “Layer” in texts about banks from 1 to 16 percent. “Constraint” in geopolitics, logistics and IT each from zero to about 9. The same words rise at the same time in fields with no thematic link: supply chains, insurance, energy, inflation.

Paired bars, share of texts before vs. April to October 2026: layer in banking texts from 1 to 16 percent, layer in inflation texts 1 to 12, durable in energy 1 to 11, judgment in work 1 to 10, constraint in geopolitics 0 to 10, rarely in insurance 0 to 9, judgment in AI 1 to 8 percent.
Seven words, seven fields, the same jump.

That points to a change in style, not in topics. The pattern fits texts written or revised with language models. It isn't proven. A word proves no origin. But it would be a punchline of its own: the industry that explains to its clients how AI changes their work apparently already lets its own texts sound that way. They just don't mention it.

And it has a consequence beyond the punchline. If all firms write about the same topic, with the same numbers on different rungs, in the same tone, then thought leadership is no longer a differentiator. Then it's noise. And any client can now produce noise on their own, for about €20 a month.

Quo Vadis: Where Consulting Is Heading

So where to? My reading, with the caveat that forecasts on this topic age just as badly as the firms' own.

The pyramid isn't getting narrower. It's breaking off at the bottom. The work that whole classes of analysts used to be hired for, collecting data, building slides, calculating market models, AI does in hours. What remains is the diamond: few juniors, a broad middle of experts and engineers, partners at the top. That's efficient. It has a price nobody shows today. The firms no longer train their own young talent, and their clients no longer pay for it along the way.

The money moves from advice to operations and liability. Accenture earns more from operations than from consulting. McKinsey bills a quarter of its fees by outcome, and PwC wants to get there. That's the actual good news. A firm that is paid for results has to measure results. And anyone who has to measure results can no longer afford numbers that wander from “exploring” to “in production” depending on the paragraph.

The justification service loses its value. Part of consulting was always legitimation: the report supplies the reasoning for a decision that has already been made. That works as long as nobody checks. Today someone checks. Clients, journalists, audit offices and language models put the studies of three firms side by side in a minute and see who quotes which rung. If you're buying a justification, you can get it cheaper now.

The new competition comes from above. OpenAI and Anthropic are building their own service companies, Palantir showed the way, and private equity is funding specialized firms. The big consultancies become partly partners, partly resellers, partly competitors of these companies. Which role goes to whom will be decided in the next two years.

What's left is not little. Good consulting is close to operations, measures its impact and stands behind it. Research shows that exactly this kind of consulting works. It just never was what the pyramid made its money with.

So that this doesn't turn into the same inconsequential future prose I just counted, here are five forecasts with dates:

Five forecasts, with a date to check back
ForecastCheckable againstCheck back
The German consulting market grows by less than 3 percent in 2026, and the BDU expectation of +4.5 percent is missed for the third year in a row.BDU Facts & Figures 2027March 2027
At Accenture, managed services again grows faster than consulting in fiscal 2027.Accenture, FY 2027 resultsOctober 2027
The share of consulting texts with “agentic” falls below 15 percent by the end of 2027, and a new term takes over.My own count, same methodJanuary 2028
At least one more big firm drops one of its own AI metrics from its reporting or changes its definition.Annual reports, study pagesDecember 2027
In mid-2027, the big firms' job postings for entry-level hires are below the mid-2025 level, those for engineers above it.Job portal analysesJuly 2027

If I'm wrong, I'll write it down here. With the reason.

Five Questions for Every Study on the Next Big Thing

For everyone who reads, quotes or pays for such studies. They apply to consultancies just as much as to vendors, associations and analysts. And they matter most when a topic shows up in one in two texts.

  1. Which rung? Does it say “see benefits,” “some value” or “measurably changed revenue and costs”? Only compare numbers when the rung, the group and the year are the same.
  2. Who was asked? All companies, only large ones, only those already using AI, only “pioneers”? In 2018, Deloitte surveyed only companies testing or using AI, and arrived at 82 percent with financial benefits.
  3. When was the text last changed? Put the address into the web archive. It takes a minute and shows whether the number always read that way and whether the date is right.
  4. Where is the forecast from three years ago? Ask for the firm's last prediction on the same topic and what became of it. If they can't show you anything, they have nothing to show.
  5. What is the publisher selling alongside this number? If there's an offer two paragraphs later, the number is part of the offer. That doesn't make it wrong. You should still read it like advertising.

In Closing

Since 2019, the consultancies have written about the future 41,594 times. Read as a measuring instrument, these texts mostly tell a story about all of us: how fast business declares a topic a must, and how fast it drops it again. New normal, metaverse, ESG, ChatGPT. Now AI agents.

For companies, that's a useful warning. When a topic shows up in one in two texts, the pressure is at its highest and it's the worst moment for hasty purchases. The better question is what it measurably changes in your own operations.

And consulting itself? This time it stands in the middle of the wave it sells, and it is rebuilding itself like it hasn't in a long time. Quo vadis, consulting? To where it gets measured.

When a topic shows up in one in two texts, the pressure is at its highest and it's the worst moment for hasty purchases.

For context

How I counted. A corpus of public pages from 28 firms, 2019 to October 2026, collected via sitemaps and the web archive (Internet Archive). Topics via fixed word families, each firm weighted equally: 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 language model helped find quotes. I used only quotes that appear verbatim in the text, and every number in this article has been checked in the original.

What the numbers can't do. The selection of texts follows keywords in the URL path. It is not random. From July 2026 on, the stock is thinner because the web archive lags behind. An ongoing comparison with the sitemaps closes the gap. A word is a hit, not an assessment. Simultaneous doesn't mean causal. McKinsey and PwC block direct requests, so for them the version comes from the archive.

What the texts stand for. I read the publications as a proxy for what sells in business right now, not as a verdict on the quality of individual firms. The hype curves count 23 terms per quarter, as a rolling average over two quarters. The topic areas use fixed word families and annual averages. That's why individual values differ slightly, depending on whether a single word or a whole field is counted.

Where the market figures come from. From the firms' annual reports, regulatory filings and press releases, from BDU and Lünendonk, and from the Financial Times, Bloomberg, Business Insider and WirtschaftsWoche. Where only a secondary source was available, it is noted below. A shrinking workforce is not the same as layoffs. Where it's about the workforce, it says workforce.

What's in it for me. I'm not neutral. With Balane I build software and AI systems in Munich, and with KI-Managementsolutions we help companies introduce AI. If big firms lose contracts, I may benefit. That's why the dated forecasts are up there and all the sources are down below.

Sources

My own analysis

  • Corpus of 41,594 texts, 28 firms, as of Oct 25, 2026. Archive versions via web.archive.org, each finding documented with a timestamp.
  • Hype terms: 23 terms per quarter, count of Oct 26, 2026, rolling over two quarters, each firm weighted equally.

The firms' studies (value from AI, agents)

  • BCG: “Where's the Value in AI?” Oct 2024. bcg.com · “Why AI pilots rarely deliver value,” Aug 11, 2026. bcg.com · “The authorization gap,” Sep 17, 2026. bcg.com
  • KPMG: Global AI Pulse Q1 2026. kpmg.com · Q2 2026. kpmg.com
  • Capgemini: “Making AI real,” Sep 22, 2026. capgemini.com · Bain: “How do companies create value with AI,” Jun 15, 2026. bain.com
  • McKinsey: “The economic potential of generative AI,” archive version of Jun 14, 2023. web.archive.org · Version of Mar 11, 2026. web.archive.org
  • McKinsey 2019 on AI semiconductors. mckinsey.com · Gartner, semiconductor revenue 2025, Jan 12, 2026. gartner.com
  • Bain podcast “AI isn't replacing consultants,” Aug 3, 2026. bain.com

Financial results 2026

  • Accenture, FY 2026 results, Oct 1, 2026. sec.gov · Annual report (10-K) FY 2026. sec.gov
  • Deloitte, FY 2026 (per CPA Practice Advisor, secondary source). cpapracticeadvisor.com
  • PwC, FY 2026, Oct 19, 2026. pwc.com
  • EY, FY 2026, Oct 8, 2026. ey.com
  • BCG, revenue 2025, Apr 23, 2026. bcg.com
  • Capgemini, half-year 2026. capgemini.com · Capgemini buys WNS. capgemini.com
  • Accenture, last separate AI figures in Q1 FY 2026, earnings call Dec 18, 2025 (Motley Fool transcript). fool.com
  • IBM drops its GenAI metric (Trefis, secondary source). trefis.com

Headcount and entry level

Divestments, acquisitions, new competition

  • McKinsey sells MIO Partners (Bloomberg). bgov.com
  • PwC, exit of the French-speaking African countries. pwc.com
  • Accenture, acquisitions FY 2027 (Channel Dive). channeldive.com
  • OpenAI: “OpenAI launches the Deployment Company,” May 11, 2026. openai.com · Funding and investors (S&P Capital IQ via MarketScreener). marketscreener.com
  • Anthropic: enterprise AI services company with Blackstone, Hellman & Friedman and Goldman Sachs, May 4, 2026. anthropic.com · blackstone.com
  • Anthropic and Accenture. anthropic.com · Anthropic and KPMG. anthropic.com
  • Financial Times: “Consultants head for a showdown with their own clients,” Aug 31, 2026 (via Irish Times). irishtimes.com

Germany

  • BDU: Facts & Figures 2026, Mar 27, 2026. bdu.de · consulting.de on headcount development. consulting.de
  • Lünendonk ranking 2026, management consulting. PDF · Study 2026, preview. PDF · IT services market 2026. luenendonk.de
  • Roland Berger, record revenue 2025, May 28, 2026. mynewsdesk.com · AI start-up with Jonas Andrulis, Feb 12, 2026. mynewsdesk.com
  • Porsche sells MHP to TCS, Aug 24, 2026 (Deutsche Börse, EQS announcement). deutsche-boerse.com
  • Handelsblatt: Deloitte cuts jobs in sustainability auditing, Oct 5, 2026, based on company sources. Original behind paywall, excerpt: autobound.ai
  • Eurostat, use of AI in enterprises 2025. ec.europa.eu

Images. Photos generated with Bria (FIBO model), charts my own.

Related Articles

Jonas Höttler builds software for mid-sized companies with Balane.