Major AI companies may still be struggling to turn a profit, but there’s little doubt that fortunes are already being spent on access to leading models. However, a new report from Deloitte has revealed that much corporate spending on AI is not coming from corporations themselves, but from their workers. Employees in the UK are spending approximately £958m of their own money per year on LLM-based tools, with this outlay coming from 17% of workplace AI users in the country.
Deloitte’s study, which canvassed 25,000 workers in Britain, finally puts a figure on a phenomenon that has increasingly come to be known as ‘shadow AI.’ And while unapproved AI usage is likely to be alarming enough for employers as it stands, some analysts suspect that this trend could become more widespread in the near future.
This is partly because UK organisations have been too slow to roll out approved AI tools, yet there may also be a tacit willingness on the part of employers to tolerate shadow AI if it delivers productivity gains. However, analysts argue that this approach could carry risks, and that businesses should endeavour to provide a greater variety of sufficient AI tools, as well as strong training and guidance on proper usage.
Shadow AI provides ‘a clear productivity benefit’
Based on responses gathered between May and June of this year, Deloitte’s inaugural GenAI Workforce Survey yields an up-to-date snapshot of UK employees and AI. 63% report “knowingly” using AI for work, while 24% use AI every single day. Among adopters, 43% use the likes of ChatGPT and Claude for searching for information, 43% for drafting emails, and 31% for creating summaries, among other things.
So far, so predictable. Yet Deloitte’s survey also included questions on subscriptions and types of tools. It found that 46% of workers are using free plans or tools, 34% are using external models paid for by their employers, and 17% are using in-house AI applications. Another 17% are using their own money to pay for at least one LLM-based tool, and Deloitte estimates that the total annual cost of this is now £958m.
Performing some back-of-the-envelope calculations, and assuming an employee population of 29.75 million, this amounts to a cost of £300 per year (or £25 per month) for every part- or full-time UK employee who’s paying out of their own pocket for AI. Given the cost of living crisis faced by a significant portion of the British population, you might be forgiven for thinking that many people would want to avoid an additional expenditure like this. However, Deloitte UK’s Head of Industry Insight, Paul Lee, suggests that workers are gaining a clear benefit from AI, making the expense worthwhile.
Workers are willing to pay for generative AI because they see a clear productivity benefit
“Workers are willing to pay for generative AI because they see a clear productivity benefit and, in some cases, believe the tools offer capabilities that outperform employer-provided alternatives,” he says.
Similar explanations apply to shadow AI, with Deloitte finding that 31% of AI workplace adopters use tools without their employer’s knowledge. According to Paul Henninger, Partner and Head of Technology & Data at KPMG UK, unapproved AI usage can largely be explained in terms of the relative inadequacy of employer-provided tools.
“There are very few controls or restrictions on consumer AI compared to enterprise AI which is almost always deployed in a way that’s safer and therefore more restricted,” he says. “People are turning to shadow AI because without enterprise controls they see an opportunity to work faster.”
No implicit disapproval if output is higher
Deloitte’s survey indicates that AI is indeed helping at least some people work faster. The UK workforce claims to be saving 70 minutes on average per week because of ChatGPT and similar models (although 31% of employees have used AI without saving any time whatsoever). A combined total of 32% workers are saving at least one hour per week because of AI, while 7% claim to save at least five hours.
Admittedly, the report does not delve into the value of this saved time and the work produced using AI. However, there seems to be some tacit belief or understanding among employers that this value is largely positive, while there’s also an implication that some may be aware that their employees are using shadow AI.
“Importantly, employers do not implicitly disapprove,” says Lee. “In fact, a significant proportion of those using generative AI without their employer’s knowledge believe their employer would approve if they knew.”
In Lee’s opinion, Deloitte’s findings point less to deliberate or would-be rule-breaking, and more to a gap between how employees now work and how their employers maintain oversight. He also suggests that many workers may simply be ahead of the curve in relation to their organisations, and that compliance teams are unable to keep up with new tools and models.
Employees are clearly eager to use AI tools they perceive as helpful
He adds, “Employees are clearly eager to use AI tools they perceive as helpful, sometimes moving faster than formal policies, governance frameworks or approved tool rollouts.”
Henninger agrees that, while shadow AI may come with operational and security risks, the growing use of such AI generally doesn’t represent malicious behaviour. Rather, it represents “employees using readily available technology to solve everyday problems.”
As such, he urges employers to prioritise the provision of “secure tools that are genuinely useful,” along with clear guidance on how to use them. Lee advocates for something very similar, arguing that the implication of Deloitte’s study for employers is that they should focus less on restricting usage, and more on providing training and approved pathways for workers to use generative AI safely and effectively.
Shadow AI is here to stay, for now
If stronger organisational emphasis on approved AI is not forthcoming, both Lee and Henninger expect shadow AI to remain a feature of work for the foreseeable future. In fact, Henninger reports that KPMG expects “there to be an increase in both enterprise and shadow AI.” This is because many UK organisations have not committed enough resources to the roll out of approved AI tools, meaning that workers have had to fill gaps. “Shadow AI will likely always be an attractive tool they can use to think through challenges in parallel,” he adds.
For Lee, shadow AI will remain a feature of the workplace for a while yet, largely because workers believe LLMs help them to “work more effectively” and access capabilities they wouldn’t otherwise access. Despite this, he does affirm that employers in Britain are increasingly making an effort to provide official alternatives.
66% of employees have relied on AI outputs without evaluating their accuracy
He explains, “We are definitely seeing employers increase their offer of generative AI tools to their employees – there is a marked contrast to just a year ago, when most large companies were focusing on in-house developed tools.”
Such a change perhaps cannot come soon enough, if only because a lack of approved tools generally appears in tandem with a lack of AI oversight and training. And as Henninger explains, this can pose considerable risks.
“KPMG’s global research found that 66% of employees have relied on AI outputs without evaluating their accuracy, while over half report making mistakes because of AI,” he says. “Businesses therefore need useful, approved options, visibility of where data is flowing and simple guardrails built into everyday workflows.”
At the same time, Henninger proposes that employees should also feel able to disclose AI usage, something which would perhaps require a change in culture within many organisations. This may entail that employers almost openly encourage their employees to use, and spend money on, their own preferred AI tools. However, this may be a price worth paying for now, since Henninger argues that, without a culture of disclosure, “usage will be driven further underground and the risks will increase.”
Major AI companies may still be struggling to turn a profit, but there’s little doubt that fortunes are already being spent on access to leading models. However, a new report from Deloitte has revealed that much corporate spending on AI is not coming from corporations themselves, but from their workers. Employees in the UK are spending approximately £958m of their own money per year on LLM-based tools, with this outlay coming from 17% of workplace AI users in the country.
Deloitte’s study, which canvassed 25,000 workers in Britain, finally puts a figure on a phenomenon that has increasingly come to be known as ‘shadow AI.’ And while unapproved AI usage is likely to be alarming enough for employers as it stands, some analysts suspect that this trend could become more widespread in the near future.
This is partly because UK organisations have been too slow to roll out approved AI tools, yet there may also be a tacit willingness on the part of employers to tolerate shadow AI if it delivers productivity gains. However, analysts argue that this approach could carry risks, and that businesses should endeavour to provide a greater variety of sufficient AI tools, as well as strong training and guidance on proper usage.
