Walk into almost any business right now and somebody is using AI to get their work done. They might be on a personal account. They might be pasting things in that they shouldn’t. And nobody has told them to stop, because nobody has decided what the rules are. That’s shadow AI, and it’s one of the main things businesses now bring me in to sort out.
The instinct is to shut it down. I think that’s the wrong move, and it misses the bigger opportunity sitting underneath it.
Shadow AI is a signal, not just a risk
When someone goes off and finds their own AI tool, they’re telling you something. They’ve found a part of their job that could be faster or better, and they’ve gone and fixed it themselves. They understand their role better than anyone else in the building, so the way they’re using AI is usually worth learning from.
So the first thing I do isn’t write rules. It’s find out what people are actually using AI for, and why. If you skip that and go straight to a ban, you lose the best practice your own team has already worked out.
The real problem: it never becomes the business’s
Here’s the pattern I keep seeing. You’ve got individuals using AI in certain places to make themselves more efficient, and they’re doing well. Take a finance lead who knows their role inside out and has built their own way of working with AI. They’re individually better off. But none of it is retained by the organisation. Nobody else benefits, and if they leave, it leaves with them.
That’s the step most businesses haven’t taken yet: going from a few people who are good with AI to an organisation that knows what’s worked and can spread it across everyone else. It’s the same idea I talk about on stage as AI-added versus AI-native.
Two fronts: compliance and productivity
I think about this on two fronts. On one side is compliance: what are the rules we need to follow to be where we’ve said we are? On the other is maximum productivity: what would let people do their best work? The job is to find the middle ground between them.
The compliance side matters more than people expect. A lot of organisations don’t realise how much of what they’ve signed up to, in their ISO standards or their Cyber Essentials, they’re quietly breaking across the business through the way AI is being used day to day.
The fix I recommend most often
This is a conversation I have almost every day. People across a business are on all sorts of AI plans, and not on an approved team or enterprise plan. That matters because of your clients’ data. Without the protections those plans give you, what staff paste in can be used to train the model, and the AI provider is acting as a third-party data processor you never agreed to. You don’t know where your liability sits, and you could be exposed.
What goes in a workable AI policy
Most of the time the work starts with writing an AI policy. These are the decisions I’d make sure it covers.
Which tools are approved. You can lock the policy to one tool or leave it open-ended. Open-ended sounds friendlier, but it means paying for every subscription on a proper business plan, because personal accounts don’t give you the data protection you need. For most smaller organisations I recommend against open-ended. It keeps cost and compliance manageable.
Who owns what. Somebody has to be the named admin for each tool, and somebody has to be responsible for the data that goes into it. That usually sits with senior leadership, and it should be written down by name.
How it fits your other policies. An AI policy shouldn’t live on its own. It has to complement what you already have, whether that’s data protection, social media or safeguarding, and it should follow the same format so staff recognise it.
Whether people can actually read it. The point is for staff to clearly understand what safe, compliant use looks like. If it reads like a legal document, nobody will follow it.
Bring people with you
There’s a change management piece here that gets forgotten. People have started to enjoy the way they’re working. If you turn round and say “you can’t do that any more, do this instead”, they won’t buy in. You have to bring them on the journey: understand what they were doing, then give them an approved set of tools that does the same job properly.
It also helps to look outside your own walls. I take a look at what good practice looks like across the rest of the industry, then bring that back into the business.
Something I’d never say to a team is “stop using that” or “you can’t do this”. It’s about reframing it: finding a compliant way to get the same outcome, and working with staff to use AI properly. An AI policy is a great place to start, as long as it puts a harness on that ambition without stamping it out.
People know their own job better than anybody else. So the ways of working they’ve created should be kept wherever possible. They’re the things people will buy into most. The real questions are how you replicate that across other departments, and how you get buy-in across the whole business.
Where I’d start
If you do one thing, do this: get everybody onto a proper team plan for the tool you’ve chosen, and configure it so it has access to the systems people actually use. In my experience that’s usually the biggest single step. It takes AI use out of personal accounts and puts it somewhere the business can see, support and learn from.
Then write down what’s working, and share it. That’s when AI stops being something a few people do quietly and starts being something the business is good at.
Let’s talk
If this struck a chord, I’d like to hear what you’re building.



