The Bottleneck That Would Not Stay Still
A New Stage, and a Schedule Nobody Could Write
A quality initiative required inserting a new production stage into an existing Make-Store-Pack coffee operation. The stage itself was not the problem. What broke was scheduling: as packaging sizes changed, the bottleneck moved — and an Advanced Planning System cannot schedule against a constraint that will not hold still.
The change
Make-Store-Pack is a common structure in food and beverage: a making process runs more or less continuously, product is held in intermediate storage, and packing lines draw from that store into finished formats. It works because the store decouples two operations that would otherwise have to run in lockstep.
The quality initiative added a stage to that chain. Every stage inserted into a coupled system changes where material accumulates and where it waits, and this one changed it enough that building a robust schedule with the plant’s APS became extremely difficult.
Why the schedule broke
The system exhibited moving bottlenecks. Change the packaging size and the constraint relocated — not because anything failed, but because a different format draws from the store at a different rate and shifts which stage is binding.
An APS optimises against a model of the plant. If that model assumes a fixed constraint, its schedule is only as good as the assumption, and it degrades quietly whenever the mix moves. The plant was not getting bad schedules because the planners were careless. It was getting bad schedules because the premise underneath them was no longer true.
Why rate-based modelling was the right tool
This is the problem that motivated discrete-rate simulation in the first place. Traditional discrete-event simulation models entities moving through queues, and it could not effectively represent the continuous-flow nature of the shifty-bottleneck behaviour here. Coffee moving through making and into a store is a rate, not a queue of items.
The team built a Make-Store-Pack simulation using discrete-rate simulation — the innovation that came out of the Extend+Industry work and later became part of ExtendSim’s core libraries, and the same approach ReliaSim packages today.
What came out of it
With a model that reproduced the moving bottleneck rather than assuming it away, the team could conceive, test and verify an entirely new scheduling algorithm — and confirm it stayed robust across a range of demand scenarios rather than working only for the mix that happened to be current.
The plant adapted its operations to the new production constraints, with a scheduling approach that had been validated against variation before it was deployed.
On the evidence
This account comes from our own project record rather than a published paper, and it is reported qualitatively — we have not put a throughput or cost figure on it, because the source does not carry one. The mechanism it describes, moving bottlenecks under changing product mix, is reproducible in the sandbox on the demo line.
Related: The Hidden Bottleneck · How big should a buffer be? · All case studies