Most SMEs already have an AI policy. It just isn't written down anywhere.

It goes something like this: everyone uses whichever tool helps right now, with whichever account happens to be open. Nobody decided it. It simply happened, because the tools arrived in the business faster than any agreement about them.

In an SME, AI governance is not a committee or a handbook. It is a decision about who may do what with which data, plus a list that shows what is actually running. Both fit on two pages.

What AI governance means for an SME, and what it doesn't

In a large company the term stands for a whole programme: an ethics board, risk classes, audits. In a business of ten or thirty people that would be theatre. Here, AI governance answers four questions:

  • What is allowed? Which tools, for which tasks, with which data.
  • What is already running? A list of every AI tool in the business, including the small ones.
  • Who decides? Who approves a new tool and who answers for a result.
  • How does everyone know? A short introduction and one place where the rules live.

What AI governance is not: a legal review. What Swiss data protection law and the EU AI Act actually require of a Swiss SME is covered in AI and data protection in Switzerland. This article is about the internal side: how do you make sure your team can follow rules it actually knows?

Shadow AI: the tool is already there

The sprawl has a name: shadow AI, AI tools in use without the business knowing about them or approving them. The Work Trend Index published by Microsoft and LinkedIn in May 2024 shows how common it is. It surveyed 31,000 knowledge workers in 31 markets, Switzerland among them. 78 % of those who use AI bring their own AI tools to work, averaged across all markets rather than a Swiss figure (Microsoft, May 2024).

The figure comes from a vendor that sells AI tools itself, and it measures habits, not damage. For your business it still says something simple: the question is not whether your team uses AI, but with which account.

The account is the heart of it. A free personal account is a contract between your employee and the provider, not between your company and the provider. What happens to the inputs, whether they are used for training and where they are stored is then not something your business controls.

A ban rarely fixes this. Ban a tool without offering a better one and the use simply moves to a personal phone. It works better to provide an approved tool on a business account and to set rules for everything else.

The AI policy on one page

An AI policy that people actually read is short. It covers six points:

  1. Approved tools: which AI tools may be used in the business, and with which account. Anything else needs approval.
  2. The data rule: in one sentence, which data never goes into an AI tool. For example: "No customer names, no health or payroll data, no passwords, no contracts."
  3. Duty to check: whatever an AI writes, calculates or summarises is read by a person before it leaves the business. Whoever passes it on answers for it.
  4. Disclosure: when you tell customers that AI was involved. For a chatbot on your website, always.
  5. New tools: how someone proposes a tool and who approves it (more on that below).
  6. Contact person: who answers questions, and to whom you can report a mistake without getting into trouble.

The last point matters most. People who are afraid to admit a mistake don't report it. You then hear about a wrongly pasted client file only when the client calls.

How to word the data rule, and how to measure a tool's benefit before rolling it out, is covered in AI strategy for SMEs.

Green, amber, red instead of a long list of bans

Nobody remembers forty prohibitions. Three levels, they do. For an accounting firm, it might look like this:

LevelExamplesRule
Greenrewording text, drafting emails without personal data, researching technical questionsfreely allowed with an approved tool
Ambersummarising minutes, drafting quotes, searching internal documentsbusiness account only, result is checked
Redpayroll, tax files, assessing job applications, decisions about peoplenot in AI tools unless explicitly approved and checked

The split is an example, not a standard. A medical practice draws the line differently from a painting firm. What matters is that the line is drawn and the team knows where it is.

The red level is not arbitrary. Where AI makes decisions about people or handles delicate personal data, data protection law bites hardest. The details are in the data protection article.

The AI register: what is actually running

A policy says what should apply. The AI register shows what does. It is a table with five columns:

ColumnExample
Tool and accountMicrosoft Copilot, business licence
Used forsummarising emails and minutes
Which datathe team's mailboxes and shared files
Who is responsiblehead of administration
Approved on3 November 2026, next review in six months

It also covers the invisible cases: AI features built into software you already use, such as the CRM, the accounting package or the phone system. They often arrive with an update, not a purchase.

A tool that reaches into your shared files finds everything the person asking has read access to. Microsoft says so explicitly for Copilot: it surfaces data to which a user has at least view permissions (Microsoft Learn, as of November 2026). If sharing in SharePoint has grown over the years, the payroll list in a wrongly shared folder turns up in answer to a simple question. That is why tidy filing and permissions come before any such tool. How to get there is covered in The digital workplace for SMEs.

Who is responsible for what

An SME does not need new positions. It needs three names:

  • Management: carries overall responsibility and signs the policy. It decides on the red level.
  • An AI lead: keeps the register, assesses new tools and answers questions. It is a side role, often held by whoever already looks after IT.
  • Everyone else: follows the policy and says what they would like to use next.

Whoever passes on an AI result is responsible for it, not the tool. That sentence belongs in the policy word for word.

The approval route for a new tool

An approval route must not take longer than the detour via a personal phone. Otherwise it gets bypassed. Five steps are enough:

  1. Proposal: whoever wants a tool names the task, the tool and the data.
  2. Check the terms: is there a business account? Are inputs used for training? Where is the data stored? Can sign-in run through the existing company account with single sign-on?
  3. Classification: green, amber or red, using the table above.
  4. Trial: two to four weeks with a small group, with a measurement before and after.
  5. Register entry: with a responsible person and the date of the next review.

Proposal, check and classification should take days; the trial afterwards takes its weeks. If you can say yes quickly, you rarely need to say no.

Training: short and recurring

A one-off presentation at launch is forgotten by the summer. Three things work better:

  • Thirty minutes at launch: the policy, the three levels and an example from your own day-to-day work.
  • One case per quarter: a real mistake or near miss, discussed without names. What happened, and what do we change?
  • New staff on day one: the policy belongs in onboarding, alongside the key and the password.

People who are new to AI need the introduction most, not the experienced users.

What the framework does not replace

An internal policy is not a legal review. It helps you meet your obligations day to day, but it does not define them. In Switzerland, AI is governed today by existing law, above all the Federal Act on Data Protection. On 12 February 2025 the Federal Council came out in favour of ratifying the Council of Europe's convention on AI and of adapting Swiss law for it in individual areas where possible (Federal Council press release, in German). Ratification itself is up to Parliament.

If you are looking for a framework for a formal management system, two templates are widely used: the ISO/IEC 42001 standard from 2023 for an AI management system, and the AI Risk Management Framework from the US institute NIST, published in January 2023, with its four functions Govern, Map, Measure and Manage. For an SME, both are reference works rather than blueprints.

What ALPENIQ does here

In AI governance we work out with you which AI tools are already running in the business and which data flows into them. We assess the tools for data protection and security, write the AI policy for your team, define roles and responsibilities and document your AI processes.

One limit belongs here: we do not give legal advice. For a data protection impact assessment or sensitive contracts, a data protection lawyer belongs at the table.

Conclusion

AI governance in an SME needs four things: a one-page policy, three levels instead of a list of bans, a register with five columns and three names who carry responsibility. Set them up before the first mistake happens, and you will need to ban less and can say yes faster.

Further reading

Frequently asked questions (FAQ)

What is AI governance?

AI governance is the set of rules, roles and processes that determines how a business may use artificial intelligence. In an SME it usually consists of a short AI policy, a register of the tools in use, an approval route for new tools and clearly named people who are responsible. It does not need a committee or a handbook.

What belongs in an AI policy for employees?

An AI policy names the approved tools and their business accounts, a data rule stating which data never goes into an AI tool, the duty to check results before passing them on, disclosure towards customers, the route for new tools and a contact person. It fits on one page and applies from an employee's first day.

What is shadow AI?

Shadow AI means AI tools used in a business without the company knowing about them or approving them, often on free personal accounts. According to the 2024 Work Trend Index by Microsoft and LinkedIn, 78 % of AI users among knowledge workers bring their own tools to work, averaged across 31 markets. The risk lies less in the tool than in the account: a personal account is not a contract with the company.

Should an SME ban ChatGPT and other AI tools?

A ban without an alternative usually just moves the use to personal devices, where the business sees nothing at all. It works better to approve a tool on a business account, set a clear data rule and put sensitive tasks into a red level that needs explicit approval.

Who is responsible for AI in an SME?

Management carries overall responsibility. For day-to-day work, one named AI lead is enough: someone who keeps the register, assesses new tools and answers questions, often the person who already looks after IT. For every AI result that leaves the business, the person who passes it on is responsible.