A business owner I know found out by accident. He was walking past a desk and glanced at a screen. One of his team, a good lad. Had ChatGPT open in a browser tab. Free version. Personal account. Talk about being bad at using AI.
And in the chat box, pasted wholesale, was a client quote. Company name, site address, pricing, margins, the lot.
“How long have you been doing that?”
“About six months. It’s brilliant for tidying up the wording. Why?”
This tore the business owner. The lad wasn’t wrong. It was making his quotes better. He’d shown initiative, found a tool, solved his own problem.
Exactly the kind of person you want on your team.
He’d also been feeding six months of client data. Pricing and margin information into a free consumer AI tool. On a personal account, the business has no visibility into or control over, and no idea where any of it goes.
Good initiative. Terrifying execution.
And if you think that’s not happening in your business right now, I’ve got news for you.
The Audit You Can’t Currently Pass
Let me ask you a few questions.
Not rhetorical ones; actually try to answer them.
Who on your team is using AI right now?
Not “are they allowed to.” Who actually is?
What are they using?
ChatGPT? Claude? Gemini? Free versions or paid? Personal accounts or company ones?
What are they putting into it?
Client names? Quotes? Contract details? Financial information? Anything you’d be uncomfortable explaining to that client over the phone?
Who’s good at it?
Someone on your team upskilled themselves at getting AI to do useful work. Do you know who? Do you know what they’ve figured out?
If you can’t answer those questions
And almost no business owner I speak to can
Then you’ve got what I’d call shadow AI.
Your team adopted a powerful tool without you. Each in their own way, on their own accounts, with their own judgement about what’s safe to paste in.
You didn’t decide this.
It just happened.
The tools were free, they were useful, and your team are problem-solvers. Of course it happened.
But “it’s happened” is not a strategy.
And what you’ve ended up with has three problems bigger than the data one.
Problem 1: There’s No Company Way. Only Personal Way
Take something simple. Quotes going out the door.
One of your team has worked out how to get AI to write a cracking quote.
Clear, well-structured, sounds like your business.
Another is getting mediocre results. Because their prompting amounts to “write a quote for this.”
A third doesn’t trust the AI stuff and does it all by hand, taking three times as long.
A fourth is using it, but the output sounds nothing like you: too formal, too American, too much waffle.
Same task. Same business.
Four different standards, depending on who picked up the job.
Now multiply that across everything your team touches.
Emails, proposals, job notes, complaint responses, reports.
AI hasn’t given your business a capability. It’s given eight individuals eight personal capabilities of varying quality.
Your clients don’t experience your team as individuals.
They experience your business. And right now, what your business looks like depends on who happened to answer.
Problem 2: The Good Stuff Lives in Personal Chat Histories
Here’s the one that should bother you.
That team member who’s got good at this, and someone has.
Has built up months of knowledge from trial and error.
Prompts that work. Ways of phrasing things. “I always paste in the site survey first, then ask it to structure the quote in our three sections.”
That knowledge is real business value.
It’s a faster, better way of doing your work.
And it lives in one place: their personal chat history, on their personal account, in their head.
When they leave, and people leave, it walks out the door with them.
Nothing written down. Nothing shared.
The next person starts from zero, fumbling through the same six months of trial and error.
You’re not building a backlog of processes or SOP for your business.
You’re renting them, and you don’t hold the lease.
You’ve spent years making sure the business isn’t dependent on any one person. Job sheets, processes, handovers.
And then AI arrived.
It rebuilt the dependency problem in the background, without you knowing. This time in browser tabs you can’t see.
Problem 3: The AI Knows Nothing About Your Business
This is the root of the problem.
Every time anyone on your team opens ChatGPT, they’re starting from a blank page.
The AI doesn’t know your pricing rules. Doesn’t know your tone, that you’re a “Cheers, Phil” business, not a “Kind regards” one.
It doesn’t know your quote structure, your terms, your standards. What good looks like, which client gets which treatment. Or the hundred small rules that make your business your business.
So every single time. It starts with someone trying to explain your business. Badly, partially, differently every time.
The quality of what comes out depends on how good that one person is at explaining. That’s the whole reason for the eight-personal-ways problem.
The blank page is the problem.
Nobody should be starting from a blank page in your business, because your business isn’t a blank page. It’s twenty years of accumulated standards and know-how.
The fix isn’t better prompting.
The fix is AI that already knows what you want it to know about your business.
What “Already Knows” Looks Like
Imagine this instead.
A team member needs to get a quote out. They don’t open a chat window and start explaining.
They open the company’s AI, and it already has your quoting process loaded.
Your structure, your tone, your rules about what gets included and what never does. They give it the job details, and out comes a quote that sounds like your business, built the way you build them.
The same way, whoever’s driving.
The complaint email that lands on a Friday afternoon? Same thing.
Your complaint-handling approach is already in there. The tone you take, the steps you follow, when to escalate to you. Whoever picks it up handles it your way.
This exists.
It’s called a plugin.
How do they work?
Well, I build them on Claude, using something called Cowork.
Anthropic built Cowork for exactly this.
If you would like to know more about AI Agents like this, check out this 4-part series.
Taking how your business works and packaging it so your whole team’s AI runs on it. Your processes, your standards, your knowledge. Loaded in advance, available to everyone, owned by the company.
Not a tool your team prompts from scratch.
A set of ready-made workflows built around how you work.
We’ve finished building one right now for a construction industry client.
Their tender process, packaged so the whole team runs it the same way.
I’ll walk you through exactly what’s in it and what it changed in Part 2 of this series.
For now, here’s the reframe that matters:
The answer to shadow AI isn’t a ban.
Bans don’t work.
The tools are too useful, and your best people will get quieter about it. They won’t tell you they are using it.
You’ll trade visibility for hidden use.
The answer is giving your team something better than their personal workaround.
Something that produces better results than their private ChatGPT habit. Because it already knows the business, on accounts you control, with rules you set.
Your people didn’t go rogue because they’re careless.
They went rogue because the company offered them nothing.
Offer them something better and the shadow problem solves itself.
Where to Start
Before any of that, you need to know what you’re actually dealing with.
Which is the audit you couldn’t pass at the top of this post.
That’s what the AI & Automation Readiness Report maps for you.
Five minutes of questions about how your business actually runs. And it shows you where AI is already in use, where the gaps are.
It’s free, and it’s the groundwork for everything else in this series.
Get the AI & Automation Readiness Report
What’s Coming Next
This is Part 1 of 4.
Part 2 goes into detail about building a Cowork plugin.
What actually goes into a company plugin, and the before-and-after. This is where you’ll start spotting your own processes in the story.
Part 3 is the bigger picture:
Your organisation’s own marketplace of AI workflows.
A menu of ready-made, approved ways of working. Sales workflows for the sales side, admin for the office, packaged know-how for the field.
Control without policing.
Part 4 is the practical one: getting your first plugin built. How to pick the right process, what the build looks like, timeline, cost.
Until then, one thing to do this week:
Ask your team, no telling-off.
Who’s using AI and what for.
What comes back will surprise you.
Someone’s doing something clever.
Someone’s doing something risky.
And I’d put money on both of them being the same person.
