You’ve probably sat in a meeting where organisational use of AI in the workplace is on the agenda. A few people nod. Someone asks sensible questions about data, ethics and whether it sits comfortably with organisational values. Someone else points out that there’s already a policy. The item is discussed, but no clear decision or action is agreed.
Meanwhile, AI is already in your organisation. This summer the ONS found that 55% of UK employees use AI for work, while only 35% of businesses say they use it. Your people are experimenting quietly, on their own, with whatever tools they’ve found.
Some of them are probably doing brilliant things, and you’re unlikely to hear about it.
The picture is similar globally. McKinsey’s 2025 survey found that 88% of organisations use AI in at least one part of the business, yet nearly two-thirds haven’t begun to scale it across the organisation. There are lots of pockets of activity and very little joined-up learning.
I’ve seen this pattern many times with new ideas and innovation projects. When there isn’t a clear remit, people go under the radar and experiments happen in isolation. If people don’t feel safe saying ‘I tried this and it didn’t work’ or ‘I’m not sure what I’m allowed to do’, the learning stays with whoever tried it. Meanwhile, AI is advancing at a rate of knots.
Three things can help.
1. Give AI an owner
The first is to give AI an owner: someone with the time and authority to move its use forward. Resist the urge to hand it to IT by default, because this is more a question about how your organisation works than about technology. It helps to think of AI as a smart but forgetful colleague in their first role, who needs clear instructions and someone to check their work.
2. Set the boundaries
The second is to set the boundaries. Governance can sound like the opposite of experimentation, but in my experience people try new things more freely when they know where the edges are. A short, clear statement of which tools are approved, what data stays out and how to flag a problem gives your team permission to explore. Without it, the cautious people hold back and the enthusiastic ones inadvertently carry risk.
3. Run proper experiments
The third is to run proper experiments, where you’re testing a way of working with a clear view of what success looks like. Pick one process where admin gets in the way of progress. Agree what you’re testing and how you’ll know if it worked. Then talk openly about what you learned, including what flopped. Pockets of unstructured tinkering can create plenty of activity, but they give you little chance to learn from what works and scale it.
Many AI tools are free or low cost to start with, so all of this is possible in your organisation, whatever your budget. The important thing is to treat AI as part of the team. That takes leaders who are curious, who make it safe to admit uncertainty, and who bring the learning out into the open where everyone can use it.
A useful question to take into your next team meeting is this: how might we work with AI to do this? The ‘we’ is the important part. It makes AI something you work with and figure out together.
If you’re a senior leader trying to get your team talking constructively about AI, I’d be glad to help you facilitate that conversation. Get in touch.
With thanks to Tayo Richards, whose AI for Charities guide sparked this post.