Change management for introducing AI: roles, phases and measuring success

How to address your staff’s concerns is covered in the guide Building acceptance of AI. This guide is about organising the change.

What is different about AI compared with other IT rollouts?

• Many tasks instead of one process: A new specialist application changes one workflow. AI changes dozens of small tasks in every area, from the doctor’s letter to the citizen’s enquiry.
• Results are drafts: Quality depends on the task, and a person always checks. That calls for new habits.
• Use is hard to order: Whether someone uses AI sensibly is decided at the workplace, not in the project plan.
• Rules on data are central: Every person needs to know which data may go into which app.
• The technology keeps evolving: New models and features arrive all the time. The introduction does not end at go-live.

Which roles does the project need?

Role	Task
Management (managing board, head of administration, executive board)	Mandate and goals, time and budget, visible support
Coordination	Plan, deadlines, collecting feedback, preparing decisions
IT and administration	Sign-in, permissions and groups, sharing, connecting systems
Data protection and information security	Data protection impact assessment, rules on data, review of new apps
Works council, staff council or employee representatives in church-run organisations (MAV)	Co-determination, agreement, the voice of the staff
Contact persons in the departments	Answering questions in the team, reporting needs, suggesting apps
People who build apps	Recurring tasks as apps, tested and shared with groups

In small organisations one person takes on several roles. What matters is that every task has a name attached to it.

In which phases does the introduction run?

1. Prepare: Define goals and first tasks, write rules on data, inform the staff representatives early (§ 90 BetrVG, the German Works Constitution Act), start the data protection impact assessment, select a provider.
2. Run a time-limited pilot: Two or three apps, a small group from different areas, an agreement for the test phase. Collect feedback every week.
3. Roll out step by step: Area by area, each with training shortly before access is switched on and a contact person in the team. Take over apps from the pilot, add new ones.
4. Embed: A fixed procedure for new apps, regular office hours, onboarding of new staff, review of the agreement after one year.

Between the phases, management decides on the basis of the feedback whether to continue and what needs to change.

How do you measure success without monitoring individuals?

Measure tasks and teams, not individuals:

• Tasks: For which tasks is there an app, and how much time does it save in the teams’ own estimate?
• Feedback: Short, anonymous surveys after the pilot and after a few months: Does it help, what is missing, what gets in the way?
• Suggestions: How many new apps do the departments suggest themselves?
• Quality: Do the contact persons report that results are being checked? Are there errors caused by incorrect use?

Usage statistics per person are not a suitable measure. Collecting them amounts to monitoring performance and behaviour and requires an arrangement with the staff representatives, see AI and monitoring of performance or behaviour.

Why do AI introductions fail?

• No mandate: The project runs on the side, nobody has time.
• Started too big: Everything for everyone at once, with no pilot and no feedback.
• Thought too small: A pilot with no plan for what comes next.
• Staff representatives involved too late: Negotiations only start once the contract has been signed.
• No rules on data: Staff are unsure and either do not use the AI at all or turn to private services.
• The introduction ends at go-live: No office hours, no new apps, no onboarding of new staff.

How does this work with basebox?

• Fast technical start: In the cloud you register in two minutes and start working straight away. As things stand today, the server from basebox reaches you three weeks after ordering and is installed in one day.
• Start small: Your administrator shares apps, connections to systems and AI models with groups from your directory service. Ready-made apps can be taken over from the App Store.
• From November 2026: In the App Builder you describe a task, and AI agents turn it into an app, with only the permissions of the person who starts it.
• Support: Training on your use cases or support over six months with office hours; schedule, services and prices are in the guide Onboarding and support.
• Partners: evoila and Systhemis for implementation, consulting and operation, Consileon for consulting on the introduction, strategy and training (Partners).
• For staff representatives: A model agreement as a non-binding draft and documents on logging and permissions (/sicherheit).