Most agency leaders would say, with some confidence, that they know how their team is using AI. New research suggests otherwise.

Deloitte's UK GenAI Workforce Survey, covering 25,000 working adults across 22 industries, found that 63% of respondents have used generative AI at work. Of those users, a third are doing so without their employer's knowledge. Half received no training at all. UK workers are now spending close to 1 billion pounds of their own money each year on AI tools for their jobs, often because the tools their employer provided either do not exist or do not fit the work. Only 35% of users think their leaders understand AI well enough to have a proper conversation about it.

This is not a discipline problem. It is a visibility problem, and it is one Go Wisely has seen up close.

When a 50-person agency brought Go Wisely in for an AI readiness diagnostic earlier this year, the assumption going in was fairly typical: some people were experimenting with AI, a handful of tools were in use, and the leadership team wanted a clearer picture before deciding what to formalise. The diagnostic, structured across six dimensions including governance, found something closer to the Deloitte pattern than anyone expected. AI was already threaded through parts of the business nobody had mapped. Not because people were being secretive, but because nobody had asked, and there was no easy way to say so.

That distinction matters. A team hiding its AI use is rarely doing so out of malice. It is usually doing so because asking permission feels slower than asking forgiveness, or because there is no clear channel to say "I've started using this for X" without it sounding like a confession. Deloitte's finding, that only around a third of users trust their leaders to have a real understanding of AI, points to the same root cause. People do not hide tools they trust their employer to have a sensible view on. They hide tools they suspect will be either banned outright or ignored.

For agencies specifically, this gap carries a sharper cost than it might elsewhere. Client work moves fast, deadlines are unforgiving, and account teams will reach for whatever gets a deliverable out the door on time. If that means running a brief through an AI tool the agency has never reviewed, on a client's confidential material, the risk is not hypothetical. It sits inside the exact contract clauses agencies sign every day, around confidentiality, data handling, and intellectual property.

The fix is not a crackdown. Following the diagnostic, the agency ran a six-session AI Accelerator training programme built around the tools its own team was already using, rather than a generic curriculum imposed from outside. The programme closed with a 4.6 out of 5 satisfaction score across 131 survey responses, and not one dissatisfied response among them. The reason it worked was not that people were told to stop improvising. It was that improvising stopped being necessary, because the agency finally had a clear, current picture of what was already happening and could build guidance around the reality rather than the assumption.

That is the real value of the Governance dimension in an AI readiness diagnostic. It is not a compliance exercise bolted onto the end of a project. It is the answer to a genuinely uncomfortable question: does anyone in this business actually know what AI is doing inside it right now? For most agency leaders, the honest answer is "not entirely," and Deloitte's numbers suggest they are far from alone.

None of this requires a big programme to start. It starts with a conversation, an honest look at what tools are already in use, on whose initiative, and why. The businesses that get ahead of this are not the ones that ban AI, and they are not the ones that ignore it either. They are the ones that go and look, properly, at what is already happening, and build from there.

Go wisely.