AI compliance in a small business: what a business without a legal department should put in place
AI compliance sounds like lawyers, expert opinions and long checklists. A business with ten people has neither a legal department nor anyone who spends all day reading regulations. Even so, you can put quite a lot in order yourself, using tools you already know from leading your team: a list, a clear responsibility, fixed rules, training, data protection that has been looked into, and a person at the points where it counts. This article describes those six things. It isn't legal advice. Where we name laws, we only repeat what they say. For your individual case, your own legal counsel is the right person to ask.
Note: This article is not legal advice. It describes how we organize the use of AI in a business. Which rules apply to your company is a question for your legal counsel.
We work in Germany, so the laws named here are EU rules as they apply to us.
What we mean by AI compliance here
Compliance literally means following the rules. Which rules those are for your business in detail depends on your industry, your data and the AI systems you use. We can't judge that from a distance, and we don't do so here.
What we describe is the operational side: the order you need in order to be able to say which rules apply to you at all. If you don't know which AI systems are running in your business, you can't have anyone check whether they're being used correctly. That's why everything starts with a list.
If you're searching for the AI Act and SMEs, you usually want to know what applies to small and medium-sized businesses. What the regulation provides for small businesses in detail, we don't assess in this article. The six points below help you regardless of how the answer turns out for your business.
1. A list of all AI systems in your business
Write down which AI systems are used in your business. That includes the chat tools that individual people on your team use on their own initiative. One line per system:
- What: the name of the tool or the AI employee
- What for: which task, which process
- Who uses it: people or roles
- Which data it sees: client data, employee data, internal texts only
- What kind: a thinking partner that someone works with in conversation, or a process employee that starts on its own and reports back
- Contact person: who is responsible for questions and emergencies
The list is our house rule. This is how we handle it: it's the starting point for everything else. Whether, and in what form, the law requires an overview like this for your business, we don't assess. With it, you can quickly show your legal counsel what this is about.
2. One person who is responsible
Without a clear responsibility, nobody maintains the list. With us, the AI employees of a business have one person on the team who looks after them together with the owner. That person keeps the list up to date, knows the emergencies and is the contact person for the team.
Whether your business needs a separate title or a formal appointment for this, we don't assess. In practice, what helps at the start is a name, a deputy and the task of maintaining the list and the rules.
3. Fixed rules for each AI system
Every AI employee that takes over processes comes with a page of rules: role, trigger, permitted actions, hard limits, escalation path, report, review and emergency brake. A template for this is in An AI governance framework for a business of ten: everything on one page.
For chat tools your team works with in conversation, a few sentences are usually enough: what they're used for, which data doesn't belong in them, and who to ask when someone is unsure.
Who checks the work of the AI employees once several processes run on their own is described in AI governance in SMEs: who checks the AI when you no longer do?.
4. Training the team
Here we name one of the two legal provisions in this article. Anyone using AI systems in a business has to take measures under the European AI Act to support the development of AI literacy among their staff (Art. 4 AI Act, newly worded since July 2026). The law doesn't require a specific level.
Which measures those are in your business is your decision. In our work, part of this literacy comes from training the AI employee. The team watches how the AI employee works, checks its first runs, decides the uncertain cases and teaches it new variants in the feedback session. Anyone who has done this once knows what the AI employee can do, where it becomes unsure and when a person has to take over.
What makes sense beyond that, for example a short introduction for everyone who works with chat tools, we don't assess for you. Kevin has written down on his blog how he approaches training: AI Literacy Under the AI Act: Training Staff.
5. Clarifying data protection for each system
For every AI system on your list, we ask the same questions. That's how we work in the program, not legal advice:
- Which personal data does the system see?
- Is there a data processing agreement with the vendor, and for exactly the plan you use?
- Where does the system run, and who has access?
- Whoever is responsible for data protection in your business: has that person read through the answers?
Our house rule on this: answer the questions first, then let real client data flow. Whether the answers are sufficient in your case is for your data protection advisor to judge. With us, the documents stay in the client's own filing system. We don't run servers for the client; we help set up the AI employees in their business. Where they run and how far that can be set up in line with data protection rules, we decide together in the first sessions.
6. A person for people, money and legal consequences
Our house rule: with us, a person stays involved when a decision affects people, when money is involved or when it has legal consequences. An AI employee prepares, sorts and reports. Deciding, approving and signing are done by a person. Every step that can't be undone has its own approval in front of it.
Where decisions affect people legally or in a similarly significant way, the law also sets limits on purely automated decisions (Art. 22 GDPR). For filing, the inbox and reports, though, our rule is an operational one, not a legal one.
At the lowest level, it looks like this: your team checks every run before anything moves on. How this level works and why we don't skip it is described in Autonomy Level 1, Assisted: when your team checks every run.
Five signs that something is still missing
You don't have to put everything in order at once. These five signs show you where to start. They're our assessment, not an authority's checklist.
- Nobody can say which AI tools are used on the team. Then the list from point 1 is missing.
- When a question about an AI employee comes up, everyone looks at each other. Then the responsibility from point 2 is missing.
- An AI employee does something nobody has written down. Then the rules from point 3 are missing.
- Client data ends up in a tool where nobody knows the plan or the contract. Then point 5 is missing.
- A decision about a person or about money went through without anyone seeing it. Then point 6 is missing, and this point is the most urgent one.
What all five have in common: order is missing, and you can create it yourself with your team. For the legal questions behind it, the next section will help.
What you're better off not settling yourself
You can tackle the six points above without a lawyer. A few questions still belong with someone who knows the field:
- The classification of individual AI systems under the AI Act, if you're unsure what a system is being used for.
- Rules for your industry. Whether your profession has its own rules is something your professional chamber or your legal counsel can clarify. What that looks like at a tax advisory firm, we describe in What a German tax advisory firm lets run on its own today.
- Contracts with vendors, for example data processing agreements.
- Anything that affects employees, for instance when an AI system processes data about staff.
For these, your own legal counsel or your professional chamber is the right place to ask. With the list from point 1 and the rules from point 3, you're well prepared for that conversation.
What this looks like for us
In Kevin's own business, 14 AI employees are at work. Each one has a personnel file, a fixed role and clear limits.
In Kevin's business, the personnel files are the core of the list from point 1. For every AI employee, they also state which tools and access it has. A person sets up the access and keeps it encrypted; it isn't written down in plain text anywhere.
If an AI employee is missing a piece of information or two sources contradict each other, it reports that instead of guessing. For this topic, that's the most important sentence in our view: a guessed detail easily goes unnoticed, a reported one doesn't. Who approves what in Kevin's business is in the article on the framework from point 3. An AI manager reviews the work of the others. That role is still on trial.
This article itself is an example of that. We checked the two sentences on the AI Act and the GDPR against the wording of the laws beforehand. We deliberately make no other legal statements here.
What we don't do: legal advice, data protection advice and the legal classification of individual AI systems. We haven't taken that on for our clients either.
Runs without me
Every two weeks, one process that runs without the owner: what came in, what the AI did, where a person decided. Two minutes to read.
Worksheet
AI Rulebook on One Page
What your AI employee may do, what it must never do, who keeps an eye on it and what is checked before release: eight fields to fill in and an acceptance checklist. Comes as a PDF by email once you confirm.
Strategy session
In a 20-minute strategy session, you tell us where things get stuck. We'll tell you honestly whether an AI employee is a good fit for you.
Book a strategy sessionFrequently asked questions
Does all this apply to a small business too?
Which rules apply to your business in detail is a question for your legal counsel. The six points above are operational order. That helps you at any size, and it makes the conversation with your legal counsel shorter.
May my team enter client data into a chat tool?
We don't give legal advice on that. Our house rule: first clarify which data the tool sees and whether there's a data processing agreement for exactly your plan. Until that's settled, no real client data goes in.
How often should I update the list?
Every time a new tool or a new AI employee is added, and on top of that once a month, during the joint look at all processes. In our program, this monthly review is a fixed part: for every process, the current level is entered and checked.
Do I need an AI officer?
Whether that's needed in your case, we don't assess. In practice, you need one person who is responsible, maintains the list and knows the rules. What you call that person is secondary.
Who is responsible if an AI employee makes a mistake?
We don't give legal advice on that either. Operationally, we make sure mistakes show up early: at the start, the team checks every run, sign-off only comes after clean runs, anything uncertain goes to a person, and every step with no way back has an approval in front of it.
Your next step
In a small business, AI compliance starts with order: knowing what's running, who is responsible and where a person decides. When the first AI employee is to take over a process in your business, we set its responsibility, rules and escalation path at the same time. Where you could start, and whether Owner Autonomy is a fit for your business, we'll go over on a call, along with how it works and what it costs.