AI-native firms have not taken strategy consulting's market

What is an AI-native firm, and how big are they?

An AI-native firm sells finished work, not software. Sequoia made the case in March 2026: sell the tool and you race the model, sell the work and each better model helps you^S49. Yet the same essay sizes management consulting at $300bn to $400bn (its estimate), calls the work "mostly judgement" and lists the best candidates as "TBD"^S49.

The capital is real. General Catalyst says it has committed billions to AI-enabled roll-ups, and that Long Lake has bought 18 businesses^S53. But note what the money buys. Reuters says enterprise AI still depends on skilled, labour-intensive services^S50. Eudia's legal offer rests on a team of 300-plus it acquired^S53. Bain took a stake in OpenAI's venture^S56; McKinsey did too, per StrategyU^S14.

In deal work the closest example is a US AI-native bank for small-company sales. By August 2025 it had raised under $20m, reported over 30 launched deals, and set a company target of $100m revenue by 2027^{S54, S55}. We found no audited revenue for it, and no aggregate funding figure for the category that we could verify.

What has actually happened to consulting?

Revenue held up, though advisory lags. Accenture's consulting line grew 3% in fiscal 2026 against 6% for managed services^S1. PwC advisory grew 4.2% against 6.5% for tax and legal^S17. KPMG advisory grew 2.9% against 7.5%^S19.

The UK is weaker. Source Global Research put 2024 at minus 3.4%^S22. The MCA called 2024 flat^S23, then reported 3% like-for-like growth in 2025, with 46% of members shrinking^S21. Canadian consulting revenue grew 6.3% in 2024, slowing after three double-digit years^S24.

Headcount is mixed. McKinsey fell from 45,100 to about 40,000^{S11, S12} and Bain from 18,254 to 17,519 between 2023 and 2025^S13, after pandemic hiring. McKinsey cites attrition and performance reviews^S11. Accenture rose 4.5% to 814,000^{S2, S3}, KPMG 1.8%^S19, and BCG added 500^S8. US consulting payrolls rose 1.5% in the year to September 2026, on preliminary data^S25.

Graduate hiring is split. UK Big Four intakes fell^S31, and PwC planned to cut US entry-level hiring by almost a third over three years^S33. But MCA members hired 10% more graduates in 2025^S20, BCG says it has not cut junior intake^S9, and PwC UK expects graduate numbers to grow^S32.

Fee models are moving. McKinsey says about a third of revenue underwrites outcomes^S10. BCG says three quarters of its largest AI cases carry variable fees, but "significantly less than a third" of all work^S9. TCS says about 80% of its business-services contracts are outcome-based^S59.

Investors are more alarmed than the accounts. Accenture fell 59% from peak to trough^S6, with a record one-day fall of about 20% in June^S5, then rose about 22% on 1 October on a forecast of 3% to 6% growth^{S1, S4}. That shows fear, not lost revenue.

How much of the work is exposed?

Exposure is high, which is not replacement. The published Science paper finds about 1.8% of jobs have over half their tasks affected by models alone, and just over 46% with software built on them^S35. The IMF says about 60% of advanced-economy jobs may be affected^S37. Microsoft scores management analysts at 0.353, among the top 40 occupations, and warns against reading scores as job loss^S38.

Task evidence is uneven. In the Harvard and BCG trial of 758 consultants, AI users finished 12.2% more tasks, 25.1% faster^S39. On a task outside AI's range, correct answers fell from 84% to between 60% and 70%^S40. A later trial of over 500 professionals found each year of model progress cut task time by 8%, with weaker gains on agent workflows^S48.

Capability is the live risk. On expert-written tests from Mercor, which licenses expert data to AI developers, the top consulting score rose from 64.8% in October 2025 to 87.0% now, and banking from 59.7% to 83.6%^{S42, S43}. In 2025 Mercor said no model met the production bar^S41. On harder agent tasks the top pass rate was 24.0% in January 2026^S44. An independent rerun now shows 47.1%^S46; Mercor's own board shows 84.2% on a changed task set^S45. Trust the direction more than the level.

Labour data lag this. A Danish study rules out effects above 2% on earnings or hours two years after ChatGPT^S30.

What happened last time?

Four analogues show one pattern: the routine layer repriced, and the profession changed shape.

Tax software. Hand filing fell from 41% to 13% between 1993 and 2003, software rose from 8% to 25%, and paid preparers rose from 51% to 62%^S67. Erosion came later: 53% in fiscal 2024^S70, though another study says about 60%^S71.

Discount brokers. After fixed commissions ended in 1975, institutional rates fell 60% by April 1979 and individual rates 18% (our calculation)^S72. Big buyers got the cut first.

E-discovery. A 2016 study said automation could cut lawyer employment by slightly over 13%^{S73, S74}. US legal services jobs rose 8.7% from 2016 to 2025^S26.

Robo-advice. Fees of 0.25% undercut a typical 1%^S77. Yet robo assets are $634bn to $754bn on a narrow count^S76, or over $1.2tn on a wider one^S75, against $176.8tn at SEC-registered advisers, whose staff grew 7.5% in 2025^S78. BLS counts 299,400 personal advisers^S79. Goldman and JPMorgan left^S75.

The limit: those tools did one task. AI works across many.

Which parts of advisory proved resistant?

Three have held. First, accountability: Deloitte agreed a partial refund on an A$440,000 Australian government report with fabricated references, though it did not say AI caused them^{S60, S61}. Someone must stand behind the number. Second, judgement on messy evidence, as the Harvard trial showed. Third, people work. In Bain's survey of 303 practitioners, 45% use AI in M&A, led by sourcing, screening and diligence^{S62, S63}, and Bain says aligning stakeholders and making decisions still take time^S62.

Evercore grew net revenue 38% in nine months and its senior banker count 32% in two years, per an analyst on its call^{S64, S65}. That is a deal cycle as much as resistance. Houlihan Lokey, closer to the mid-market, fell 4.9% over six months to September, though it rose 4.8% in the latest quarter (our calculations)^S66.

Does the evidence support "kill"?

Not on evidence to October 2026: the large firms still grew, while price and junior hiring in analysis-heavy work are under real pressure.

Other causes matter too. Australian Big Four firms blame over-hiring and weak demand^S34, PwC UK blames the economy^S32, and Accenture cites the Middle East war^S5. Forecasts are not results: BLS projects management analyst jobs up 8.8% to 2034^S28, or 10% to 2035 in its newer handbook^S80, and MCA firms forecast 6% growth in 2026^S20.

Our assessment. That AI-native firms have killed strategy consulting: no, with high confidence. That analysis-heavy work keeps losing price and junior seats over three to five years: likely, with medium confidence. That judgement stays protected after that: unknown, with low confidence.

What does this mean for an independent firm?

Expect the price of analysis to fall, and big buyers to get the cut first, as with brokers. Separate what you charge for analysis from what you charge for judgement. Boutiques hold one advantage: success fees already pay for outcomes, the model McKinsey is moving towards.

The exposed end is small deals. One report says that bank's model makes even a $5m sale profitable^S54. As with tax software, its first customers may be owners who never had an adviser, but brokers at that size overlap directly.

Check everything. The Harvard trial shows AI makes wrong answers arrive faster, and the Deloitte refund shows who pays. Protect training too: Bain's UK head says juniors must work with experts to build judgement^S15. Watch three signals: pricing language on Accenture's calls, the share of MCA members shrinking, and independent benchmark reruns.