The Office for National Statistics has a measurement problem. It can count more businesses using AI, but it says that does not yet tell us what the technology costs them or what benefit it delivers. Its September paper is about the whole UK economy. The same gap can appear inside one business: everyone knows a tool has been bought, but nobody can say which job became easier.
The ONS's earlier survey found that around 35% of UK businesses with at least 10 employees reported using an AI technology in June 2026, up from around 12% in late 2023. The average number of AI technologies used by adopting businesses moved only from about 1.4 to 1.6. Those figures describe use, not hours saved or better decisions. They also do not tell us whether any particular business should buy another tool.
Start with one piece of work
Take a job somebody can describe without saying “AI”: preparing tomorrow's sales figures, answering a routine account query, or finding open quotes that need a response. Watch it happen a few times. Record how long it takes, where information has to be checked, and how often the person has to go back because something is missing or wrong.
Then agree what a good result looks like. A report should be current and traceable to its source. A reply should use the right account facts. A quote reminder should go to a customer who has not already answered. If the proposed system makes the first step faster but leaves somebody cleaning up mistakes, that work belongs in the measurement too.
Count the awkward minutes
For two weeks after a change, compare like with like. Count preparation time, review time and exception handling. Note the days the system failed, needed a manual restart or found a case it could not safely complete. Ask the person doing the work whether they would trust the output tomorrow without rebuilding it themselves.
That last question matters. A fast report that someone checks from scratch has shifted the work, not removed it. Equally, a person spending five minutes reviewing a clear exception may be a better result than a system that silently sends the wrong answer.
You do not need a national statistics project to find out whether one workflow improved. You need a baseline, a result people can check, and the honesty to count the work that remains. If the tool has not changed the day, change the process or stop paying for the tool. A licence count is not the outcome.
Source and scope
The ONS's Measuring artificial intelligence in the UK economy using a thematic account, revised 21 September 2026, says measuring AI adoption does not yet produce a value-based estimate of costs and benefits; this methodology page is not itself official statistics. The 35%, 12% and 1.4-to-1.6 figures come from the ONS's Artificial intelligence in UK businesses: 2023 to 2026, released 20 July 2026, based on its voluntary Business Insights and Conditions Survey. They are UK survey estimates for the stated business-size group, not a forecast for WeekFound clients.