Adoption is not implementation

Progress on AI in British business is usually reported through adoption statistics. On their own terms, these show real progress. The Office for National Statistics reports that the share of businesses with ten or more employees using AI rose from around 12% in late 2023 to around 35% by June 2026.

The detail beneath that headline is less encouraging. Among businesses with ten or more employees that use AI, 10% describe their use as extensive, and 11% say more than half of their workforce has been trained to use it. The technology most often in use is the large language model, reported by 18% of firms. These measures of depth are new, and the ONS asks that they be read as early indications.

What the headline measure does not capture

The ONS is candid about the limits of its own figure. It notes that the measure "treats all reported uses as equivalent and does not distinguish between light use, user-level use, and more embedded, production-level adoption", and says that limited depth would imply "relatively limited transformative impacts to date for most AI-adopting firms".

International evidence points the same way. Between November 2025 and January 2026, researchers working with the Bank of England, the Federal Reserve Bank of Atlanta, the Bundesbank and Macquarie University surveyed nearly 6,000 senior executives across the United States, the United Kingdom, Germany and Australia. They found 69% of firms actively using AI, and 89% reporting no effect on productivity over the previous three years.

Where firms have been asked what they changed as a result of adopting AI, a large share report little. A survey of 3,235 senior leaders in 24 countries, published in January 2026, found 37% using AI at a surface level, with little or no change to existing processes. Neither survey is representative of all UK firms, and both rely on what respondents report.

Individuals are ahead of their employers

A second gap sits beneath the first. Between May and June 2026, 55% of people in work in Great Britain reported using AI for work or education, against 35% of UK businesses with ten or more employees reporting any use of AI. Most of that use is individual, unseen by the business and unconnected to anything the business measures.

It also introduces a risk. Where staff use personal AI tools of their own choosing, the business has no visibility of the commercial, client or personal data being entered into them.

What implementation means

Much of what is counted as AI adoption in smaller businesses is convenient and easy to demonstrate: a voice assistant on a mobile phone, automatically generated email replies, a browser extension. These applications rarely have any bearing on the financial performance of the business.

Implementation is different. It means integrating AI into the processes through which the business operates: how quotations are produced, how work is scheduled, how a customer enquiry is resolved, and how work passes from one person to the next. Among the small group of survey respondents whose organisations attribute 5% or more of operating profit to AI, nearly three-quarters had redesigned workflows around it, against about one in four of all others.

AI is delivering results at the level of the individual task, and those results are not reaching the financial statements. The technology is effective. The shortfall is in how it is implemented.

The Implementation Gap sets out this evidence in full, together with the five disciplines we apply to close the gap.