Bottleneck

Managing the Bottleneck: A Working Constraint Playbook

One process step sets the output of the whole line, and an hour lost there is lost forever. This playbook covers finding the constraint, squeezing it before spending on it, and the daily cadence that protects it.

One process step sets the output of your entire line, and every dollar of overhead in the building flows through it. An hour lost at the bottleneck is an hour of output lost for the whole system, unrecoverable, while an hour lost anywhere else usually costs nothing. Price it accordingly: a line generating 3,000 dollars an hour of throughput margin makes bottleneck downtime worth 50 dollars a minute, which reframes a 20 minute changeover as a 1,000 dollar event. Plants that cannot name their constraint are managing 15 or 20 assets with equal energy when exactly one of them is writing the revenue number.

Find the constraint with arithmetic, then confirm with your eyes. Compute good-unit capacity per step: rate times available hours times yield. Say three steps run 80, 72, and 95 units per hour; the 72 unit step is the constraint, and at 20 available hours and 96 percent yield it caps the system at 1,382 good units a day, whatever the other steps can do. The Bottleneck Capacity calculator runs this step-by-step comparison and flags the limiter. Then walk the floor: the constraint is where WIP piles up ahead and the downstream step waits, and if the math and the pile disagree, trust the pile and fix the data.

Exploit the constraint before spending a dollar to elevate it, because most bottlenecks turn out to be only 70 to 85 percent utilized once measured honestly. Standard exploitation moves: run the constraint through breaks and lunch with relief staffing, worth 45 to 60 minutes a shift, about 10 percent more system output by itself. Move inspection upstream so the constraint never wastes a cycle on a part that is already scrap. Offload any work the constraint does that another machine could do, even at worse efficiency. Cut constraint changeovers with SMED; a 40 minute changeover cut to 20, three times a day, buys back an hour of system output daily.

Subordinate everything else, then elevate. Keep a time buffer of 2 to 4 hours of WIP ahead of the constraint so upstream stoppages shorter than the buffer cost nothing, and schedule upstream steps to hold that buffer between its minimum and maximum rather than to chase their own efficiency numbers. Release material at the constraint's pace, or WIP grows without adding one unit of output. Only when the constraint runs above 90 percent utilized with exploitation exhausted does capital make sense, and then the math is easy: if 8 more units an hour is 240 dollars an hour of margin, a 150,000 dollar machine pays back in about 4 months of two-shift running.

The failure modes are classic. Improving non-constraints because they are easy, which produces reports, WIP, and zero output. Letting the constraint starve; plants that finally measure it typically find 5 to 15 percent of constraint hours lost to upstream starvation. Missing a wandering bottleneck: after you elevate step two, step one becomes the constraint, and the old buffer and staffing rules now sit in the wrong place, so re-run the capacity math after every significant change. And the quiet one, a policy constraint masquerading as a machine constraint: a second shift never staffed, a changeover window fixed by habit, a batch size set in 2015.

Run the constraint on a protective cadence. Hourly, the constraint has its own board, plan versus actual, and any lost hour gets a reason code within the hour. Daily, the tier meeting starts with the constraint: buffer status, yesterday's lost minutes, today's risk, one countermeasure. Weekly, review constraint utilization and the starvation and blockage log, and re-check the step capacity table if mix shifted, since a mix change of 20 percent can move the bottleneck between steps. Monthly, review elevation options against the throughput-dollar math and re-confirm which step is the constraint before approving capacity spending anywhere else.

World-class constraint management is visible in the numbers: the bottleneck runs above 90 percent of available time, starvation under 2 percent, changeovers on the constraint at half the plant average, and every capital request answered first with which constraint it relieves. Plants operating this way typically lift system output 15 to 25 percent in the first year without new equipment, purely from exploitation and subordination. The cultural marker is that everyone from the value stream manager to the forklift driver can name the constraint, because the whole building is organized to keep one machine full, fed, and running.

Published 2026-07-02.