The most expensive answer to "how do we get started with AI?" is "we buy licences for everyone and see what happens". The answer that holds up: one process, four weeks, three figures, one written decision. This guide describes that 30-day pilot, what it costs and what it lets you decide.

Before the pilot: choosing the right process

Take a week and note, team by team, the tasks that keep coming back: data entry, reminders, letters, meeting notes, answers to the same questions. Look for repetition, not for "AI": repetition is what creates the gain. A task that comes up ten times a week and takes twenty minutes each time adds up to more than three hours a week. That is the kind of task that justifies a pilot.

Keep just one, using the gain / effort / risk grid described in our guide on finding the three right use cases: plenty of hours, little set-up effort, and a mistake that stays in-house or is reviewed before it goes out. A good subject fits in one sentence with a volume: "invoice reminders, around fifteen a week, fifteen minutes each".

Scoping: half a day with the people who do the work

The pilot is prepared during a 4-hour AI Workshop, on site or remotely, with the people who run the process. The session serves as both initial training and scoping. Five things are set:

  1. The actual flow. Who receives what, who does what, where it gets stuck. Not the process described in the quality manual.
  2. The start and the end. Where AI comes in (the incoming email, the document) and where it stops (the draft ready for review). Nothing more during the pilot.
  3. Permitted data. What may go into the tool and what stays out, according to the three levels internal, client and protected in our security guide.
  4. Review. Who reviews each AI output before it goes out. During the pilot, everything that leaves the company is reviewed.
  5. The decision date. Day 30, in the diary, with the managing director.

You leave with a written action plan, ready-to-use instructions for the chosen task and a one-page sheet on data. The half day costs CHF 1,900 excluding VAT for a group of up to 20 people.

The 30 days, week by week

  • Week 1: building the habit. The people in the pilot use AI on every case in the process. Everything is reviewed. Time spent is recorded, corrections included.
  • Week 2: refining the instructions. A fifteen-minute check-in: what works, what has to be rewritten from scratch, which cases AI should no longer handle. You adjust the instructions, not the tool.
  • Week 3: tackling the difficult cases. The exceptions arrive: an unhappy client, an incomplete file, missing information. You decide which ones fall outside the scope.
  • Week 4: measuring. You take the three figures from day 0 again and prepare the decision.

Three figures, measured on day 0 and day 30

  • Time per task, in minutes, including review and corrections. That is the gain.
  • Correction rate: the share of AI outputs that had to be substantially rewritten or thrown away. That is the risk.
  • Actual use: how many people use the tool on the process, and how many times a week. It is the best indicator of what will happen if you extend it.

A pilot with ten indicators tracks none of them. These three are enough to decide, provided day 0 was measured before AI arrived.

Day 30: a written decision

You compare the figures and write down one of three sentences:

  • "We extend it": time goes down, corrections remain rare, and the team uses the tool without being reminded.
  • "We adjust and run another 30 days": the gain is there but runs into a specific obstacle (instructions, incomplete data, a frequent exception).
  • "We stop on this process": the gain is not there. You move on to the second use case on the list.

A pilot that ends without a decision has proved nothing, even if everyone found the tool "brilliant". The return calculation that goes with the decision is explained in our ROI guide.

What the pilot does not prove

  • That AI will transform the company. It proves that AI holds up on one process, with one team.
  • That the chosen tool suits everything. Another process may need another tool.
  • The budget for a full rollout. The pilot gives orders of magnitude (volume, time, corrections); the budget is then built on those figures.

What the first cycle costs

The AI Workshop (CHF 1,900), business licences for the people in the pilot for one month (a few tens of francs per user, at the vendor's public price) and your team's time. A few thousand francs in total, not tens of thousands.

If the decision is to extend, equipping the team and follow-up are provided under AI Adoption Support: at a fixed price on quotation, or as a subscription from CHF 1,800 a month for 10 hours, with a three-month minimum. The full breakdown of budget items is in our analysis of what AI costs a Swiss SME.

Who does what

The managing director chooses the process, books the day-30 date and signs off the decision. A member of the team, the future AI lead, follows day-to-day use and records the figures. Our team runs the workshop, prepares the instructions and helps you read the results. Without this split, the pilot becomes the managing director's personal experiment, and the team does not follow.