Shadow AI use booms, finds new Deloitte survey
As the push for different types of artificial intelligence moves through corporate finance, AI use at the employee level may be just as important for risk-aware CFOs.
One in three workers who use generative AI are doing so without their employer’s knowledge, a practice known as shadow AI, according to a new Deloitte survey of 25,000 working adults in the U.K. across 22 industries. The finding illustrates a much murkier picture of AI adoption than the software licenses or approved pilot programs some CFOs are leading would suggest.
On top of that finding, nearly two-thirds (63%) of those surveyed said they have used generative AI, and half of the people using it have received no training from their employer on how to do so.
Employees surveyed said they have largely settled on simple use cases, such as writing emails and producing summaries. Though superficially those tasks are low risk, employers still do not know which tools some workers have chosen or what types of information workers are feeding those models. Many teams, according to the survey, have no view into the company information being entered or where the output ends up.
This hidden use makes an already difficult investment case harder to read. Though finance teams can track spending on the AI tools implemented by the organization itself, they cannot easily measure the time saved through tools it does not know its employees are using. This, especially amid the debate of a single source of truth, creates several issues for CFOs.
The company’s AI count may be wrong
Generative AI adoption at an organizational level is often measured through approved licenses and usage rates. Deloitte’s findings, however, suggest those figures could miss a meaningful share of activity.
That leaves finance teams working with an incomplete set of data to take action off of. Low use of an approved platform can look like limited interest in AI or that it’s not working, but employees may simply prefer a public tool they already know. The license or usage data that decisions are being made off of would not capture that activity.
Despite the impact on decision-making datasets, leadership may not be giving workers much reason to come forward. Only 35% of generative AI users said their leaders speak about the technology with a good understanding of it.
This is another example of why executive communication and knowledge demonstration skills are so important, because policy written by people who appear unfamiliar with the tools is unlikely to draw hidden users into the open.
The report does not explain how companies can build an honest inventory before making another round of AI purchases. But finding out which tools employees already use and why they chose them would be a good place to start. Those answers may be more useful than another top-down adoption target.
The productivity payoff remains uneven
The survey also puts some distance between AI use and measurable productivity.
Thirty percent of all respondents said generative AI saves them at least some time each week. Slightly more (31%) had used it at work and reported no time savings. Just 7% said the technology saves them five hours or more per week.
It’s worth noting here that results varied sharply by industry. In information and communications, 21% of respondents reported saving at least five hours a week. That figure fell to 5% in healthcare and social work.
The variation matters when leadership teams are deciding where to fund AI projects. A broad mandate to increase usage will not say much about the return if employees are using AI under the radar, how and when they want. Finance needs to see whether a tool reduces the time required for a specific process by implementing a specific type of tool for that process. If the work still needs extensive review, the productivity gain may be slow to come to fruition or was never there to begin with.
Training and psychological safety are also major parts of that equation. If half of generative AI users say they have received no training whatsoever, companies may be judging the technology based on workers figuring it out as they go. The high performers will likely find use cases and productivity gains, but others will add a new step to the job without improving the result.
In the U.K., and probably in the States too, AI use has already spread through the workforce, and a sizable portion of it is happening out of sight. How leaders can remedy that is still up in the air.
