Share with your CFO
A Deloitte survey of 25,000 U.K. workers across 22 industries finds that one in three generative AI users are doing so without their employer’s knowledge, and half of all generative AI users have received zero training. Sixty-three percent of respondents report using generative AI at work. Productivity gains remain thin and uneven: only 7% save five or more hours weekly, while 31% report no time savings at all. Variation by sector is sharp, with information and communications workers five times more likely to report significant time savings than healthcare workers.
What this means for your business
Finance teams that measure AI adoption through approved license counts and usage dashboards are almost certainly working with a materially understated picture. If a third of generative AI activity is invisible to the organization, then low utilization rates on a company-sanctioned platform don’t tell you whether employees reject AI. They may just prefer a tool IT never approved. That gap corrupts the data CFOs rely on to make the next round of AI investment decisions, and most finance teams have no systematic way to close it.
The productivity numbers deserve more attention than the adoption headline. A 7% share reporting five-plus hours saved weekly sounds modest, but the industry spread tells the real story. Sectors where work is already document-heavy and text-driven, like information and communications, show a 21% hit rate on meaningful time savings. That’s not a marginal effect; it’s a signal about where AI tooling actually earns its budget. Funding a broad AI mandate across a mixed-industry enterprise based on blended averages will produce blended mediocrity. The return is hiding inside specific workflows, not across the org chart.
The deeper problem is what the survey reveals about shadow AI as a governance signal. When only 35% of generative AI users say their leaders demonstrate real understanding of the technology, the rational employee response is to work around the policy, not comply with it. That means the compliance gap isn’t primarily a technology problem or even a procurement problem. It’s a credibility problem, and another layer of policy written by executives who appear unfamiliar with the tools will not fix it. The budget decision to weigh here isn’t another enterprise license. It’s whether the internal AI literacy investment is real enough to bring hidden usage into the open, where it can be measured, managed, and actually improved.
Concept deep-dive: Shadow AI
Shadow AI refers to employees using artificial intelligence tools, typically publicly available ones like ChatGPT or Claude, without their employer’s knowledge or approval. It mirrors “shadow IT,” the decades-old pattern of workers adopting unauthorized software because sanctioned options feel slower or less useful. The business risk isn’t just security: it’s that company data enters external models, outputs land in work products without review, and the organization can’t measure, govern, or learn from any of it.
Based on reporting from Shadow AI use booms, finds new Deloitte survey, originally published 2026-09-18 10:00:00.
