One of the few firm-level studies to attempt a causal estimate, a working paper published by the Bank for International Settlements in January 2026 using data on more than 12,000 firms in the European Union and the United States, found that adopting AI raised labour productivity, measured as turnover per employee, by 4% on average among EU firms, with no adverse effect on employment in the short term.
The gain was larger where firms also invested in software, data or workforce training, and it was concentrated in firms with 50 or more employees: the study found no significant effect below that size. That is consistent with the wider evidence: the return follows from implementation.
UK evidence on smaller firms is thinner. A University of St Andrews working paper, drawing on the government's Longitudinal Small Business Survey, estimated productivity 27% to 133% higher among small and medium businesses that had adopted AI. It has not been peer reviewed, its authors describe the wider evidence as sparse and mixed, and we do not rely on it in the figures below.
The figure in pounds
Official statistics allow the 4% figure to be expressed in pounds. UK businesses with 10 to 249 employees generate about £214,000 of turnover per person employed. A business of 50 people at that average turns over about £10.7 million. A 4% gain is worth about £8,600 per person, or about £428,000 a year.
| Size of business | Annual value of a 4% gain |
|---|---|
| 50 people | £428,000 |
| 100 people | £856,000 |
| 200 people | £1.71 million |
At the 4% average, each month in which a business of 50 people has adopted AI without implementing it forgoes about £36,000 of turnover.
Two businesses, one difference
What separates a business that realises this gain from one that does not? Consider two businesses of the same size in the same market. Each employs 50 people and has paid for the same AI tools. The first has added the tools alongside the way it already works. The second has rebuilt three of its core processes around them.
| Business A: adopted AI | Business B: implemented AI | |
|---|---|---|
| Where AI sits | Alongside existing processes | Built into three core processes |
| Ownership | No named owner | A named owner for each process |
| Training | None organised | More than half of the team trained |
| Measurement | Number of licences in use | A baseline per process, reviewed monthly |
| After a year | No measurable change | A measurable gain against the baseline |
Method and limits. This is an illustration and not a forecast. Turnover is not profit, and turnover per person varies widely by sector; the official figure counts working owners and excludes financial services. The 4% is an average among EU firms between 2019 and 2024, in a study that counts big data analytics as AI and found no significant effect below 50 employees. Most firms surveyed report no productivity effect to date. Neither business in the comparison is a client. Substitute your own turnover and headcount to obtain a figure for your own business.
Sources
- Aldasoro, Gambacorta, Pal, Revoltella, Weiss and Wolski, "AI adoption, productivity and employment: evidence from European firms", BIS Working Paper 1325, January 2026.
- Department for Business and Trade, Business population estimates for the UK and regions 2025, October 2025.
- Brown, Sun, Cowling and Dowell, "Can We See it in the Numbers Yet? Artificial Intelligence, Firm-Level Productivity and SMEs", SSRN working paper, April 2025.
Full references are given in the report.