Most businesses run at least three kinds of AI and have only counted one. There is the AI that arrived inside software you already pay for, the AI someone in the business built, and the AI staff use privately. Each needs a different question, and you cannot answer any of them until you have listed all three.
Why can't a business just say how many AI tools it uses?
Because most of it was never bought. Ask an owner and you will get a number, but it is usually the number someone has counted, not the number in use. The rest arrived without a decision: as an update to existing software, as something a staff member set up to save an hour, or as a tab someone opened on their own account.
None of those went through a purchase, so none of them are on a list. That is the whole problem, and it is also why the first governance step is dull and practical. Count what you have.
What are the three kinds of AI in a small business?
A useful way to sort them is by who put them there, because that decides what you can see and control.
1. AI inside software you already pay for
The assistant in your email or documents, the summary button in your meeting tool, the helper in your accounting package. Nobody chose it. It showed up in an update, and the whole team can use it from day one.
The question to ask: what is it allowed to see? These tools often have access to your mail, files and calendar by design. Know what is in scope and whether you can narrow it.
2. AI someone in the business built
A person set up an automation or a small assistant in a no-code tool to save themselves an hour. It might send emails or update records on its own. It works, so nobody reviews it.
The question to ask: what is it allowed to do, and who answers when it gets one wrong? This is the kind that acts, not just answers, which makes it the one most likely to cause a real problem quietly.
3. AI staff use privately
Someone is using a chatbot in a browser, on their own account, with whatever is in front of them. Often with good intentions.
The question to ask: what should never be pasted into it? This is a rule and a conversation, not a technical control.
Which kind is easiest to miss?
The first two. Nobody ever signed anything for them. The third at least involves a person doing something visible, even if you do not know about it. The first two sit inside systems that look like part of the furniture.
This is also why the framing helps. Different kinds of AI carry different risks and need different controls. A single "AI policy" that treats them all the same tends to fit none of them well.
How do I count them?
Make a list under the three headings. Start with what you know, then ask your team to add what you missed. Frame it as a way of finding out what is useful, not a hunt for who is doing something wrong. People tell you what they use when it feels safe to.
For each item, write one line: what it is, who set it up, and the question for its kind. That is enough. You are not building a register for an auditor. You are making the invisible visible.
Expect the list to be longer than you thought. That is normal and it is the point.
What do I do with the list?
Start with the second kind, because those act on their own. For each, check who is accountable and whether anything needs a human sign-off. Then look at the first kind for scope: what can each assistant reach, and under whose login. We cover that in more detail in what your AI assistant can reach and under whose login. For the third kind, agree a short rule about what never goes into a personal account.
At Handiwork we start every governance conversation with this list, before a single policy gets written, because the policy depends on it. For the wider picture, see our guide to using AI safely in a small business, and if you are letting any of it act on its own, how to use AI agents safely. Counting is also the first step of the work we do under our services.
What to do this week
Write down every AI feature that is switched on across the business, under those three headings. Ask your team to add what you missed. Put the list somewhere you will find it again. That is it.
Ready to find out where you stand?
If you want a straight look at what AI your business already depends on, our free AI Readiness Check takes about five minutes: AI Readiness Check.
Frequently asked questions
Do I really have AI I don't know about?
Very probably. AI features are being added to software that businesses already pay for, often switched on by default, and staff use personal accounts for quick jobs.
Is this the same as shadow IT?
It is closely related. The difference is that AI features can read and act on business data in ways older software did not, so it is worth listing separately.
Do I need special software to do an AI inventory?
No. A shared document with three headings is enough for a small business. Tools can help later, but the first list is a conversation.
Should I ban staff from using personal AI accounts?
A ban tends to push the use out of sight. A clear, short rule about what never goes in, plus a good approved option, usually works better.
How often should I update the list?
Quarterly, and whenever you add or change a major piece of software.
Ready to find out where you stand?
Take the free five-minute AI Readiness Check. There is no pitch at the end of it.
Take the AI Readiness Check
Sources
- The Definitive Guide to AI Security - Zenity (vendor guide; the idea that different AI deployment types carry different risks is used as background, no product or statistic is cited)
