Training teams in AI without drowning them

In many SMEs and mid-sized companies, AI training follows the same script. A half-day plenary, a presentation of "what is possible", access to a tool, then silence. Three weeks later, a few enthusiasts use it heavily, most people never do, and leadership concludes that teams "are not ready". The problem is rarely the teams' ability. It lies in how the training is designed: too broad, too abstract, and too early relative to real needs.

This guide offers six steps to train without overwhelming. It builds on our analysis of why AI tools end up in the drawer and belongs to our Organization & adoption practice.

Why teams get overwhelmed

Three mechanisms recur. The first is information overload: ten use cases are presented when a person will retain one, and only if it matches what they do on Monday morning. The second is missing context: a demonstration on a generic example does not transfer to a client file, a quote or a month-end close. The third is missing permission: without clear rules on what may be entered into the tool, caution wins, and caution translates into non-use.

Step 1: start from tasks, not from the tool

For each team, identify three or four repetitive, text-heavy or document-heavy tasks that take time and whose output is easy to check: summarizing meeting notes, drafting first replies to recurring requests, preparing proposals, searching internal documents. Ask team leaders, not IT, to nominate them.

If a task is poorly defined or depends on ten unwritten exceptions, AI will not make it clearer. The process needs to be clarified first, which falls closer to our work in Operations & processes. Training comes after.

Step 2: separate three role levels

A single programme for everyone is the most common cause of saturation. We recommend three levels.

Users

They need little: how to phrase a request with context and constraints, how to check a result, how to recognize the tool's limits. Two short sessions of 60 to 90 minutes, centered on the Step 1 tasks, are typically enough to get started.

Champions

In each team, one or two volunteers, not necessarily the most technical, who go deeper, collect good practices and answer colleagues' questions. They keep usage alive between sessions.

Decision makers

Executives and managers do not need to write advanced prompts. They need to understand what the tool does and does not do, which risks it carries, how to set priorities and what to measure. A two-hour session built around decisions to make is worth more than a general talk.

Step 3: set a minimal rulebook

Before training, write one page. It states which tools are approved, which data must never be entered (personal data of clients or employees, trade secrets, confidential information), who validates content before it goes outside, and whom to ask when in doubt.

This is not a formality. In Switzerland, the revised Federal Act on Data Protection (FADP), in force since 1 September 2023, already sets transparency and personal data protection requirements that apply to AI use. For companies operating in or selling into the European Union, the AI Act, in force since 1 August 2024, applies in phases: prohibitions since February 2025, obligations for general-purpose AI models from August 2025, most other obligations from August 2026. The text also includes a requirement for staff who work with such systems to have an appropriate level of AI literacy. Documented training, proportionate to roles, therefore helps beyond adoption. Each company should still assess its own scope with its legal advisers.

Step 4: learn on real work

The rule is simple: every participant leaves a session with a usable result on a file they already had to handle. Bring real documents, anonymized where needed, and work on them. Then schedule a short follow-up, 20 to 30 minutes, one to two weeks later, to share what worked and what did not. Habits form in that second moment.

A concrete illustration: in getting an AI assistant adopted in customer service, progress rested on cases from the current week, not on demo scenarios. Likewise, to speed up the onboarding of seasonal staff, the knowledge base was built from the questions newcomers actually asked.

Step 5: build relays, not events

A training session is an event, adoption is a process. Give champions half an hour a week, a simple sharing channel and a small collection of tested instructions per team. Managers must legitimize learning time: if nobody frees a slot, training becomes one more item on an already full list.

Step 6: measure what matters

The number of people trained measures activity, not outcome. Track instead, for each Step 1 task, frequency of use, time saved as estimated by the teams themselves, and observed quality (rework, corrections, feedback from internal clients). Three indicators are enough. Review after two months: drop tasks that bring nothing, reinforce those that work.

An order of magnitude, with its assumptions

Take a company with 60 employees involved. Assumptions: 3 hours of initial training, then 30 minutes a month for six months, so 6 hours per person. Total: 360 hours. At a fully loaded cost in the order of CHF 80 per hour, that is roughly CHF 29,000 of working time, excluding external fees. If half of the people, 30, save one hour a week on their tasks, that frees 30 hours weekly and the invested time is recovered in about twelve weeks. These figures are illustrative: the point is to set your own assumptions before starting, then compare them with measured results. If the real gain is ten minutes a week, it is better to know early.

What leaders can do this week

Choose two pilot teams, not the whole company. Have their leaders nominate the tasks. Write the one-page rulebook. Schedule two short sessions on real work and a follow-up. Set three indicators. Then extend team by team, reusing what worked.

Training in AI without overwhelming means doing less, but closer to the work. If you want to establish your starting point and define a path proportionate to your organization, our diagnostic is a sound first step.