Ask a plant manager where the bottleneck is and you’ll get a confident answer. Sometimes it’s right. Often it names the work centre that’s most visibly busy, which is not the same thing — and improving the wrong station is one of the more expensive mistakes in manufacturing, because it feels like progress while changing nothing.
Why The Obvious Answer Is Often Wrong
The station everyone points at is usually the one with visible queues and stressed operators. But a queue can form because of upstream batching, not because that station lacks capacity. Meanwhile the real constraint might be a machine that looks calm because it’s starved half the time — the plant can’t feed it fast enough, so it never appears busy.
Constraints also shift with product mix, shift pattern, and order profile. The bottleneck in March may not be the bottleneck in September. A fixed belief about where the constraint lives becomes wrong quietly, and stays wrong until someone re-examines it.
What The Data Reveals
Given job-level timestamps — when work arrived at a station, when it started, when it finished — several signatures identify a true constraint.
Queue growth. Work in progress accumulating in front of a station faster than it drains, sustained over time rather than in bursts.
Starvation downstream. Stations after the constraint sitting idle waiting for input, which is the clearest signal of all and the one most often unrecorded.
Throughput correlation. When this station’s output rises, does total plant output rise with it? If yes, it’s a constraint. If plant output stays flat, you’ve found a station that’s busy but not limiting — and any investment there is wasted.
That last test is the one that separates real analysis from intuition, and it’s difficult to run by eye across months of data.
Constraints That Aren’t Machines
Plenty of real constraints have no equipment attached. A single certified operator for a critical process. Quality inspection sign-off. A tool or fixture shared across lines. Material availability for one component. A model looking at wait-time patterns finds these because it isn’t assuming the answer is a machine — which human analysis frequently does.
What to Do Once You’ve Found it
Before spending on capacity, exhaust the free options. Ensure the constraint never runs dry — buffer material in front of it deliberately. Move quality checks upstream so it never processes parts that will later be scrapped. Take changeovers off it where possible. Run it through breaks and lunch by staggering operator schedules. These typically recover a significant share of constraint time at no capital cost.
Then re-run the analysis. Relieving one constraint always creates another somewhere, and knowing where it moved is how you keep improving instead of guessing. Plants that treat bottleneck identification as a continuous measurement rather than an annual exercise tend to find a steady sequence of small, cheap improvements — which compounds better than one large capital project aimed at whichever station looked busiest the day the decision was made.
ticktick.ai identifies constraints from live job flow data and re-evaluates as mix and volume change.
