AI training for employees: what the team should be able to do
What AI training for employees should achieve
AI training for employees is finished when every participant can do six things at their own desk: name the rules and approved tools, choose the right tool for a task, give the tool enough context, check the output, stop when a task is unsuitable, and report a problem. Tool demos and prompt lists alone are not enough. If someone cannot say on Thursday whether the customer email in front of them may go into the company chat, something is still missing.
This is for managing directors and team leads in a Swiss SME who are about to book or plan training for 5 to 80 people. The policy and allow-list should exist first: see the AI usage policy template and how to approve AI tools. Training teaches the rules. It does not replace leadership decisions.
- Knows the rules, the approved tools and where the policy is
- Chooses the right tool for a task, including no tool at all
- Gives context: purpose, reader, material, format
- Checks facts, figures, names and tone before anything is used
- Stops when a task is unsuitable and does it another way
- Reports a problem the same day to a named person
The six skills, each with a test in the room
Each skill needs a test you can run in the room. Without one you are measuring attendance.
Rules and tools. Hand out three real snippets: a paragraph from your website, an internal checklist, a customer email with name and address. Each person sorts them into the data classes of your policy and names the tool they may use, in under a minute.
The right tool. Give five tasks from the team’s week: translating a supplier email, summarising a long contract, the weekly report built from three exports, a reply to an angry customer, a figure for the annual accounts. People say which tool fits, or whether none does. The weekly report is not a chat task. It repeats every week and should eventually run on its own.
Context. The typical weak request is one line with no material. Have people rewrite one of their own requests: what it is for, who reads the result, which material the tool may use (within the data class), what format, what to avoid.
Checking. Hand out a draft with two planted errors, for example a wrong date and a promise to the customer nobody made. Can they find both? Many trainings skip exactly this skill.
Stopping. Signals to stop: the task needs data that may not go in, the answer would have to be right without checking, the third attempt is still wrong, or the task is a judgement about a person. Give five cases and ask which ones they would stop.
Reporting. Everyone can name the contact person from the policy and knows that a wrong paste reported on the same day is not a disciplinary matter.
Half a day, full day: limits and cost
Half a day covers rules, approved tools and the six skills practised on three or four real tasks. A full day adds time per function and room to chain several steps on the team’s own work.
A realistic half day: 45 minutes on rules and tools, using examples from your own inbox. About two hours in which each person works on a task they brought. The last half hour on stop cases, checking and the incident contact. Sales, office and finance share the same tasks, which keeps the day together but thin for each function.
A full day adds separate tasks per function, longer documents to check and time to connect tools into a small flow. What neither format does: build a connected workflow between your systems, or make everyone an expert. Nobody in the room will have the Monday report automated by the evening.
Our AI in Business workshop follows that split: half a day or a full day, each for up to 20 participants, with a preparation call beforehand, an example library, a recording and written take-aways for the first weeks. On site or hybrid in Switzerland. Current prices are on the offer page. That is the workshop, not the Process Check fee.
What training does not replace
Training gives shared rules and practice. It does not build an interface between finance and the CRM. It does not replace an allow-list. It does not answer which single workflow to tackle first.
If the pain is recurring glue between two systems, the next step is not another training day. The Process Check fits: one task, 30 minutes online, a concrete case or a clear stop. The check is free.
How this differs from implementation and from security training
Workshop or implementation separates a team question from a build question. This article is the team side: same rules, same practice. Security awareness training covers phishing, fraud and reporting paths. It does not replace an AI policy or practice at the approved company chat. You may need both, in a different order: approved tools and two pages of rules first, then the day in the room.
A made-up example: half a day
A made-up example: a wholesale company with around 30 people, all on Microsoft 365. Copilot Chat with work accounts is on the allow-list, and a two-page policy is signed. A week before the training, each participant sends one task from last week that took too long, with names removed.
On the day, the rules are explained using three of those tasks, not slides of the law. Then everyone works on their own task, in the approved tool, with the policy on the table. The last half hour: two stop cases and a planted-error draft. At the end, each person writes down which weekly task they will do differently, and which one they will stop trying in the chat.
The team lead keeps those notes. Four weeks later they are the basis for the check below.
How to see whether the training worked
Measure the work, not the logins. Usage shows that people open the tool. It does not show that the work got faster or better.
adesso GenAI Impact Report 2026 for Switzerland (100 executives): measure GenAI success by process efficiency. 45% by usage rate, 30% name missing qualification as a hurdle. Their figures, not ours.
A small office can do better with little effort:
- Before the training, pick three recurring tasks and note roughly how long they take and how often they need rework.
- Four to six weeks later, look at the same three tasks again.
- Read five outgoing drafts from people who attended. Were they checked?
- Count questions to the contact person and reported incidents. A few reports are a good sign.
Does every SME need AI training?
No. Without an approved tool and a policy, training reinforces the wrong habits. With rules and a list in place, the day makes sense when many people should use the same tools the same way.
Who should not book training yet
Rules first, then the day: the allow-list exists, the policy is signed, and several functions should apply the same rules. Clarify the workflow first: the pain is a recurring report or import, or one person retypes between two systems. In both cases the training day is the wrong first step.
Next step
If the rules and the list exist and the team should apply them the same way, book the AI in Business workshop, half a day or a full day, on your team’s real tasks. If you already have one specific recurring task in mind, the Process Check is the right hour: one task, 30 minutes, a case and a price range as a guide figure.
Written by
Aurum Avis Labs
Builds and ships at Aurum Avis Labs. Writes here about what we learn working with founders and SMEs in the DACH region.
Related Articles
You might also be interested in these articles
Workshop or implementation: which one you actually need
AI workshop or Process Check: when the team needs shared rules, and when a single workflow should be built.
AI usage policy template for SMEs: a page you can copy
AI usage policy template for Swiss SMEs: eight clauses from approved tools to data classes and human review, ready to copy. Not legal advice.
Approve AI tools at work: an allow-list in five steps
How to approve AI tools at work: take stock, move to business accounts, check the Microsoft 365, Google Workspace and ChatGPT terms, announce the list.