Finding your bottleneck: how to read where work is actually piling up

Eli Goldratt's Theory of Constraints is the framework most production thinking traces back to. The core claim is simple and weirdly under-applied: a system's throughput is limited by exactly one constraint at a time. Improvements anywhere except the constraint don't change throughput - they just create more pile-up upstream of the constraint, or starve work downstream of it. The first useful question for any production team isn't "how do we go faster?" - it's "which phase is the constraint?"

The bottleneck shows itself in three places on a healthy production board. One: the phase with the longest queue (cards waiting to enter, plus cards in the phase) is almost always the constraint - work piles up just before the slowest step. Two: the phase whose median time-in-phase is rising over the last 30 days. Constraints don't get better on their own; their durations creep. Three: the phase where blockers cluster. If you tagged blockers honestly during stand-up, the bottleneck phase has more of them than its neighbours.

Goldratt's prescription is the famous five steps: identify, exploit, subordinate, elevate, repeat. The two that matter most for a small production team: exploit (squeeze every minute out of the bottleneck phase - no meetings, no context-switching for the people working it, no pulling them into other phases as a courtesy) and subordinate (other phases must serve the bottleneck's pace, not their own). The instinct of the rest of the team is to keep their own phase humming; the discipline is to slow your phase down on purpose so the bottleneck doesn't drown.

The most common mistake after a successful bottleneck fix is treating the original bottleneck as permanent. It isn't. Improving one constraint reveals the next - the new bottleneck is somewhere else, often a phase that was previously "fine". Most production teams should re-run the three-signal check monthly. {PRODUCT}'s dashboard surfaces all three signals on one screen so the check is a five-minute visual scan, not a weekly meeting.

Two things teams often misdiagnose. One: the busiest phase. Busy ≠ bottleneck; the busiest phase is sometimes simply the broadest scope. The bottleneck is where work waits, not where the most work happens. Two: the slowest individual person. Production thinking is about phases, not people. If a phase has only one person and runs slow, the constraint is the staffing of the phase, not the person. Naming a person as the bottleneck is almost always the wrong move and corrodes the team's relationship with the board.