We describe a real world case study that involves the monthly planning and scheduling of the sand-casting department in a metal foundry. The problem can be characterised as a single-level multi-item capacitated lot-sizing model with a variety of additional process-specific constraints. The main objective is to smooth production. We present a hierarchical approach, in which we use a combination of mixed integer linear programming, shortest path algorithms, and iterative local improvement. The quality of the production schedules obtained in this way is by far superior to the quality of the schedules constructed by a very expert production planner with no other tool than a plan board. Furthermore, the planning effort is significantly reduced: the manual method requires about 2–3 days, whereas a typical planning session with a prototype decision support system takes no more than half an hour.