Governance has a poor reputation in many organisations, often deservedly. It is introduced as a control instrument, grows into a multi-stage approval process, and leads motivated teams to find ways around it. The end result is more risk than before, because usage now happens out of sight.
The purpose of AI governance is different. It should answer in advance the questions every team would otherwise have to ask individually.
Three classes, not one process
A simple split of use cases into three classes, each handled differently, works well in practice. For the large majority of everyday uses — drafting text, summarising, research without personal data — no approval is needed, only clear rules and enablement. For applications touching personal or customer data, defined guardrails and a named responsible person are needed. For systems that influence decisions about people, a formal assessment, documentation and human oversight are required.
The gain is that roughly eighty per cent of cases fall into the first class and run without waiting. The remaining cases receive genuine attention in return.
Governance that treats every case alike treats no case appropriately.
Regulation as a prompt, not a goal
The EU AI Act also affects Swiss companies as soon as they serve customers in the EU or deploy systems there. The pragmatic response is not to launch a compliance project, but to build the overview you need anyway: which systems are in use, what data they process, who owns them, and which risk class they belong to.
In most organisations that overview is the real bottleneck. It is missing not through negligence, but because AI use grew from the edges, usually faster than any inventory.
The simplest first step
Start with a register. A list of every AI-supported system in the company, with its purpose, data basis, responsible person and a rough risk assessment. That list can be produced in a few weeks, costs almost nothing, and changes the quality of every discussion that follows.
It almost always surfaces surprises. That is uncomfortable, and exactly why it is worth doing.