Guides / GUIDE-008

Change management and acceptance: communicating with staff, AI champions, pilot groups, handling fears

Bring employees, team leads and the works council into an AI project early so that people actually use the system, using clear communication, a network of AI champions, a representative pilot group and an honest way of handling fears.

Most AI projects in SMBs fail because nobody changes how they work, not because the technology breaks. Licences go unused, people keep doing things the old way, or they quietly use private tools. This guide shows you how to involve staff from the first idea onwards, how to set up AI champions and a pilot group, and how to deal with fears openly. At the end you will have a plan for communication and participation that you can run with the people you already have.

Why it matters

  • Fears are real and specific. In a representative Bitkom survey, 77% of people in Germany named possible job losses as a disadvantage of AI at work. 71% said it was unclear who is responsible for errors, and 63% said it was unclear where the data goes [3]. These are the questions your staff will ask. If you have no answers, rumours will supply them.
  • Shadow AI fills the gap. In 2025, 8% of German companies with 20 or more employees reported widespread private AI use by staff, up from 4% in 2024, and a further 17% reported individual cases. Only 26% provided access to generative AI and only 23% had rules on its use [4]. If people are not involved and given a good tool, many will use their own.
  • Training is rare. In 2025 only 20% of employed people said their employer had trained them on AI, and 70% said their employer offered no AI training at all [5]. Since 27 July 2026, Art. 4 of the AI Act requires deployers to take measures to support the AI literacy of their staff (see GUIDE-006). It does not guarantee any particular level, but doing nothing is not an option.
  • Involvement improves acceptance. In the Weizenbaum Institute's 2025 "KI im Betrieb" survey of 440 companies, 53% of works councils were involved by management when AI was introduced. Companies that involved them more often reported higher acceptance and less often reported work intensification [2]. The Plattform Lernende Systeme also names early participation of employees and their representatives as a key factor for acceptance [1].

When you need it

  • At the idea stage, when you collect and prioritise use cases (GUIDE-001). The people who do the work know where the tedious steps are: copying data between systems, sorting emails, cleaning up spreadsheets, preparing documents. These small automations and the data preparation behind them are often the best first projects. Only staff can point them out.
  • Before the pilot (GUIDE-009). Choose the pilot group, set the communication rhythm and inform the works council in good time.
  • Before rollout, when the system reaches people who were not in the pilot.
  • When a vendor switches on AI features in software you already use, such as M365, your ERP or your ticketing system. For staff this is a change even if you did not sign a new contract.
  • When something goes wrong, such as a visible error or a data incident.

Who is involved

Role Part in change management
Management / owner States the goal and the employment position in its own words, provides time and budget, takes the decisions
Department lead (process owner) Defines what changes in the process, frees up time for the pilot, reports problems
End users / pilot group Test the system in real work, give structured feedback, help redesign the process
AI champions Peers in each team who help colleagues, collect questions and pass feedback to the project
Works council (if elected) Information and co-determination rights; often a strong ally for acceptance (GUIDE-004)
HR Training, role changes, job descriptions, links to the qualification plan (GUIDE-006)
IT Accounts, access, support channel, monitoring of usage (in line with the works agreement)
DPO Answers the question "what happens with my data?" in plain language (GUIDE-005)

In the public sector the staff council (Personalrat) takes the works council's role under the BPersVG or the state staff representation laws.

Step by step

1. Write a one-page "why and what for" (half a day, management + project lead). Describe in plain language the problem you are solving, what the system will and will not do, which data it uses, and what it means for jobs. Be honest. If you do not plan redundancies, say so in writing. If you do not know yet, say that too, and say when you will know. The Plattform Lernende Systeme recommends agreeing the goal and purpose with employees and their representatives before introduction and explaining how the system works [1].

2. Inform and involve the works council early (if you have one). Under §90(1) no. 3 BetrVG you must inform the works council in good time when you plan work processes involving AI. Most AI tools that log user activity are technical systems suited to monitoring behaviour or performance. The Federal Labour Court treats objective suitability as enough, so §87(1) no. 6 BetrVG co-determination usually applies (for example to M365 Copilot or call analytics). Under §80(3) BetrVG, engaging an AI expert counts as necessary. Treat the works council as a design partner, not as an obstacle to clear at the end. Details are in GUIDE-004.

3. Map stakeholders and their concerns (2 hours, project team). List every affected group (users, leads, data owners, customers seeing AI output, IT) with what they gain, fear and need to know.

4. Recruit AI champions (1 per 10 to 20 users, about 2 hours a week each during rollout). Pick respected practitioners, not only the most technical people. Include at least one sceptic. Give champions early access, extra training, a direct line to the project team and visible recognition, for example in their objectives or as part of their working time. Champions answer everyday questions, show colleagues good examples and report what does not work.

5. Set up a representative pilot group (8 to 20 people, 6 to 12 weeks). Mix experienced and new staff, enthusiasts and critics, and include different ages and shift patterns. Agree up front what the pilot measures (GUIDE-009), and that individual performance data from the pilot will not be used for evaluating employees. Pilot and trial phases are the main tool for gathering experience before a company-wide rollout [1].

6. Run a communication rhythm, not a single announcement. - Kick-off with management present: goal, timeline, what stays the same, who to ask. - Short monthly update: what we learned, what we changed, what comes next. - An open question channel, such as a Teams or Slack channel or a mailbox, plus a published FAQ that you keep updated. - Show-and-tell sessions where pilot users show real examples, including failures. In production, use the shift briefing, not email.

7. Handle fears directly. Turn the Bitkom concerns into FAQ answers [3]: - Will my job disappear? State the company's position and what happens to the freed-up time. - Who is responsible for errors? The human who uses and approves the output, supported by clear review steps. Say which steps need a human check. - Where does my data go? Name the provider, the hosting region and the data processing agreement, and say whether any usage data about individuals is analysed. - Will I be monitored? Explain what is logged, who can see it, and what the works agreement allows. Hold small-group sessions where people can raise worries without their manager in the room. Then report back on what you changed because of them.

8. Train in context (see GUIDE-006). Use the employee's own tasks, not generic slides. Make sure everyone gets basic training on limits, errors, confidential data and your AI policy (GUIDE-007). Champions get deeper sessions.

9. Redesign the work, not just the tool. Update process descriptions, approval steps and job profiles. Agree how freed-up time is used, and check whether the workload is going up [1][2]. Recognise the often invisible work of data preparation, such as cleaning master data and tagging documents (GUIDE-003). Give it time in people's schedules.

10. Measure adoption and act on feedback (monthly). Track active users, the use cases people actually use, ticket volume, the most common questions and a short pulse survey (3 to 5 questions). Report adoption at team level, not individual level, unless the works agreement allows otherwise. Close the loop: "You said, we did".

Checklist

  • [ ] One-page goal and purpose statement, including the employment position, published
  • [ ] Works council informed under §90 BetrVG and co-determination clarified (§87(1) no. 6)
  • [ ] Stakeholder map with concerns and messages per group
  • [ ] AI champions named, with time allowance and training
  • [ ] Representative pilot group, including sceptics, with agreed rules on pilot data
  • [ ] Communication plan: kick-off, monthly updates, question channel, FAQ
  • [ ] FAQ answers on jobs, liability, data, monitoring
  • [ ] Task-based training scheduled; AI literacy measures documented (Art. 4 AI Act)
  • [ ] Process descriptions and roles updated
  • [ ] Adoption metrics and pulse survey at team level; feedback loop published

Templates & tools

  • Plattform Lernende Systeme: "Einführung von KI-Systemen in Unternehmen", a whitepaper with a four-phase requirements catalogue (goal setting and impact assessment, planning and design, preparation and implementation, evaluation and adjustment) [1].
  • Weizenbaum Discussion Paper #53, on how management and works councils shape AI introduction, useful as an argument for early involvement [2].
  • Mittelstand-Digital: free webinars and workshops for SMEs, and the free online course "KI-Kompetenz für ihr KMU" (Mittelstand-Digital with appliedAI), listed by the IHK [6]. The current centres are funded until the end of 2026, so check with your regional centre what is still available.
  • Model works agreements on AI, published by employer associations and unions, as a starting point for negotiations (see GUIDE-004).

Common pitfalls

  • Announcing instead of involving. A memo saying "from Monday we use Copilot" produces resistance and shadow AI. Involve people while there is still something to decide.
  • Only enthusiasts in the pilot. The results look great, then the rollout stalls. Include sceptics on purpose.
  • Champions without time. Champion work done "on top" stops after four weeks. Make it part of working time.
  • Silence about jobs. If management says nothing, staff assume the worst. State the position, even if it is "no redundancies planned; we will review in 12 months".
  • Using pilot data for performance reviews. This destroys trust and can breach the works agreement. Agree the rule beforehand and keep it.
  • Ignoring the small stuff. Staff often get the most relief from small automations and better-prepared data, not from the flagship project. Collect these ideas and deliver some quickly.

Sources

  1. Plattform Lernende Systeme, AG Arbeit/Qualifikation, Mensch-Maschine-Interaktion: "Einführung von KI-Systemen in Unternehmen – Gestaltungsansätze für das Change-Management", Kurzfassung. https://www.plattform-lernende-systeme.de/files/Downloads/Publikationen/Zusammenfassungen/AG2_Change_Management_Paper_Kurzfassung.pdf (full paper: https://www.plattform-lernende-systeme.de/files/Downloads/Publikationen/AG2_Whitepaper_Change_Management.pdf). Accessed 2026-10.
  2. Weizenbaum-Institut / idw, press release of 24 March 2026 on Discussion Paper #53 "Zunehmender KI-Einsatz in der Arbeitswelt" (survey "KI im Betrieb" 2025, 440 companies). https://idw-online.de/de/news868100. Accessed 2026-10.
  3. Bitkom, press release "KI-Einsatz bei der Arbeit: Beschäftigte sind geteilter Meinung", 4 April 2024 (1,004 respondents, 511 employed). https://www.bitkom.org/Presse/Presseinformation/KI-Einsatz-bei-der-Arbeit. Accessed 2026-10.
  4. Bitkom, press release "Beschäftigte nutzen vermehrt Schatten-KI", 21 October 2025 (604 companies with 20 or more employees). https://www.bitkom.org/Presse/Presseinformation/Beschaeftigte-nutzen-vermehrt-Schatten-KI. Accessed 2026-10.
  5. Bitkom, press release "Ein Fünftel im Job zu KI geschult", 7 July 2025 (1,005 respondents, 513 employed). https://www.bitkom.org/Presse/Presseinformation/Ein-Fuenftel-im-Job-zu-KI-geschult. Accessed 2026-10.
  6. IHK Lüneburg-Wolfsburg: "Kostenfreie Unterstützung bei der Einführung von KI". https://www.ihk.de/ihklw/produkte/beratung-und-service/ecommerce-und-internet/kuenstliche-intelligenz/kostenfreie-unterstuetzung-bei-der-einfuehrung-von-ki-6396292. Accessed 2026-10.
  7. Betriebsverfassungsgesetz (BetrVG), §§ 80, 87, 90. https://www.gesetze-im-internet.de/betrvg/. Accessed 2026-10.