Whether AI takes hold in everyday work rarely depends on the technology. It depends on how you involve your team. A tool that nobody uses changes nothing. If everyone uses it according to their own rules, risks arise for data and quality.
In between lie four steps: clarify responsibility, take concerns seriously, start small and write down rules. Responsibility comes first, because every further question needs someone to answer it.
Clarify responsibility
AI produces drafts, people decide. Set out who checks and approves results. Also appoint a contact person for AI in the company.
Under Article 4 of the EU AI Act, any company that uses AI must take measures that support the development of AI literacy in its team. More on this in the article AI literacy under Article 4.
In a company with twenty employees, responsibilities could be divided like this:
- Management: decides which tools are used and puts the rules into effect.
- Contact person for AI: collects questions, checks new tools before they are approved and keeps rules and templates up to date.
- Sign-off on the work: whoever is responsible for a result today remains responsible for it with AI. Whoever currently signs a quote also checks the AI's draft.
The last point saves a lot of discussion. AI changes how a draft is produced. Who answers for the result stays the same.
Take concerns seriously
Many people wonder what AI means for their work. Talk openly about what will change and what will not. Show which tasks AI takes off their hands and what that frees up time for.
Typical questions in the team are: Will I be replaced? Do I have to be able to use it, even if I am not familiar with it? Who is liable if the AI makes a mistake? Will it be used to monitor me? Management should have an honest answer to each of these questions before announcing the topic.
A guide for a first team meeting:
- What we plan to do: which tasks, which tool, from when
- What stays the same: who is responsible and who approves
- What we do not plan to do: for example, measure performance with AI, if that applies
- How everyone learns: training, a contact person, time to try things out
- Open discussion: collect questions and concerns, follow up with answers where there are none yet
Write the questions down. They show where the rules will need to be more precise later. The question of monitoring is also a matter for the works council, if there is one, as described below.
Start small
Choose two or three tasks where you can soon see whether AI helps. Let the team try things out themselves and collect templates that work in everyday use. The article Five tasks to get started with AI describes which tasks are suitable.
A before and after from an administrative office with twelve employees:
- Before: Management announces in an all-staff email that everyone can now use an AI tool. Some try it, others do not. After a few weeks, nobody knows who uses it for what, or whether it achieves anything.
- After: Three volunteers spend a few weeks testing two defined tasks: minutes and replies to frequent enquiries. They save successful prompts in a shared folder and report back in the team meeting on what worked. Management then decides whether to extend the tasks to the whole team.
The second approach takes longer to produce a first result. In return, the templates come from your own daily work, and colleagues see in real cases what AI can do.
Write down the rules
Record which tools are permitted, which data may go into them and how results are checked. One page is often enough. More on this in the article Can my team use ChatGPT?
Wording you can adopt and adapt:
- “For work, we only use [tool] with the company account. Private accounts are not permitted for work content.”
- “We only enter names, contact details, customer numbers and anything that could identify a person into tools covered by a data processing agreement (DPA). If in doubt, we replace them with placeholders.”
- “Confidential documents such as quotes, contracts and internal figures do not belong in freely available tools.”
- “A person checks every result that leaves the company. Responsibility lies with whoever would be responsible for it without AI.”
- “Questions and suggestions for new tools go to [contact person].”
- “These rules apply from [date] and will be reviewed after [period] at the latest.”
Date the rules. That way, you can show later what applied at which point in time.
Involve the works council early
If there is a works council (the body elected by employees to represent them), involve it early. It must be informed in good time about the planned use of AI (§ 90 of the German Works Constitution Act, BetrVG). The Act requires consultation to take place early enough for the works council's suggestions and concerns to be taken into account in the planning.
If a tool is capable of monitoring behaviour or performance, the works council also has a right of co-determination (§ 87 BetrVG). Settle this question before the tool is introduced.
Invite the works council to the first draft of the rules. Its questions then feed into the version that later applies to everyone.
Build skills and keep them up
Article 4 requires measures, not a particular format. We recommend a combination: the basics for everyone, more in-depth training for those who work with AI every day, a contact person and regular refreshers, because the tools change quickly.
Keep a record of what you have done:
- who attended which training and when
- the current rules on AI use, with the date
- which AI tools are in use and who uses them
To keep the knowledge alive in everyday work, a standing item in the team meeting is often enough: What worked, which template is new, where were there problems? New templates go into the shared collection, new questions to the contact person. When new tools, new tasks or new staff come along, plan a refresher.
Next steps
A sequence you can start with over the coming weeks:
- appoint a contact person and decide who approves results
- hold the first team meeting using the guide and write down the questions
- choose two or three tasks and a few volunteers
- write the rules on one page, date them and discuss them with the works council, if there is one
- plan training and record attendance
- after a few weeks, review the results and decide how to proceed
Who in your team would be the right contact person for AI?
Frequently asked questions
Who should be the AI contact person?
Someone curious who has the backing of management. Expert knowledge of AI matters less than a willingness to collect questions and pass them on.
What if someone does not want to take part?
Do not push. Show what colleagues are achieving and make room for questions. A successful example from your own daily work is often what convinces people.
Should we ban AI until rules are in place?
A ban rarely holds. People who use AI anyway then often do so with private accounts, without anyone knowing. A short interim rule is more helpful: which tool, which data must not go in, who checks.
Does everyone in the team need the same level of skill?
No. Article 4 requires measures that support the development of AI literacy. Nobody has to guarantee a particular level. We recommend the basics for everyone and more in-depth training for those who work with AI every day.
How long should an AI policy be?
One page is often enough. It should answer the questions that come up in everyday work: which tool, which data, who checks, who helps.
Sources
- § 90 BetrVG, German Works Constitution Act (in German)
- § 87 BetrVG, German Works Constitution Act (in German)
- Regulation (EU) 2026/1744 (AI Omnibus), EUR-Lex
Updated: . This article is not legal advice.
