Switzerland runs on AI now. Quietly.

In May 2026, EY surveyed 604 people in Swiss companies and found that 89 percent already use AI solutions in their daily work. Not piloting. Using, daily. The same study's second finding explains why nobody is celebrating: most organizations are still in the early stages of scaling AI systematically and embedding it strategically. Among SMEs the curve is steep too: adoption rose from 22 to 34 percent in a single year, and two thirds plan to integrate at least one AI tool by the end of 2026, but only 18 percent have a structured plan for it.

That is the Swiss pattern in one line: usage everywhere, operating model unchanged. Quiet, pragmatic, tool-first adoption, and then a plateau. EY's respondents name the blockers themselves: a lack of technical talent and skills (30 percent) and difficulty identifying viable use cases (27 percent).

The use-case problem is a legibility problem

I have spent twenty years building and operating companies in and around Zürich, a good part of it automating financial back offices, and I have never once seen a company that truly lacked AI use cases. What they lack is processes described precisely enough to hand over. A use case is just a process someone can write down: inputs, steps, exceptions, outputs, and who checks the result. If nobody can describe the process on one page, no model can run it and no vendor can save you.

This is why I think Switzerland is sitting on an unfair advantage it has not cashed in. Swiss companies document. Process discipline, audit trails, quality norms: the cultural stereotype is real, and in the Framework's terms it is exactly the raw material AI needs, because machines can only absorb process that has been made explicit. The country that wrote everything down is better positioned for delegation to machines than the countries that move fast and document nothing.

What a Swiss SME should actually do

• Pick one process you can already describe end to end on a single page. Invoice intake, first-draft offers, support triage. One.
• Hand that process to a machine, keep a named person as its owner and reviewer, and measure cycle time and error rate for a quarter.
• Only then take the second process. A structured plan is not a strategy deck; it is this loop, repeated.

The 89 percent number says the tools have already won the argument. The 18 percent number says the operating model has not caught up. Quiet adoption suits Switzerland. Quiet transformation would suit it better, and it is the leadership layer, not the workforce, that decides which one we get.