AI's impact on the global strategy and corporate-strategy industry, covering management consultancies, the advisory arms of the Big Four and in-house corporate-strategy functions, as at October 2026.
The usual worry about artificial intelligence in strategy consulting is that it is not good enough. Strategy is judgement, the argument runs, and a machine that invents citations cannot be trusted with a board's biggest decision.
The controlled evidence points the other way. In a field experiment run by Harvard Business School researchers with 758 Boston Consulting Group consultants, those given AI completed 12.2 per cent more tasks, 25.1 per cent faster and at more than 40 per cent higher quality, on work within the tool's capability. McKinsey says 72 per cent of the firm is active on its in-house assistant, Lilli, which answers more than 500,000 prompts a month.
On 1 October 2026 Accenture, the largest listed professional-services firm, reported fourth-quarter new bookings of $22.2 billion, up 4 per cent in US dollars. The same earnings release lists among its risks that AI could harm the company's business, including by reducing demand for its services.
Both are true at once because of how the industry is paid. Most strategy work is sold by the hour and delivered by a pyramid of junior analysts doing exactly the research, modelling and slide production the tools now compress. An analysis that took sixteen hours and now takes four is worth the same to the client and, on an hourly invoice, earns three-quarters less. The juniors whose hours disappear are also the apprentices who become the partners. The threat to strategy consulting is not that AI cannot do the work. It is that it can.
Six risks, ranked by how badly they bite
Worst first; the evidence follows.
1. The billable hour. Hourly billing turns every efficiency gain into a smaller invoice. An analysis that drops from sixteen hours to four earns a quarter of the fee unless the firm changes how it charges, and clients who know the work is machine-assisted will ask why it has not.
2. The missing apprentices. The work AI absorbs first is the searching, synthesis and first drafting that juniors used to do, and it is where they learned judgement. A firm can keep its partners for a decade and still find nobody trained to replace them.
3. Evidence nobody checked. Strategy firms sell evidenced analysis. Published reports from Deloitte, KPMG and EY Canada were found in 2025 and 2026 to contain incorrect or fabricated citations. Each was a failure of human review, and each struck the one asset the firm trades on.
4. Pilots that never pay. MIT's NANDA initiative found that about 95 per cent of organisations were getting no measurable return from generative AI, and Thomson Reuters found only 18 per cent of professionals say their organisations track the return on AI at all.
5. Client data walking out. Strategy work runs on the most sensitive documents a company holds. Zscaler counted 410 million data-loss-prevention policy violations tied to ChatGPT alone in 2025, and organisations are deploying AI agents faster than they feel ready to secure them.
6. Regulation, lower than it looks. Most strategy work sits outside the heaviest tier of the European Union's AI Act. The exposure arrives when a deliverable crosses into a listed high-risk area such as employment, where the stricter rules for stand-alone systems apply from 2 December 2027.
What AI already does in strategy consulting
Strategy consulting does not run on bespoke AI. It runs on assistants built over a handful of general-purpose model families. By Menlo Ventures' count, Anthropic, OpenAI and Google together accounted for 88 per cent of enterprise use of large-language-model APIs in 2025, and their order inverted in two years: Anthropic rose from 12 to 40 per cent of enterprise spending, OpenAI fell from 50 to 27 per cent, and Google climbed from 7 to 21.
The large firms have tied themselves to those suppliers. Anthropic and Boston Consulting Group announced a partnership in September 2023 to bring Claude to BCG's clients; Bain announced a global services alliance with OpenAI in February 2023, with Coca-Cola as its first client. Across professional services, 40 per cent of professionals told Thomson Reuters in 2026 that their organisations use generative AI, up from 22 per cent a year earlier, while 15 per cent reported agentic AI, the kind that carries out multi-step tasks rather than answering a single prompt.
The same field experiment that showed the gains also showed their edge. On a task deliberately chosen to sit outside the tool's capability, consultants using AI were 19 percentage points less likely to reach the correct answer than those without it. The researchers called this the jagged frontier: tasks that look equally hard fall on different sides of it, and nothing tells the user which side they are on.
| Measure | Effect of AI access |
|---|---|
| Tasks completed, within AI's capability | 12.2% more |
| Speed, within AI's capability | 25.1% faster |
| Quality of output, within AI's capability | More than 40% higher |
| Correct answers on a task chosen to sit outside AI's capability | 19 percentage points less likely |
Note. Inside the frontier the gains are large. Outside it, the same tool made skilled consultants worse, and the frontier is not marked on any map.
Buy the full report
AI Impact - Industry
AI's impact on Strategy & Corporate Strategy
Published: 4 October 2026
89 pages138 checked sources
- Industry
- Strategy & Corporate Strategy
This published copyUSD 29.99
Buy this reportConfigure this report new at today’s date (USD 99)Sixteen hours, invoiced as four: the billable-hour problem
Most professions can bank a productivity gain. Strategy consulting, priced by the hour, often cannot. The analysis that took sixteen hours and now takes four is worth the same to the client and earns the firm a quarter of the fee. Under the old commercial model the efficiency arrives as a smaller bill rather than a larger margin.
The largest firm in the sector now states the risk in its own words. Accenture's earnings release for the fourth quarter of 2026 warns that AI could harm its business, including by reducing demand for its services, in the same document that reports $22.2 billion of new bookings. Selling the transformation while being exposed to it is the position of the whole industry.
The way out is commercial, not technical. A firm that prices by outcome turns the four-hour analysis into margin; a firm that prices by time hands the saving to the client. A client who can see the hours shrinking will ask why the invoice has not.
The consulting pyramid loses its base
The consulting pyramid rests on a wide base of junior analysts who gather data, build models and assemble slides, funding a narrow tier of partners. That base is the work generative AI handles best. McKinsey reports that colleagues using Lilli save up to 30 per cent of the time they spend searching for and synthesising knowledge, which is the core of a first-year analyst's week.
That is a gain for the firm and a quiet loss for the analyst. The research, synthesis and first drafts that a tool now produces are the tasks through which juniors learned to judge whether an answer was right. A firm that automates them keeps the output and loses the training.
The jagged frontier makes the loss sharper. A senior consultant who learned the work by doing it can sense when a model has crossed the line into confident error. A consultant who has only ever edited the model's output has no such reference, and the experiment shows that crossing the line makes skilled people measurably worse.
Flight 447 and the deskilled partner
On 1 June 2009 Air France Flight 447, an Airbus A330 flying from Rio de Janeiro to Paris, was lost over the Atlantic with 228 people aboard. According to the French air accident investigator, the BEA, ice obstructed the speed probes, the speed readings became erroneous and the automatic systems disconnected. The crew did not identify the approach to the stall, failed to diagnose the stall itself, and made none of the inputs that would have made recovery possible.
Strategy firms are building the same arrangement on purpose. A junior analyst who no longer builds the bottom-up market model or structures the problem by hand is not acquiring the craft that later becomes senior judgement, and a deskilled junior becomes a deskilled partner.
The cost is invisible while the system works. It arrives on the day a model changes after a silent vendor update, or a confident recommendation rests on an assumption nobody audited, and there is no bench left that can rebuild the reasoning from first principles.
| Failure mode | On Air France 447, 2009 | In a strategy function now |
|---|---|---|
| Loss of reliable inputs | Iced probes made the speed readings erroneous and the automatic systems disconnected | A model changes, or is fed bad data, and its output stops being reliable without warning |
| Failure to recognise | The crew did not identify the approach to stall | Users cannot tell a grounded finding from an extrapolation or a fabrication |
| Failure to diagnose | The crew failed to diagnose the stall situation | Fluent, confident output is accepted without being tested against the evidence |
| No recovery | No inputs were made that would have made recovery possible | Nobody left who can rebuild the reasoning by hand when the model breaks |
Note. Each failure mode pairs what happened in the cockpit, which is history, with the same failure in a strategy team, which is a forecast.
AI hallucination: when the evidence is wrong
The most damaging failures so far have come from the industry's own published work. Deloitte carried out an independent assurance review of Australia's Targeted Compliance Framework, the system that applies penalties to job-seekers, for the federal Department of Employment and Workplace Relations. The department states that Deloitte confirmed some footnotes and references in the review were incorrect, that a further investigation found generative-AI tools had produced the errors, and that a corrected report replaced the original.
GPTZero, a company that builds tools to detect AI-generated text and citations, has published its own investigations of two more. In KPMG's October 2025 report Total Experience: Redefining Excellence in the Age of Agentic AI, it found only 5 of 45 citations accurately pointed to real sources. In EY Canada's cyber-security report on loyalty-programme fraud, Points of Attack, it found more than half of the titles among 27 references did not correspond to real sources. GPTZero calls the pattern 'vibe citing'.
In each case the model behaved as such models do. The failure was a review process that let fluent output reach readers unchecked, and it struck the one product a strategy firm sells, which is confidence that its evidence is real. Each was caught by an outsider, so the public cases are a floor on the problem, not a ceiling.
| Firm | The published work | What was found, and by whom |
|---|---|---|
| Deloitte | Independent assurance review of Australia's Targeted Compliance Framework, 2025 | The Department of Employment and Workplace Relations states that Deloitte confirmed some footnotes and references were incorrect and that generative-AI tools had produced the errors; a corrected report replaced the original |
| KPMG | Total Experience: Redefining Excellence in the Age of Agentic AI, October 2025 | GPTZero found only 5 of its 45 citations accurately pointed to real sources |
| EY Canada | Points of Attack, a cyber-security report on loyalty-programme fraud | GPTZero found more than half of the titles among its 27 references did not correspond to real sources |
Note. None of these was a model malfunction. Each was a review process that let fluent output reach a reader unchecked.
Ninety-five per cent of organisations, no return
Adoption is close to universal and return is rare. McKinsey's 2025 survey found 88 per cent of respondents' organisations use AI regularly in at least one business function. Yet MIT's NANDA initiative, drawing on 150 interviews, about 350 surveys and an analysis of 300 deployments, found that despite $30 to $40 billion of enterprise investment, about 95 per cent of organisations were getting no measurable return from generative AI.
Others point the same way. RAND, the American research institute, reports estimates that more than 80 per cent of AI projects fail, twice the rate of IT projects that do not involve AI. Gartner expects more than 40 per cent of agentic-AI projects to be cancelled by the end of 2027, citing escalating costs, unclear business value or inadequate risk controls. Only 6 per cent of 1,500 chief financial officers surveyed by IBM and Oxford Economics describe their finance function as transformation-ready, with AI embedded at scale.
The causes MIT identified are organisational: unsuccessful pilots tended to use generic tools and to target sales and marketing. This is the gap strategy firms are paid to close for their clients, and it runs through their own buildings. The firms that close it inside their own delivery will have something to sell.
| Measure | Share | Population surveyed |
|---|---|---|
| Organisations using AI regularly in at least one business function | 88% | Global survey respondents, McKinsey, 2025 |
| Professionals whose organisations use generative AI | 40% (22% a year earlier) | Professionals, Thomson Reuters, 2026 |
| Organisations scaling an agentic AI system somewhere | 23% | Global survey respondents, McKinsey, 2025 |
| Professionals whose organisations track the return on AI | 18% | Professionals, Thomson Reuters, 2026 |
| Finance functions transformation-ready, with AI embedded at scale | 6% | 1,500 chief financial officers, IBM and Oxford Economics, 2026 |
| Organisations getting a measurable return from generative AI | About 5% | MIT NANDA, 2025: 150 interviews, about 350 surveys, 300 deployments |
Note. Read down the column: nearly everyone uses the tools, fewer than one in five tracks the return, and about one organisation in twenty gets one it can measure.
AI data leakage: the data that walks out
Strategy teams feed AI the most sensitive material a company holds: a target's data room, unpublished results, board papers. Zscaler's 2026 AI security report, drawn from its customers' traffic in 2025, counted 410 million data-loss-prevention policy violations tied to ChatGPT alone, including attempts to share source code and personal records, and found enterprise data transfers to AI applications up 93 per cent in a year.
Agents widen the gap. Prompt injection, an instruction hidden inside a document or web page that an AI system reads and then obeys, ranks first on the OWASP list of risks for large-language-model applications. A planted page in a data room could tell a research agent to send out what it reads. Cisco's 2026 survey found 83 per cent of organisations planned to deploy agentic AI, but only 29 per cent felt ready to do so securely.
Note. Intent runs far ahead of the means to secure it. An agent that can read a data room and send email is a door as well as a tool.
What it comes down to
The fair rating for strategy consulting is high disruption, at the edge of transformative. The industry's core task, turning information into a defensible recommendation, is the task generative AI compresses fastest. What holds it short of transformation is value capture and accountability: most organisations still cannot show a measured return, and responsibility for a strategic decision stays with a person.
The names at the top are likely to persist. Trust, client relationships and the capital to buy capability are moats AI reinforces, which is why incumbents are buying the threat: Accenture completed its acquisition of the UK AI firm Faculty on 16 March 2026 and says it expects to deploy about another $5 billion in acquisitions in its 2027 financial year. What will not persist is the arrangement underneath, the hourly invoice and the pyramid of juniors it paid for.
A firm that prices by outcome turns the four-hour analysis from a lost fee into a margin. A firm that keeps the hour and cuts the juniors gets the worst of both: smaller invoices now, and in ten years nobody trained to check the machine. If AI took over the first five years of every career in your own organisation, who would be left to know when it is wrong?
Drawn from TheRiskAgent's report AI Impact on the Strategy & Corporate Strategy Industry (October 2026); every figure here was checked against its original publisher. Produced with AI research tools and reviewed before release; reference material, not advice. The full report carries the detailed analysis.