AI promises to 'optimise' shifts, but the real value depends on data quality and governance. If the input is bad (incomplete records, incidents without reasons), AI amplifies the chaos. And if the model is opaque, the team will reject it. The key is using AI as explainable support, not as a judge.
1) Use cases that genuinely add value
Peak prediction, reinforcement suggestions, absenteeism pattern detection, insufficient rest alerts, and incident prioritisation. These cases help make faster decisions without replacing human judgement.
Example: the system detects that every Friday there are closing-time extensions and suggests an overlap. The manager decides whether to apply it and communicates the change.
2) Risk 1: bias from incomplete data
If one group records worse (for example, more forgotten clock-ins due to lack of a kiosk), a model may interpret this as 'poorer performance' and penalise them with worse shifts. That is not AI; it is automated injustice.
The solution is to improve the data first: accessible clock-in method, traceable correction flows, and consistent rules.
3) Risk 2: opaque decisions ('black box')
If people do not understand why they were assigned a shift, they perceive arbitrariness. Any recommendation must be explainable: coverage, skills, rest periods, fairness, and preferences.
Example: 'X was assigned because they cover forklift duties and respect the rest period' is explainable. 'The system decided it' destroys trust.
4) Governance: human oversight and the right to review
Define who supervises, how corrections are made, and how they are documented. Automation needs a human review channel, and corrections must be recorded.
Example: if a recommendation fails due to incorrect data, the manager corrects the data with a reason, and the system learns. Without that governance, the error repeats.
5) Win-win: less repetitive work, more fairness
For the company, AI can reduce planning time and improve coverage. For the worker, it can improve fairness if used with clear rules and transparency.
The win-win emerges when AI does not impose: it suggests, explains, and allows review. Trust is the condition for success.