Contents7
In short
The major surveys on AI in business come from firms that sell AI consulting, and they question groups whose revenue often exceeds a billion dollars.
They collect executives’ statements, not accounts. Their authors say so themselves, in the small print.
None of the ones we have read describes a team of 20 to 500 people.
Official statistics do count small companies, and show an adoption gap that widens with size, for lack of the fixed resources needed to get started.
What a 50-person company can keep from the consulting firms: a direction of travel. Not figures to copy.
Who these studies actually survey
A consulting firm’s survey almost always opens with a striking figure and ends with a methodology note that few people read. Yet that is where everything is decided.
of the 1,330 executives in the 2026 survey by BCG, which sells AI consulting, work in groups with more than 5 billion dollars of revenue (BCG, 2026)
at least, and over a billion for three quarters of them: the threshold for the 2,110 executives in the Global AI Pulse by KPMG, which sells audit, consulting and AI services (KPMG France)
of the 3,235 organisations in Deloitte’s State of AI in the Enterprise 2026 already use AI every day; Deloitte sells consulting, audit and AI services (Deloitte)
Even the studies that present themselves as close to mid-sized companies remain far off. The “middle market” of RSM, which sells audit, consulting and managed AI services, starts at 30 million dollars of revenue and goes up to 10 billion. The closest sample we found is Grant Thornton’s International Business Report, Grant Thornton also selling AI consulting: companies with 50 to 3,000 employees.
From the field
What our research register shows
We keep a register of figures published on business AI. Closed on 14 October 2026, it lists 489 figures, of which 278 were read at the source and are ready to quote. No consulting firm study in that register describes teams of 20 to 500 people. This finding changed the way we write: whenever we quote a firm, the size of its sample is in the sentence.
What they measure: statements
Most of these surveys ask executives what they think, plan or observe. They do not read the accounts. BCG acknowledges a perception bias in its 2026 report, and specifies that only 33 of its 41 capabilities are comparable from one year to the next. When it announces that the share of large companies getting measurable value from AI rose from 26% in 2024 to 48.5% in 2026, that is a declared trend, on a scale that has changed.
The difference shows when the same firm changes method. The BCG Institute, the group’s research arm, read the accounts of more than 600 US listed companies with a market capitalisation above 5 billion dollars, without a survey. It finds only 6% AI leaders. Two methods, two orders of magnitude.
Academic research with no commercial interest, published by the NBER in 2026 and based on nearly 6,000 executives in the United States, the United Kingdom, Germany and Australia, gives a third picture: 69% of firms actively use AI, but nine executives out of ten see no effect on employment or productivity in their company over the past three years.
Who publishes them: firms that sell what they recommend
This is not an accusation, it is a business model. According to BCG itself, more than 40% of its 2025 revenue comes from services centred on AI and technology. A study concluding that AI value comes from a multi-year transformation programme, led by top management, also describes the product the firm sells.
That does not invalidate the conclusions. It means reading them the way you read a manufacturer’s study of its own products: useful, well documented, and never neutral. We apply the same rule to our own articles, since we also sell an audit.
A big consulting firm’s study
Groups worth several hundred million to several billion
Executives’ statements
A sponsor that sells AI consulting
Trends and gaps between “leaders” and “laggards”
Official statistics
All companies with 10 or more people, by size class
Use declared by the company, on a fixed definition
An institute with nothing to sell
Adoption rates, purposes and barriers
What official statistics say instead
INSEE in France and Eurostat for the EU survey tens of thousands of companies, small ones included, with the same definition from one year to the next. What they show, no consulting firm puts forward: the gap between small and large companies is widening.
Gap in AI use between the smallest and largest companies, in points
France (fewer than 50 employees vs 250 or more)
European Union (10 to 49 people vs 250 and more)
INSEE attributes it to the “fixed costs of adoption”: choosing a tool, training people, securing data and adapting a process costs roughly the same at 40 people as at 400. Eurostat offers the same reading: implementation complexity, economies of scale and costs.
Barriers also differ with size. Among EU companies that considered AI without adopting it, a lack of expertise is cited by 70.9% of those with 10 to 49 people, and a lack of usefulness by 19.2% of them, against 9.8% of those with 250 and more. A small company doubts more that AI will serve it, and lacks someone to check.
Finally, the way companies equip themselves has nothing in common. In France, 80% of companies using AI acquire it by buying ready-made software, according to INSEE. In the EU, 25.8% of small companies using AI go through an external provider, against 45.8% of large ones. The multi-year transformation programmes the consulting firms describe look nothing like the daily life of an SME switching on an AI feature in software it already pays for.
What a 50-person company can keep
Throwing these studies away would be a mistake. Their core conclusions match what we see:
- Value follows measurement. According to BCG, only 12% of the large companies surveyed track AI value directly in their profit and loss account. The link it draws between measurement and value is declared, not proven, but the lesson holds: without a starting measurement, you will never know whether AI paid off.
- The effort goes into people and processes. The companies BCG rates as the most advanced devote nearly three quarters of their AI priorities to people, organisation and processes, and only 9% to algorithms.
- Foundations matter. According to Bpifrance, the French public investment bank that also supports these companies, an SME already committed to digitalisation is 5 times more likely to use AI. A statistical link, not proof of cause, but consistent with the rest.
What to leave behind: the gain percentages, payback periods and “leader” rankings. They describe groups with a data department, multi-year budgets and dedicated teams. Copying them into an SME’s plan means setting yourself someone else’s target.
Key takeaway
A consulting firm’s study shows which direction to look in. To know where a 50-person company stands, you need official statistics for the context, and your own measurement for the rest.
Our view
The lack of data on mid-sized companies is real, and it is no one’s fault in particular: the consulting firms mostly survey large groups, and official statistics only publish broad size classes. The European Commission’s 2026 report on France even recommends publishing AI adoption by company size.
In the meantime, the right approach stays modest: start from a task, measure what it costs today, and judge AI on that task. That is the spirit of our article on AI return on investment and of our guide to integrating AI into a business. And it is what the AI Audit Sprint does: an assessment of your tasks, costed on your data rather than on a 5-billion-dollar group’s. And if AI is already running in your company and the bill is rising without you knowing what pays off, our AI at the right cost offer measures the cost of each use.
Frequently asked questions
Are BCG, Deloitte or KPMG studies reliable?
They are well documented and useful on large groups. Their limits lie in the sample (companies with several hundred million dollars of revenue and more), the method (executives’ statements) and the sponsor, which sells the consulting it recommends. They should be quoted with those three pieces of information.
Are there serious studies on AI in SMEs?
Yes, mostly public ones: Eurostat and national statistics offices such as INSEE (companies with 10 or more people), the French France Num barometer (fewer than 250 employees), Bpifrance (French SMEs and mid-caps with more than 10 employees) and the ECB survey. Each counts a different kind of use: do not compare them with one another.
Why do small companies adopt AI less?
According to INSEE, because of the fixed costs of adoption, which weigh more when headcount is small: choosing a tool, training people and securing data cost roughly the same at 40 people as at 400. Eurostat adds the lack of expertise, cited by more than seven small companies in ten among those that considered AI without adopting it.
Should consulting firms’ figures be ignored?
No. Keep their qualitative conclusions (measure, train, rework processes) and leave their gain percentages, which describe companies other than yours. To set yourself a target, start from what the task you want to hand to AI costs today.
Sources
- BCG, 2026 (new tab)web-assets.bcg.com
- KPMG France (new tab)kpmg.com
- Deloitte (new tab)deloitte.com
- NBER (new tab)nber.org
- BCG itself (new tab)bcg.com
- Eurostat (new tab)ec.europa.eu
- 2026 report on France (new tab)ec.europa.eu
Jim · founder of Brain Plus AI
Also in business ai
Stepping back11 min
AI return on investment: why it is taking so long
The figure of 95% of AI projects with no return went round the world. What it really measures, why the other studies say roughly the same thing, and the practical lessons to draw from them for an SME.
Practical guide11 min The topic guide
Integrating AI in business: private and at the right cost
For teams of 20 to 500 people, integrating AI comes down to answering five questions in order: which task, where the data goes, which model, at what price, under what rules. This guide takes them one by one and points to our detailed articles.
Stepping back10 min
The human role alongside AI: knowing how to judge
AI excels at some tasks and is confidently wrong on neighbouring ones. The human's value shifts: knowing where the line runs, keeping your judgement, answering for the result.