UK workers pay for AI tools at work

That matters economically because it shows AI is beginning to raise productivity from the bottom up, not just through enterprise software rollouts and capital spending. Workers told Deloitte they most often use AI to search for information, draft emails and create summaries, saving an average 70 minutes a week. Even if those tasks are relatively basic, the aggregate effect across millions of employees could be material for labour productivity, operating efficiency and business formation in services-heavy economies such as the UK.

The survey, based on 25,000 workers polled by Ipsos for Deloitte, suggests employers are not yet capturing the full value of the technology they are helping normalize. Two-thirds of workers said they had tried tools including ChatGPT, Claude, Google Gemini and Microsoft Copilot, and nearly a quarter were using them daily. Deloitte’s chief AI officer Hayley McKelvey said workers were “not want[ing] to wait for permission,” while partner Paul Lee warned the biggest gains would come to firms that move beyond simply giving access and instead build guardrails and purpose around use.

For investors, the findings help explain why the AI build-out is becoming a competition not just among software vendors, but across the broader productivity stack. Microsoft, whose Copilot is among the tools cited, stands to benefit if enterprises convert informal use into paid subscriptions and wider deployment. Nvidia remains exposed to the same adoption cycle through demand for the compute underpinning AI services, while listed AI names such as C3.ai reflect a market where enthusiasm for future monetisation has outpaced clarity on enterprise uptake. The data also point to a mixed picture for corporate margins: businesses that formalize AI use could lower administrative costs, while those that ignore shadow usage risk leakage, compliance problems and missed productivity gains.

The tension for management teams is straightforward. Letting employees experiment can accelerate adoption and reveal useful use cases. But unmanaged use raises data-security, privacy and governance risks, especially if staff are entering sensitive information into public models. The companies that win will likely be those that convert this ad hoc spending into approved platforms, training and controls rather than treating it as a fringe behaviour.

The broader implication is that AI adoption is no longer waiting for boardroom strategy to catch up with frontline behaviour. If workers are already paying for the tools themselves, the next phase for investors will be watching which employers turn that organic demand into measurable productivity gains — and which are left paying for the risks without capturing the upside.