Finite Capacity Scheduling: Why Infinite-Capacity Production Plans Always Slip

There is a specific moment familiar to every production planner: the ERP says the order is scheduled for Tuesday, the shop floor says Tuesday is impossible, and both are correct within their own logic. The ERP scheduled backwards from the due date using standard lead times and never asked whether the machine was free. The shop floor is looking at a machine that is already committed. This is not a communication failure. It is what happens when a plan is built on an infinite capacity model.

Infinite capacity scheduling assumes that any amount of work can be performed in any period. It is computationally cheap and produces a plan quickly, which is why it remains the default in most ERP systems. It is also, in a busy factory, systematically optimistic — every plan it produces is achievable only if nothing competes for the same resource, which in a busy factory is never true. Finite capacity scheduling, known in German-speaking manufacturing as Feinplanung, does the harder thing: it plans against resources that have limits.

Those limits are more varied than machine hours. A finite capacity model has to account for machine availability including planned maintenance, operator availability and skills, tooling and fixtures that can only be in one place at a time, material availability from the MRP run, and shift patterns that differ by area. An order that needs a specific die and a certified operator on a specific press is constrained by whichever of those three is scarcest, and a scheduler that only models machine time will schedule it wrong.

Changeover and sequence dependency are where finite scheduling earns most of its return. In many processes the time to switch from one product to the next depends entirely on which two products they are — light colour to dark is quick, dark to light requires a full clean. Sequencing by due date alone can double the total changeover time across a week compared with sequencing that groups compatible products. A finite capacity scheduler with a sequence-dependent setup matrix optimises this automatically; a planner with a whiteboard optimises it approximately, on good days.

The bottleneck deserves special treatment because it sets the throughput of the whole line. Every factory has one constraint at any given time, and time lost at that constraint is time lost for the entire factory, permanently. Finite capacity scheduling makes the constraint explicit rather than incidental — you can see which resource is at ninety-plus percent load, schedule protective buffers ahead of it, and stop wasting effort optimising resources that were never limiting anything.

What finite scheduling produces that infinite scheduling cannot is a reliable answer to when. When a customer asks whether an order can ship on the fifteenth, a finite capacity model can simulate the insertion — where it fits, what it displaces, whether the promise holds. That capability, capable-to-promise, is commercially valuable in a way that is easy to underestimate. Quoting fewer dates and hitting them consistently changes the customer relationship more than quoting aggressively and apologising.

The realism trap is worth naming. A finite capacity schedule is only as good as its constraint data — if your routing says a job takes four hours and it reliably takes six, the scheduler will produce a beautifully optimised plan that is wrong by fifty percent. Factories that succeed with APS almost always tighten their routing and standard time data first, usually using actual production data captured by an MES. The scheduler does not need perfect data, but it needs data that is honest about the direction of its errors.

Rescheduling frequency is a practical decision that shapes how the tool is experienced on the floor. Reschedule too rarely and the plan drifts from reality until people ignore it. Reschedule continuously and the sequence changes under an operator's hands mid-shift, which destroys trust faster than a bad plan. Most working implementations freeze a near-term horizon — the current shift or day — and reoptimise beyond it as actuals arrive.

For a factory currently planning in spreadsheets, the diagnostic question is simple: how often does the plan survive contact with the shift? If the answer is rarely, the problem is unlikely to be planner skill. It is more likely that the plan was never constrained by the things that actually constrain the factory.