Your employees already use AI. The real question is whether they use it well, and with which data.
In 2024, the Microsoft and LinkedIn Work Trend Index found that 75% of knowledge workers use AI at work, and 78% of those users bring their own AI tools, rising to 80% in small and medium-sized companies. Meanwhile, France's statistics office INSEE reports that only 18% of French companies with 10 or more employees said they used at least one AI technology in 2025. In other words, AI has often come in through the back door, with no shared method and no rules about data.
A good prompt already saves time. But imagine if everyone on your team knew how to connect AI to their tools, rely on an assistant that knows your business, and build their own recipes for the manual tasks they repeat every week. That is what good AI training for employees aims at. Here is how to run it so it actually sticks.
What your teams gain: far more than a good prompt
Most training stops at prompting, meaning how to phrase a request. Useful, but it is only the first level. There are three more, and they are the ones that change the working day.
1. Asking well. Give context, an example and the expected format, then check the answer before using it. A salesperson who does this writes a personalised follow-up in two minutes instead of fifteen.
2. Connecting AI to their tools. Today's assistants can connect to email, shared drives, calendars and the CRM (the software where you track customers and deals). No more copy and paste between five windows: the AI fetches the last exchange with the customer and the current quote, then drafts the reply.
3. An assistant that knows your business. Your offers, procedures, prices and internal rules: a shared assistant that answers from your own documents and shows its sources. New hires find the right answer without interrupting the whole team. We explain the idea in our article on the company brain.
4. Building their own "skills". A skill is a recipe the AI follows every time for a recurring task: drafting a quote from a request, writing meeting minutes, checking a supplier invoice. The recipe holds the steps, the rules and an example of a good result. You write it once and the whole team reuses it. Depending on the tool, it is called a custom GPT in ChatGPT, a Gem in Gemini or a Skill in Claude.
The fourth level is where manual work really shrinks. Your people become AI-augmented employees: they hand formatting, copying and checking to the AI and keep their time for what needs their judgement.
Train by role, not in a lecture hall
A generic "discover ChatGPT" session for 40 people in one room feels good on the day and is forgotten three weeks later. Everyone leaves with examples that have nothing to do with their job.
Training that sticks is done by role, in small groups:
- Leadership: prepare a decision, summarise a file, analyse a table of figures, set the rules for the team.
- Sales: prepare a meeting, write a proposal, follow up, update the CRM without spending the evening on it.
- Admin and finance: read and file documents, check invoices, prepare reporting.
- Marketing: content plans, variations of a message, visuals that respect your brand guidelines.
- Customer service: reply templates, sorting requests, searching the documentation.
Each group works on its own cases. Nobody wastes time, and everyone sees in the first hour what AI changes for them.
Start from this week's real tasks
The simplest and most effective rule: before the training, ask each participant to list three tasks they repeat every week and find tedious. Answering the same customer questions. Copying data from a PDF into a spreadsheet. Preparing the same report every Monday.
During the workshop, you work on those exact tasks, with the company's real tools. At the end, everyone leaves with at least one recipe that works on their own computer, tested on a real case. The next morning, they use it. That jump to immediate use is what makes training last.
Those task lists have another benefit: they show which processes deserve deeper automation. If ten people name the same task, it is a serious candidate for a proper project. Our AI use case workshop exists to rank exactly those by what they are worth.
Data rules: what never gets pasted
This is the point most often forgotten, and it can be the most expensive. When an employee pastes a customer contract into a free AI tool on a personal account, the company loses control of that data.
Good AI training gives simple rules, written on one page:
- Approved tools. A short list of validated tools, used with business accounts. The business plans of the major providers generally commit not to train their models on your data: see for example OpenAI's commitment for its business offers. Check it for every tool.
- What never gets pasted. Personal data about customers or staff (names, contact details, health, salaries), documents under a non-disclosure agreement, passwords and access keys, unpublished financial figures.
- What to do instead. Anonymise (replace names with "Customer A"), summarise the situation yourself, or use the internal assistant connected to your documents with the right access rights.
- Who checks what. Anything the AI produces is reviewed by a person before it goes to a customer, into a contract or into the accounts.
It is also about compliance. The GDPR applies to any personal data you give an AI tool. And Article 4 of the EU AI Act requires companies that use AI systems to take measures to build their staff's AI literacy, an obligation that has applied since 2 February 2025. According to the European Commission, the amendment that entered into force in mid-July 2026 no longer sets a specific level, but the obligation remains. Training your teams is also how you stay compliant. For more, read our article on AI agents, GDPR and the EU AI Act.
Coaching after the training: where it is won or lost
Without follow-up, most participants slip back into old habits within a few weeks. Not out of bad will, but because the first time a recipe fails, nobody is there to help.
What works:
- Short coaching sessions over a few weeks, on each person's real blockers.
- AI champions in each team: one or two motivated people who help colleagues and report needs.
- A shared library of recipes, so one salesperson's good idea benefits the whole sales team.
- A channel for questions, in the messaging tool you already use.
That way, the know-how stays in the company, not in one person's head.
Measure the gain, simply
AI training should be judged on what it brings, not on the satisfaction score at the end of the session. You do not need a complex dashboard:
- Before the training, pick three to five tasks that come back every week and time them on a few examples.
- At 30 days and again at 90 days, measure the same tasks again.
- Track real usage: how many people use AI every week, how many recipes are created and reused.
- Look at quality: less retyping, fewer errors, fewer customer complaints about vague answers.
These figures are yours, measured in your company. They tell you where to extend the approach, and which tasks deserve the next step: full automation.
How Qartmina trains your teams
Qartmina is an AI studio in Nice and an OpenAI Select Partner. Our AI training and consulting offer follows exactly this method: workshops by role on your teams' real tasks, hands-on learning of skills, tool connections and a shared assistant, clear data rules and coaching after the training. We work on site in Nice and on the French Riviera, or remotely anywhere in France.
Frequently asked questions
How long should AI training for employees take? Short and repeated works best: a first workshop per role of half a day to a full day, then short coaching sessions over a few weeks. What makes the difference is not the length of the first session, it is the follow-up on real tasks.
Is AI training mandatory for companies in the EU? Article 4 of the EU AI Act requires companies that use AI systems to take measures to build their staff's AI literacy. It has applied since 2 February 2025. Since the amendment adopted in 2026, no specific level is required, but the obligation to address it remains.
What data should employees never paste into an AI tool? Personal data about customers or staff, documents under a non-disclosure agreement, passwords and access keys, unpublished financial figures, and anything you would not email to a stranger. Everything else goes only into tools the company has approved, with a business account.
How do you measure the gain from AI training? Pick three to five tasks that come back every week, time them before the training, then again at 30 and 90 days. Add real usage (who uses AI every week) and the number of reusable recipes the team actually shares.
Your teams already use AI, each in their own way? Book a free AI consultation and we will work out together which team to start with.

