Deloitte UK: workers spent nearly £1bn of their own money on GenAI

You've probably used a free chatbot on a work task because the official tool was slower, locked or missing. British workers spent nearly a billion pounds of their own money so their employer could pretend it had a GenAI programme. That is not transformation. That is a tax on the people who still care enough to pay.

Hayley McKelvey, chief AI officer at Deloitte UK, oversaw the largest workplace GenAI survey I have seen in a single country: 25,000 UK workers, aged 18 to 70, with Ipsos UK in the field from 7 May to 10 June. They will repeat it every six months. Treat this as a baseline, not a mood.

Sixty-three per cent of UK working adults in that age band say they knowingly use GenAI for work. Of the users, 46% are on free tools, 34% on external tools the employer pays for and 17% on in-house tools. One in six users (17%) pay for at least one tool themselves. Deloitte's estimate of that personal spend across UK workers is £958 million a year. That is the number in the release. The headline says nearly £1 billion. Same object.

Permission is not the bottleneck. Cover is.

Thirty-one per cent of GenAI users say they use it without their employer's knowledge. Twenty-four per cent of UK workers now use it every day. Sixty-five per cent of users report a lack of convincing leadership on how it should be used. McKelvey's line is that workers do not want to wait for permission. The challenge is no longer getting people to use GenAI. It is meeting demand in a way that is secure, responsible and actually worth something.

Claimed time saved is 70 minutes a week on average. Most of that time goes back into more work for the same employer. Twenty-three per cent think there is a stigma. Sixty-four per cent of weekly users worry their manager will decide the tool can do their job. So of course they hide it. You built a surveillance story and then acted surprised about shadow AI.

Around half of the people using GenAI at work say they have had no formal training on using it safely and effectively. The top three jobs for the tool are searching for information (43%), drafting emails (43%) and creating summaries (31%). That is not transformation. That is a slightly faster Tuesday.

Paul Lee, partner and head of industry insight at Deloitte UK, said the story is not that workers are using GenAI. It is that they are using it despite limited training, patchy guidance and, in some cases, without the employer knowing. Advantage, on his telling, goes to the organisations that move past access and put guardrails and purpose around the work.

If you are a United Kingdom CHRO or CAIO, the practice is ugly and cheap: stop celebrating adoption dashboards. You already have adoption. You do not have cover, training or a manager who can tell the difference between a summary and a decision. Train on the workflow they are already hiding, not on a module titled "Introduction to Copilot" that the shadow users will skip because they are already paying ChatGPT Plus out of their own pocket.

The mix of tools is the other tell. Free, employer-paid, in-house and bring-your-own can all be true in one team. If your policy only names the in-house bot, you have written a policy for 17% of users. The 46% on free tools and the 31% in the shadows will keep going. McKelvey is right that permission is no longer the product. Cover is. Cover means a blessed path that is faster than the personal card.

Seventy minutes a week, dumped back into more work for the same employer, is why the stigma line exists. People can feel the tool making them look optional. Sixty-four per cent of weekly users already worry the manager will draw that conclusion. If you then publish a poster about higher-value work while cutting contractors, you will train them to lie.

What this means for L&D

Lee's search, email and summary list is the floor. It is also an indictment. If that is still the top three after a year of licences, the programme has not reached a process. It has reached a toolbar. Repeat the survey in six months. If the top three have not moved, you did not transform. You subscribed.

Put £958 million and half-of-users-untrained on the same slide. Name McKelvey and Lee. Repeat the survey. Do not call search-and-email a transformation programme.

Sources: Deloitte UK press room, 16 September 2026.