When employees buy their own work tools, they may be signalling that a recurring need is going unmet. The response should begin with understanding the work, while keeping clear rules about company information and approved systems.

What happened

In a release dated 16 September 2026, Deloitte estimated that British workers were spending £958 million a year of their own money on generative AI for work. Its survey covered 25,000 UK working adults aged 18-70, with fieldwork in May and June. Among respondents using generative AI, 17% paid personally and 31% used it without their employer’s knowledge. Those percentages are not shares of every worker. Deloitte provides the findings and methodology.

The expenditure is a survey-based estimate. It is not an audited account of purchases or proof that the tools generated a corresponding productivity benefit.

Why it matters

For an employer, the useful diagnostic question is which tasks people are trying to make easier. The answer may reveal a missing capability, an approval process that moves too slowly or a tool that employees have not been taught to use effectively.

I would investigate through a clearly explained discovery exercise. Ask teams about their recurring tasks, the assistance they seek and the barriers they encounter. Make it clear how the information will be used, and involve the people responsible for data and security requirements.

The exercise should not encourage staff to ignore existing rules. It should give them a practical route to describe unmet needs and request a suitable option. A policy is easier to follow when employees know where to take a reasonable request and when they can expect an answer.

Some needs may be met through better training or a change to the process. Others may justify an approved tool. Buying a licence for every request without checking the task would be as superficial as assuming every personal purchase is unnecessary.

The bigger shift

My reading is that AI adoption needs a service model inside the organisation. Employees need more than a list of prohibited products; they need a usable way to obtain support for legitimate work.

That model could include a small set of approved options, clear instructions about permitted information and a route for evaluating new use cases. It should explain who owns the decision and how exceptions are assessed.

The organisation should also measure whether its response works. Are employees able to complete the intended task? Does review create excessive additional work? Are there requests repeatedly waiting for an answer? Those observations can help distinguish a technology gap from an operating gap.

Spending should remain separate from value. A popular tool may save effort, improve quality, provide reassurance or simply feel easier to use. A business case needs to establish which benefit exists and whether it survives the complete workflow.

This is why AI strategy should begin with practical business needs, rather than with the assumption that more subscriptions equal progress.

My take

I would treat personal workplace AI spending as a reason to listen carefully, then make a clear decision.

Choose a few frequently requested tasks and test approved ways to support them. Publish the permitted use, the review requirements and the route for feedback in language employees can act on.

Then revisit the demand. If people still struggle to get useful help through the approved route, the organisation has more work to do. The goal is a process that combines usable assistance with accountable handling of company work, so staff do not have to guess how to proceed.