AI Workshop Switzerland.

The day has four blocks: where you stand, collecting use cases, trying them out, ranking and deciding. Less than an hour of it is presentation. The rest is work on your own tasks.

• Block 1, about 90 minutes — where you stand. What is already being done with AI in your business, officially and unofficially? In almost every company, individuals have long been using tools management knows nothing about. We collect that without apportioning blame, because it is the most honest starting point — and because the first data protection questions usually surface right there.
• Block 2, about two hours — collecting use cases. We walk through a team's week looking for tasks with three characteristics: they repeat, they arrive unstructured, and their result can be checked quickly. From a pool of twenty to forty ideas, five to eight serious candidates typically survive.
• Block 3, about two hours — trying them out. Two or three candidates get tested directly, on real documents and real cases from your business. This is where it shows which morning idea fails in the afternoon on missing data, unclear rules or simply insufficient quality. This block is the reason the day pays for itself.
• Block 4, about 90 minutes — ranking and deciding. Every candidate gets an assessment of effort, expected value, data risk and preparatory work needed. Out of that comes an order, a ninety-day plan and a decision about who inside the business owns what.

There are breaks between the blocks, and the day ends at 17:00. We work with the tools you already have or could have — the workshop is not tied to a vendor and sells no licences. If it becomes clear during the day that a use case goes beyond straightforward tool use, we place it: for tasks that have to run several steps by themselves, see AI Development.

Six to twelve participants from the operational work, plus someone with system access and someone who can approve a budget. Below six people you lose the range of tasks; above twelve the hands-on part becomes a spectator event.

The format suits SMEs approaching the subject in a structured way for the first time, teams inside larger companies preparing a decision for their area, and management groups having to choose between contradictory recommendations. It explicitly suits sceptical teams too: a day on which two use cases fail in front of everyone and one works produces more clarity than three months of discussing potential.

It is less suitable when a single person is sent along to convince everyone else afterwards — that brief is unfair and almost always goes wrong. Equally unsuitable when the decision has already been taken internally and the day is meant to confirm it. Tell us beforehand; then we save you the day and talk about delivery instead. The terms that keep coming up during a workshop are documented publicly — for instance prompt engineering, AI agents and RAG, the method by which a model reaches your own documents.

By the end of the day you hold four results: a ranked use-case list, two or three workflows actually run through, a tested set of instructions and a written data rule. All of it is usable without us.

• The use-case list holds, per entry: what would be automated or supported, how often the case occurs, which data is needed, how the result gets checked, and a rough placement of the effort. The list is deliberately written so that an internal IT supplier could read and implement it.
• The workflows run through show on real cases what comes out — including the places where the quality is not good enough. Those negative findings are the most valuable part, because they stop you spending money later on a case that will not carry.
• The set of instructions consists of the phrasings that actually produced usable results in the hands-on block, with your terminology and your examples. That saves everyone in the team the first fortnight of trial and error.
• The data rule records which categories of information may go into which tool and which may not — client data, HR data, contracts, internal costings. One page, in plain language, so it gets applied in daily work.

On top comes the ninety-day plan: three to five steps with owners and dates. We send the documents within five working days of the workshop. Whether you work with us afterwards is open — the day is built to be worth it on its own. If a use case points towards content and visibility, our article on AI content strategy for Swiss SMEs is the right thing to read next.

For the workshop we deliberately publish no list price; after a short briefing call you receive a written quote with a fixed amount. The reason is not secrecy but that three variables drive the effort, and a flat rate would either be too expensive for small groups or too tight for preparation-heavy days.

• Number of participants. From about ten people we run the hands-on block with two facilitators, otherwise nobody gets a turn.
• Preparation. An impulse day with no preparation is a different thing from a day for which we review your workflows, document storage and tool landscape beforehand so we can start on real cases.
• Location. At your offices, in our studio in Zollikon, or online — including travel for locations outside the Zurich area.

What comes afterwards is transparent. If the workshop produces a use case someone wants built, it starts here as a validation prototype from CHF 10,000 — a working first version tested on real data, solid enough to carry the decision about building it out. If the day shows instead that your real problem is being found inside AI answers, ongoing GEO work starts at CHF 2,500 a month. Classic search engine optimisation starts at CHF 800 a month, local visibility at CHF 300, and a website runs between CHF 3,000 and CHF 25,000 depending on scope.

We name these figures here so you can place the day: the workshop is the cheap part, and its job is to stop you spending the expensive part in the wrong place. If the honest answer by the evening is that none of it pays off in your business at the moment, that is a valid result — and we will say so.

Switzerland has no overarching AI act. Existing law applies, above all the Federal Act on Data Protection, which is technology-neutral and therefore covers AI applications. The Federal Office of Justice is preparing a consultation draft by the end of 2026 that is expected to address transparency, data protection, non-discrimination and supervision; the Federal Council is pursuing sector-specific adjustments rather than adopting EU law one to one.

For many Swiss businesses the EU AI Act is nonetheless relevant as soon as they address people in the EU. Its transparency obligations under Article 50 have been enforced since 2 August 2026: a system interacting directly with people has to be recognisable as AI where the interaction happens, and synthetically generated content has to be marked in machine-readable form. The stricter requirements for high-risk systems were postponed to December 2027 by the Digital Omnibus that came into force at the end of July 2026.

In the workshop we treat this not as legal advice — we cannot and may not provide that — but as a practical operating rule. Concretely we settle three questions: which categories of data may go into an externally operated tool at all? Where is the data processed, and can the location be fixed? And who inside the business decides when a new use case appears? Those three answers on one page prevent more trouble than any training slide. For anything beyond that we refer you to a lawyer rather than pretending to be one.

A workshop creates clarity and a list, but no running system — and it has no effect if nobody decides anything afterwards. We name these limits before you book, not after.

• No finished system. What emerges in the hands-on block is workflows run on real cases, not a production-ready solution. Getting from a working run to a reliable process with permissions, logging and approvals is a project in its own right.
• No data preparation. If your knowledge sits in paper folders, in scanned images with no text recognition, or in people's heads, the day will show that mercilessly — it cannot fix it. Often the data preparation turns out to be the real first project.
• No substitute for everyday training. One day changes a team's capability only so far. What sticks is your own tested examples. Lasting competence only appears when someone inside the business maintains the collection afterwards.
• No legal advice and no certification. We place the applicable duties in practical terms and document your decisions. Binding legal assessments and conformity evidence belong with a law firm.
• No effect without a decision. The most common reason such days lead nowhere is that nobody with budget authority was in the room. If that person cannot attend, move the date.

There are three usual routes after the day, and all three are valid outcomes — including the third. Which one it becomes is decided by the list, not by us.

Route one: you implement it yourselves. The two or three simplest cases need no project, they need discipline — a named owner, the maintained set of instructions, and a date six weeks out to check what is actually being used. For this route the workshop is entirely sufficient, and many smaller businesses do best with it.

Route two: one use case gets built. Where a case goes beyond plain tool use — several steps, access to your systems, approvals before anything is written — it becomes a development project. What such an undertaking involves is described under AI Development, and the first version is built as a prototype from CHF 10,000.

Route three: guidance over months. Where several departments are affected, governance has to be settled or the tool landscape has grown organically, an ongoing engagement follows instead of a single project. Scope, sequence and boundaries are set out at AI Consulting Switzerland; for engagements with an agency relationship in the city of Zurich we also run AI Agency Zurich and the service module AI advisory.