A shift plan is a puzzle with many rules: availability, contracted hours, qualifications, rest periods, holidays, requests. Planning for 40 people every week takes hours. AI-powered software promises to solve the puzzle in seconds. Often it does. The tricky part is not the calculation but who decides in the end, and who was asked beforehand.
What the survey shows
For a Flash Eurobarometer, the European Commission surveyed about 12,800 owners and managers of companies with 1 to 749 employees in all 27 member states in March and April 2026. 42 % already useU AI or algorithmic management, and another 21 % plan to.
| Area | Share of users |
|---|---|
| Customer service and customer relations | 47 %U |
| Monitoring performance and productivity | 36 %U |
| Logistics, planning and staff scheduling | 30 %U |
Almost everywhere, people still decide: 90 % say humans make the decisions, and only 2 %U report fully automated processes. 49 %U do not trust AI with personnel decisions. The biggest barriers named are investment costs (50 %), legal uncertainty (49 %) and lack of expertise (48 %). The announcement contains no figures for Germany.
The legal situation in Germany in five points
- The rota is co-determined: The start and end of daily working time and its distribution across the week are subject to the works council's co-determination (Section 87(1) No. 2 Works Constitution Act). Software that generates rotas does not change that.
- The software is co-determined: If the system can evaluate behaviour or performance, Section 87(1) No. 6 applies. When planning work procedures that use AI, the employer must inform the works council in good time (Section 90(1) No. 3).
- High-risk depending on function: Systems that allocate tasks based on individual behaviour or personal traits, or that monitor performance and behaviour, are high-risk under Annex III No. 4 of the AI Act. Following the Omnibus Regulation, the obligations apply from 2 December 2027R. Deployers must inform worker representatives and staff beforehand (Art. 26(7)). Planning based only on availability and contracted hours usually falls outside this.
- No purely automated decision: Decisions with significant effects must not be based solely on automated processing (Art. 22 GDPR). A planner approves.
- The Working Hours Act sets the limit: At most ten hours a day and at least eleven hours' rest (Sections 3 and 5). These rules belong in the system as fixed constraints, not in a prompt.
More on the deadlines is in our article on the AI Act after the Omnibus.
The workflow
- 01StartDefine the use: demand forecast, staffing suggestions or both
- 02CheckDoes the system evaluate the behaviour or performance of individuals?yes: high-risk under Annex III, plan for obligations from 2 December 2027
- 03PersonInvolve the works council early, agree a works agreement on data, rules and reports
- 04SystemDefine inputs: availability, contracted hours, qualifications, working-time rules
- 05AISystem drafts the plan with a reason for each assignment
- 06PersonPlanner reviews, adjusts and approves
- 07SystemStaff see the plan in the app; swaps and sick notes flow back into the system
- 08ResultApproved plan, deviations from the suggestion documented
For its own Copilot rollout in Germany, France and the Netherlands, Microsoft describes an approach that transfers to smaller companies: works council members joined the first test wave, their feedback went to the product team, and only then was the rollout extended.H For shift planning this means: a test month in which the works council can see suggestions and approvals builds more trust than any presentation.
What the software can do
- Planday: When you create a shift, it suggests suitable people, ranked by fit and with a reason. It considers availability, contracted hours and working-time rules. Every suggestion can be overridden.H
- Quinyx: Combines a demand forecast with automatic schedule creation.H
- Your own check: Even without specialist software, an approved AI assistant can check a finished plan against your rules. Only use a tool with a data processing agreement and replace names with initials.
Check the following draft shift plan against these rules: at most 10 hours per day, at least 11 hours of rest between two shifts, at most 5 working days in a row, every shift with at least one person holding qualification [qualification]. List every violation with initials, date and the rule broken. Do not change the plan and do not suggest who should work.
Limits and risks
- Hidden evaluation: If sick days, punctuality or customer ratings feed into assignments, planning turns into performance evaluation. Define in the works agreement which data are off limits.
- Fairness: A system that always suggests the most flexible people overloads them. Review monthly how weekend and night shifts are distributed.
- Traceability: Without a reason for each assignment, nobody can check why someone was scheduled. Choose software that shows reasons.
- Where the numbers come from: The Eurobarometer surveys company managers; it does not measure. Product details come from the vendors.
How to start
- Measure for four weeks how long planning takes and how often plans change afterwards.
- Invite the works council before you compare vendors, and agree together which data the system may use.
- For each tool, check whether it evaluates behaviour or performance, and record the result in writing.
- Run a parallel test for a month: the system suggests, the planner decides as before and notes deviations.
- Review time spent, deviations and the distribution of unpopular shifts, then decide on the rollout.
Sources
- UEuropean Commission: Flash Eurobarometer on AI and algorithmic management in SMEs, 14 Oct 2026
- RRegulation (EU) 2024/1689 (AI Act), Annex III and Art. 26
- RRegulation (EU) 2016/679 (GDPR), Art. 22
- RGerman Works Constitution Act, Section 87
- RGerman Works Constitution Act, Section 90
- RGerman Working Hours Act, Sections 3 and 5
- HMicrosoft Inside Track: Rollout with works councils, 5 Feb 2026
- HPlanday: Intelligent shift assignments, 24 Mar 2026
- HQuinyx: AI-powered scheduling