In many factories, the plan for tomorrow is made the same way every afternoon: a spreadsheet, a few phone calls, a holiday list and the experience of one or two people who know who can run which machine. It works — until someone calls in sick, a rush order comes in or the person who knows how it all fits together goes on holiday.
AI can take a large part of this work off planners' hands. Not by replacing them, but by doing the puzzle and leaving them the decisions. This article shows what that looks like in practice, which data it needs and how to start small.
Why manual planning reaches its limits
A good plan has to satisfy many conditions at the same time. Each one is easy on its own; together they quickly become more than anyone can keep in mind:
- Qualifications: Not everyone may run every machine. Training, certifications and experience differ from person to person.
- Availability: Holidays, part-time work, sick leave, shift preferences and rest periods between shifts.
- Capacities: Machines and lines have different output, planned maintenance and unplanned downtime.
- Orders: Delivery dates, priorities and changeover times decide what has to run when.
The result is plans that work but are rarely good: overtime to close gaps, idle machines because the right person is missing, and hours of planning time every day.
What AI actually does — and what it doesn't
An AI planning system learns from your historical shift and production data how your operation works, and combines this with today's situation. From it, the optimiser builds a plan that meets the conditions above and, for example, keeps overtime low or meets delivery dates reliably.
What it does not do is decide on its own. The planner gets a proposal, sees where conflicts would arise, adjusts what only a human can know — a promise to an employee, an informal agreement — and approves the plan. When someone drops out at short notice, the system proposes a new plan within moments instead of starting from scratch.
The key point: the plan is only as good as the acceptance of the people who work with it. Proposals that can be understood and changed are worth more than a supposedly perfect black box.
Which data you need
You don't need a data lake to start. Four things are enough, and most companies already have them somewhere:
- Qualification matrix: Who is qualified for which machine or workstation — often an existing spreadsheet.
- Shift model and availability: Shift times, working-time models, holidays and absences, for example from the time-recording system.
- Machines and lines: Capacities, planned maintenance and, if available, the current machine status.
- Orders and delivery dates: From the ERP or production planning system.
In practice, the qualification matrix is usually the biggest piece of work: it is often incomplete or out of date. Tidying it up pays off twice, because it also shows where training is needed.
In practice: an automotive supplier with three lines
A mid-sized automotive supplier spent several hours every day on manual shift planning for three production lines. Holiday peaks, qualification profiles and machine capacities all had to be coordinated by hand.
RALUS built a planning system that combines historical shift data, qualification profiles and machine status. Every day, the model proposes an optimised shift plan, and the planner checks it in minutes instead of hours. After a six-week rollout, the system was fully part of daily operations: planning effort fell by 50%, and overtime caused by poor staffing by 31%.
How to start without a big-bang project
- 1Choose one area: Start with one line or one department where planning is particularly time-consuming.
- 2Collect the data: Bring together the qualification matrix, shift model, capacities and orders for this area.
- 3Plan in parallel: For two to four weeks, the system proposes plans alongside the manual planning. Planners compare and give feedback.
- 4Measure: Track planning time, overtime and shift coverage before and after — then decide on the next areas.
Typical pitfalls
- Outdated qualifications: If the matrix is wrong, the plan is wrong. Agree who keeps it up to date.
- Rules nobody wrote down: Many rules exist only in planners' heads. Collect them in the parallel phase.
- Leaving people out: Involve planners, team leaders and employee representatives early. A plan that people trust gets used.
Conclusion
AI does not make production planning disappear, but it changes it: from a daily puzzle to reviewing and deciding. Companies that start with one area, clean data and their planners on board see results within weeks.
Want to see what this looks like with your data? Request a demo of the production planner.