AI in Romandy SMEs: where to start without mistakes

In many SMEs across French-speaking Switzerland, generative AI is already in the building. Someone uses ChatGPT to rephrase emails, someone else tried Copilot, a third person watched a demo of an agent that “does everything”. Motivation is rarely missing. A starting point almost always is.

Before talking about tools, here are the three mistakes we see most often in a first AI project, and how to avoid them.

Mistake 1: starting from the tool instead of the problem

“We need a chatbot.” That sentence opens more doomed projects than any other. A technology is picked first, then a place to plug it in is found. The result: a tool installed, barely used, and a team concluding that “AI isn’t for us”.

The right question is not which tool? but which task eats time, repeats itself, and follows rules clear enough to be written down? Sorting incoming requests, drafting a quote from a template, summarising meeting notes: these are serious candidates. Next year’s sales strategy, much less so.

Mistake 2: aiming too big, too fast

The second trap is badly measured ambition. A project touching five departments, three business applications and the customer database from day one is very likely to get bogged down in integration and compliance.

A first use case should be:

• bounded: one team, one flow, one measurable result;
• reversible: if it doesn’t work, you go back to the old way without damage;
• visible: gains should show within weeks, not quarters.

That first success is less about “making AI pay” than about building trust, in the team and in management.

Mistake 3: forgetting the people

A tool nobody knows how to use properly produces nothing. Generative AI requires a new skill: writing a precise instruction, checking an answer, knowing when not to trust the model. It can be learned, but it can’t be improvised.

That is why training should never be a line added at the end of the budget. It is part of the rollout, just like the technical setup.

The method: diagnose before you automate

To avoid those three mistakes, we work in four steps.

1. Map the workflows. Interviews and field observation, team by team: which tasks, which tools, where the friction and double entry are.
2. Diagnose the opportunities. For each friction point, we ask whether generative AI or a no-code automation can help, and at what cost in effort and risk.
3. Prioritise by expected return. A simple gain-versus-effort matrix surfaces the quick wins and the longer projects.
4. Write the roadmap. A phased plan, with team training built into every milestone.

The deliverable is not a report that gathers dust. It is a short, ranked list of projects, each with a scope, a cost estimate and the skill to acquire.

What about data?

It is often the first concern, and a legitimate one. Since 1 September 2023, Switzerland’s revised Federal Act on Data Protection (nFADP) regulates the processing of personal data more strictly. In practice, before putting anything into an AI tool, you need to know where the data is processed, by whom, and for what purpose.

A good audit builds this in from the start. Some use cases can run on mainstream cloud tools, while others justify an in-house solution, such as self-hosted n8n or a model deployed in Europe.

In short

• Start from a concrete task, not a technology.
• Start small, measurable and reversible.
• Train people while you deploy.
• Settle the data question before, not after.

Want to know where AI can genuinely help your organisation? See our audit method or get in touch.