A one-page policy
- Which tools are approved and with which account
- What data may never go in (personal, confidential, secrets)
- Human review for anything external or affecting people
- Labelling AI content where required
- Whom to ask and how to report mistakes
Training that sticks
A short common basis for everyone (how AI works, checking, data, security), then role-specific practice with real tasks. Keep a record of who completed what — the AI Act recommends it, and it helps onboarding. Repeat yearly or when tools change.
Champions, not only rules
Name one or two AI champions per team who collect good prompts, help colleagues and report problems. Share a monthly “what worked” example. Adoption follows people, not documents.
What to measure
- Time per task before and after (including review)
- Quality: corrections needed, customer feedback, errors found
- Adoption: active users per week
- Incidents: data pasted where it shouldn’t be, wrong outputs sent
Talk about jobs openly
People worry that AI will replace them. Say clearly what you want to achieve (less routine, more time for customers, growth) and involve employees in choosing use cases. Trust decides whether they share what works.