A proposal has to be out by the end of the day. A long email thread needs summarising before a meeting that starts in ten minutes. Somebody opens a new browser tab, finds a free AI tool, pastes in what they have and gets a usable draft back almost immediately. It works. It happens again the following week. Nobody signed it off, because nobody was ever asked.
The short answer is that an AI usage policy for a business your size fits in an email. Three lines, sent to everybody, and they can genuinely be three sentences: which AI tools the business is happy for people to use, what must never be pasted into any of them, and who to ask when something sits between the two. It takes about half an hour to write. It is probably the best half hour a small business owner will spend this month.
The first line is the one people get wrong. A note that names no tools at all is a ban with extra steps.
Should we just ban AI tools at work?
No, and the ban is usually where this goes wrong.
A ban takes away the approved route. It does not take away the proposal that has to be out by the end of the day. The person still needs the hour back, and the tool moves onto a personal phone or a personal browser account, where the business can no longer see it at all. The behaviour survives the ban. Your view of it does not.
There is a version of this that works, and it costs money. Pay for a good tool, look at it properly before you buy it, set it up so it belongs to the business, and tell people its name. Then “please don’t use anything else” is a fair thing to ask, because nobody is being asked to give up the convenience. They are being asked to get it from a different tab.
We are firmly in favour of small businesses using this stuff, and we say so on client calls. Nothing here is a reason to use it less.
Which AI tools should a small business approve?
Start with what you already pay for. Plenty of small businesses are buying software with AI built into it, and the AI inside a subscription you hold already sits in the account your work lives in. That is a much shorter conversation than taking on a new supplier.
We are a Microsoft house, so for most of the businesses we look after the first name on that line is Microsoft 365 Copilot, on the tenant they have had for years. Yours might be something else entirely. The point of the line is that it has a name on it.
Somebody has to read the terms and the settings before anything gets approved, and it should be the same somebody every time. What happens to the material people put in, and whether any of it is used to train anything. Where the data is held and who can see it. Whether sign-in can be tied to your existing company account, so access arrives and leaves with the person. And whether the paid business plan is a different agreement to the free version of the same product, because it usually is.
Free tools get paid for somewhere. Often that is with whatever people put into them, on terms almost nobody opens.
What should never go into an AI tool?
Information that belongs to somebody else. That is the whole of the second line, and it is the heaviest of the three.
Your own draft copy, your own notes, a spreadsheet you built yourself: fine, into the approved tool it goes. A client’s contract, an export from a client’s system, a staff member’s sick note, a CV somebody sent you in confidence: those stay out of anything the business has not chosen and paid for.
The reason is duller than a data breach and far more likely to come up. Somebody puts a client’s document into a tool nobody at your business has looked at. That client’s information is now being handled by a company the client has never heard of, on terms nobody at either business has read. Your agreement with that client probably says something about who handles their information and where. This is a question about that agreement long before it is anything more dramatic. We are an IT company and not a firm of solicitors, so the wording of your own client agreements is a conversation for whoever writes them.
Say the line plainly and stop there. Overspecifying the second line is how a note turns into a document nobody finishes.
What should the note actually say?
Three lines, in this order.
The first line names the approved tools. At least one of them, by name, so the answer to “can I use AI for this?” is not simply no. Name the tool people already have and say plainly that this is the one you are happy for them to use.
The second line says what never goes in: information belonging to clients or to other people. One sentence. If people can recite it, it works.
The third line names a person. A human being who is happy to be asked “is this one alright?” and can answer the same day. A shared mailbox will not do the job, because people do not ask mailboxes awkward questions. Most of what arrives will be easy, and the point of the third line is that it arrives at all.
Write your own three lines. Copying somebody else’s note gives you somebody else’s words, and the ones that land name the tools your business actually holds and the work your team actually does.
What we do about this in our own business
We pay for the AI tools our team uses. Every one of them came through the same route: somebody read the terms and the settings properly, the tool got set up so it belongs to the business, and only then did we buy it. Our team knows not to use anything outside that list, and the reason it holds is that nothing was taken away from them. They still get the fast answer. They get it from a tool we chose.
We also run a discovery call before we quote for anything, and the same thing comes up over and over. We meet capable, AI-forward businesses doing genuinely good work with these tools, well ahead of most of their competitors. Then we ask what the team has been told, and the answer is nothing yet. We mention shadow AI, meaning any tool people are using that nobody signed off, and most owners have not heard the term. Nobody has been careless. They were never handed the three lines.
The way to make people stop reaching for a free AI tool is to pay for a good one and tell them its name.
The spend is small, the set-up is a one-off, and the note that follows is three sentences long.
Write those three sentences today and send them to everybody before you close the laptop, and the most capable person on your team stops having to guess where the line is.
Quick answers
Does this need to be a formal policy that everyone signs?
For most small businesses, no. An email everybody has actually read does more than a signed document nobody remembers. The exception is when somebody outside asks. Tenders, insurance renewals and large-customer questionnaires are starting to ask how a business governs AI use, so keep the note as a dated file somewhere findable.
Somebody has already put client information into a free tool. What now?
Write down what went where, and into which tool. Delete the conversation history and close the account where the tool allows it. Then send the three lines. The gap was never the person, it was that nobody had drawn the line, and the people who do this are almost always the ones getting the most done.
What about AI features already switched on in software we pay for?
Those count, and people forget them. Meeting transcription, note-takers that join calls, summaries in the mail client: all of it is AI use, and some of it is on by default. Name them in the first line so people know they are approved. Then say something about telling other people on a call when a note-taker is in the room, because that is their meeting too.
Should the note say anything about AI on personal phones?
One sentence, yes. What matters is the information, and that does not change with the device. A client’s document is out of bounds from a work laptop and out of bounds from a phone on the sofa. Saying that out loud stops the note being read as a rule about devices.
Related reading
If you would like any help or advice, get in touch today!
