Is AI Safe for a Small Swiss Business? What to Check Before You Start
AI can be used safely in a small Swiss business when five things are settled before the first customer name goes into a tool. Training on your data is switched off. You know in which country the data is stored. Every tool runs in an account that belongs to the business. Nothing leaves the business without a person approving it. And one page of writing says which tool holds which data. All five are decisions an owner can take in an afternoon.
Most small firms have not taken them yet. In March 2025 the research institute Sotomo surveyed 300 Swiss SMEs for AXA. Among the firms already using AI, only a third had a clear rule on data protection. Among firms with five to nine employees, it was 23 percent.
So the tools are in use, and the rules are mostly missing. This article is about closing that gap without a legal department.
One thing first. I am a coach and I set up AI for small businesses. I am no lawyer. What follows is the practical routine I use, and I will point out where a lawyer belongs.
Where the Risk Sits in a Small Business
When owners think about AI risk, they often picture something dramatic: a hacked system, a leaked database. In a business of eight people, the everyday risk is much smaller and much more likely. Somebody copies a customer email into a free chatbot on a private account because it is the fastest way to get a polite answer written.
Three things are wrong in that small moment. The account belongs to the employee, so the business has no control over what is stored there. Several consumer versions of AI tools use conversations to improve their models unless someone switches that off. And nobody decided whether customer emails should go into that tool at all.
It happens without any bad intent, because the tool is helpful and no rule exists.
The legal side is less exotic than it sounds. The revised Swiss Data Protection Act has applied since 1 September 2023, and it applies to a business of four people as much as to a bank. Your customer list is personal data. So are the emails your customers write to you. If an AI tool processes them, you should be able to say which tool, where it stores the data, and why it is allowed to. That is the point where your privacy notice comes in, and where a short conversation with a lawyer is money well spent.
Everything before that point is housekeeping, and an owner can do it.
Five Checks Before the First Customer Name Goes In
First, the training switch. Every serious AI tool has a setting that decides whether your conversations may be used to train future models. Business versions usually have it off by default, while free and personal versions often have it on. Find it, switch it off, and take a screenshot of the setting for your records.
Second, the country. Ask where the data is stored. For Swiss customer data, I choose storage in Switzerland or the EU whenever the tool offers it. Many tools let you pick the region once, when the account is created, and never again. That makes it a decision for day one.
Third, the account. Every tool should run in an account that belongs to the business, paid by the business, with the owner as administrator. When an employee leaves, the account stays. And once the work with an outside helper like me is over, nothing of yours sits on their side.
Fourth, the approval. Decide what AI may do alone and what needs a person. My own rule is simple. My AI agents may send email only to my own inboxes. Anything addressed to another person is saved as a draft, and I read it before it goes. The same applies to posts: nothing is published without my yes. A small business can adopt exactly this rule from the first day.
Fifth, one page of writing. List each tool, what data it sees, where that data is stored, and who in the business may use it. Add one sentence for the team on what must never go into any AI tool. For most small firms that is health details, payment details and anything a customer told you in confidence. This page is what your lawyer will ask for when the privacy notice is updated, and it is what you will be glad to have if a customer ever asks.
What Stays With You
Safety is also a question of who decides, long after the settings are done.
An AI assistant can be wrong in a calm and confident tone. It can draft a reply that promises a delivery date you never agreed. The protection against that is a habit more than a technical setting: the assistant prepares and a person sends. I read what goes out in my name, and I read every conversation the AI front desk on my website has with visitors. It takes a few minutes a day and it keeps the judgment where it belongs.
Good tools also have limits of their own. In July 2026 one of my AI agents refused an instruction from me five times in a row, and explained why. I was irritated in the moment. Looking back, I would rather work with a tool that says no out loud than with one that nods and does something else.
A common worry is breaking something. The answer is to start where little can break. Begin with material that contains no customer data at all: your own texts, your product photos, supplier price lists, the wording of your standard offer. Let the assistant draft from those for a few weeks. You will learn how it behaves, and the customer list can follow once the five checks are done.
If you do one thing this week, open the settings of the AI tool you already use and look for the training switch. It takes two minutes.
Source for the survey figures: AXA SME Labour Market Study 2025, carried out by Sotomo.
AI setup and coaching for owners of small businesses is available at The Change Republic, in Zürich and across Switzerland, with the privacy routine described here built into every setup. Find out more at www.thechangerepublic.com/ai-setup-small-business
Tünde Lukacs is an executive coach, keynote speaker on AI and leadership, and founder of The Change Republic. She sets up AI for owner-led small businesses across Switzerland and coaches their owners to run it themselves.