6 November 202610 min read

AI Training for Teams: How to Plan One That Sticks

An overhead projector with a glowing amber lamp sits on a bare table beside three pencils, a small stack of blank worksheets and a kitchen timer, in front of a half-lowered projection screen.

Good AI training for a team teaches colleagues to apply AI to their own daily tasks, check the results, and follow shared rules on data. It is not a tool tour. The format that fits most Swiss teams is a hands-on workshop of half a day or less, built on real work and followed by a measured trial. This guide shows how to decide what to teach, pick a format, run the session and check that it paid off.

TL;DR

Start with a common baseline (AI limits, checking outputs, data rules), then add role-specific workflows. Microsoft’s guidance on AI literacy and upskilling draws the same line.
Map your gaps before booking anything: pick two or three roles, watch the work, and ask people which tasks they want help with.
A focused workshop can be short. ETH Zurich’s generative AI template runs 180 to 210 minutes for 4 to 12 people.
Write down which data may go into which tool before anyone practises on real material.
Record a baseline on one real task, then compare time, corrections and quality after the training.

What AI training for a team should cover

Teams often confuse three things: knowing what AI is, knowing how to use it in a given job, and being ready for how roles will change. Harvard Business School Online separates them as AI fluency (fundamentals, capabilities, governance), role-based AI skills (applying AI to role-specific tasks) and AI upskilling (preparing for future roles). A good program has all three layers, in that order.

The baseline layer is literacy. According to Microsoft’s overview, that means knowing AI’s limits, verifying outputs and following data-handling guidelines. Upskilling builds on it with role-specific workflows, real tasks, measured impact and quality controls.

In practice, the content of a solid session usually includes:

How generative AI behaves: why it sounds confident when it is wrong, and what that means for review.
Prompting basics: giving context, a goal and a format, then iterating.
Writing, editing, research and summarizing: the tasks most office roles share.
Spreadsheets and data interpretation: where many finance, operations and sales teams get the quickest gains.
Privacy and data handling: what never goes into a tool.
Quality control and escalation: who checks what, and when a human decides.

That list follows the modules in Coursiv’s 2026 guide to employee AI training. The D2L guide, adapted from the U.S. Department of Labor’s AI Literacy Framework, reaches a similar set of five competencies: fundamentals, prompt engineering, output evaluation, responsible use and workplace applications.

One more element is easy to skip: the worry. Go1’s guide lists change readiness as part of effective training, meaning time to address the fear that AI will replace people. A team that is anxious learns less. Say plainly in the first ten minutes what the training is for.

Find your gaps before you pick a format

A generic course teaches everybody the same thing, and that wastes most people’s time. A short diagnosis costs a few days and pays for itself.

Start with two or three roles

The Conference Board of Canada’s Work Reimagined report recommends a task redesign audit. You identify two to three roles where AI has already changed daily work, using job-shadowing and short interviews. Then you update job profiles, build a skills-gap matrix, deliver short focused training, and monitor completion and feedback before scaling.

For a team of 10 to 40 people, this might be customer support, marketing and finance. Sit with someone for an hour in each. You will see where copy-paste between tools happens, where reports are rebuilt by hand and where people already use AI quietly.

Ask the team what they want help with

A short survey beats guessing. The DACH needs-assessment template from Sprad, written for the Swiss, German and Austrian market, covers role and work context, how often people use AI tools, their confidence in trying new software, their grasp of the basics, and their top three tasks they would like AI to support. Those last answers are gold: they become your workshop exercises.

Do not apply one standard to everyone

DL Academy argues that the most common mistake is a single standard across the organization, and that the riskiest gap is where confidence runs ahead of capability. WPMS AI Consulting adds that one job can contain both low-risk drafting and high-risk decision support, so a job title alone does not tell you what skill is needed.

A practical way to use this: list tasks, not titles. For each task, note how sensitive the data is, what happens if the output is wrong, and who reviews it. Training for the high-impact tasks goes deeper than training for the low-impact ones.

Choosing a format: workshop, series or coaching

The right length depends on what you want to change. Three formats cover most needs.

Format	Typical length	Best for
Hands-on workshop	About 3 to 4 hours	Getting a whole team to a shared baseline and a first set of workflows
Strategy workshop	120 minutes	Decision-makers choosing which use cases to pilot
Workshop plus follow-up coaching	Workshop, then several weeks	Teams that need new habits to stick in real projects

Two Swiss references show how compact a good session can be. ETH Zurich’s workshop template for generative AI is designed for 4 to 12 participants and takes 180 to 210 minutes. Participants sign in to at least one tool beforehand, so no time is lost on accounts. The Swiss AI Association’s 120-minute guide is for 4 to 8 decision-makers and ends with a top three of use cases, a pilot and an owner.

Small groups matter. With more than about a dozen people, hands-on time per person shrinks and the quiet ones stop asking questions. For a larger company, run the same session several times by department rather than one big lecture.

On-site, online or hybrid is mostly a question of your team. On-site works well when people need to compare screens and help each other. Remote works when the team is spread across Zurich, Geneva and Lugano, as long as every participant has a working login and a quiet hour.

What a good AI workshop includes

The best workshops spend most of their time on exercises, not slides. ETH Zurich’s template is a useful model because it states the time boxes. It contains a 35-minute block on generative AI for text and a 35-minute block on data analysis. Then come a 15-minute use-case brainstorm and a 10-minute peer-support closing, with action items and responsibilities agreed between colleagues.

Here is a running order you can adapt to a half-day session.

Ground rules (15 minutes). What the session is for, what data is off limits, and how AI errors look in practice.
Text tasks (45 minutes). Drafting, rewriting and summarizing using your own anonymized material. Participants compare outputs and note what they had to fix.
Data and spreadsheets (45 minutes). Cleaning a table, asking questions of a dataset, producing a chart, then checking the numbers by hand.
Role breakouts (45 minutes). Sales builds a call-prep workflow, finance a variance explanation, HR a job-description draft. Each group writes down the prompt that worked.
Use-case brainstorm (20 minutes). Everyone names tasks that eat their week and rates them by value and risk.
Owners and next steps (15 minutes). Each chosen workflow gets an owner and a date to report back.

The reusable prompt is the main takeaway. If people leave with a saved, tested prompt for a task they do every week, adoption follows. If they leave with a general impression, it does not.

Which tools to practise

Practise in the tools your company actually approves. ETH Zurich’s template names ChatGPT, Gemini and Perplexity AI for text, Gamma.app and Gemini for presentations, Claude.ai and the ETH Leadership Companion for coaching, and ChatGPT Data Analyst, Julius AI and Google Sheets with Looker Studio for data analysis. That is a sample, not a shopping list.

If your company is on Microsoft 365, practise in Copilot. If it is on Google Workspace, practise in Gemini. Teaching a tool nobody is allowed to use at work creates shadow AI, which Microsoft names as a risk alongside hallucinations and over-reliance. Its advice is to pair training with leadership support, clear guardrails, and secure, approved tools.

Data protection and responsible use

The rule belongs at the start of training, not the end. ETH Zurich’s template reminds participants not to enter personal or confidential information into generative AI tools, and to anonymize personal data if it must be used.

Turn that into a one-page internal rule before the workshop:

Green: public information, your own published texts, generic questions.
Amber: internal material that is not personal or client-specific, only in approved company tools.
Red: customer data, personal data, contracts, source code under NDA. Never in a tool unless your IT and legal teams have cleared it.
Who decides in doubt: name a person, not a department.

Swiss companies handling personal data should involve their data protection officer or legal counsel in setting the red list, since your obligations depend on your sector and the tool’s hosting terms. Finish the workshop by asking the team to test the one-pager against three real examples. Gaps show up quickly.

Accuracy needs its own habit. Teach a simple review routine: check every figure against the source, check every claim that sounds specific, and ask who is accountable for the final version. WPMS describes good performance as spotting unsuitable confidential inputs, detecting unsupported claims, and escalating a high-impact output. Those three behaviors make a good checklist.

Make it stick, and measure it

One workshop rarely changes habits alone. What helps is a small follow-up structure.

Pick one representative task per team and time it before training. Record how long it takes, how many corrections it needs and how good the result is.
Repeat after two to four weeks. Compare the same three numbers. Treat any gain as something you measure, not a percentage you promise. This approach is described in aiworkshop.ch’s business workshop outline, and it works with any provider.
Keep a shared prompt library. One document or wiki page, with an owner, where tested prompts live.
Hold a 30-minute review monthly. What worked, what failed, what needs a new rule.

Digital Switzerland’s AI playbook, published in German, describes a similar path. A pilot group is active, feedback loops run and acceptance is measured. Then KPIs are defined, results are compared against a baseline, and a reporting rhythm is set. For a small team, a spreadsheet with three columns is enough.

Watch for the confidence gap too. People who use AI a lot and never check it are a bigger risk than people who avoid it. Peer review of outputs, in pairs, catches this without anyone feeling singled out.

Where our masterclasses fit

We are a small Zurich studio that builds websites, web apps and platforms, and we also run AI masterclasses in person or remotely. Because our team works daily on AI automation and software for business processes, our sessions are designed to start from your actual tasks and tools, not a generic deck. Senior people stay involved from the first conversation to handover, so the person who scopes your session is the person who teaches it.

If, after training, your team finds a repetitive process worth automating, we can also build and run that automation. You can see our services on the pricing and services page, or start from the Ampersand homepage and tell us about your team.

FAQ

How long should AI training for a team be?

For a first session, three to four hours is enough to cover text, data and a role-specific exercise. ETH Zurich’s template, for example, runs 180 to 210 minutes. Add short follow-ups over several weeks so the new habits get used on real work.

How many people should join one workshop?

Keep it to roughly 4 to 12 participants for a hands-on session, as in the ETH Zurich template. A strategy workshop for decision-makers works with 4 to 8 people. Larger companies are better served by repeating the session per department.

Do employees need technical knowledge?

No. Generative AI tools work in plain language, and a workshop should start from what each person already does. What people do need is a login to an approved tool before the session and a clear rule on which data stays out.

What should we never enter into an AI tool?

Personal data, customer data and confidential company information, unless your IT and legal teams have approved the tool for that use. Where personal data must be used, anonymize it first. Write the rule down and name who decides in unclear cases.

How do we know the training worked?

Time one real task before and after, and compare the time needed, the corrections required and the quality of the reviewed result. Add a monthly review of what the team actually uses. Judge success by changed working habits, not by attendance.

Sources

Talk to us

Have a project this touches on?

A free 10-minute call is the fastest way to find out whether we are the right studio for it.

Book a free 10-min call

Keep reading