A bottleneck sets the pace for everything upstream of it
💎CORE ARGUMENT
Every flow of work has one step that is slowest, and the whole flow moves at that step's speed. Work done faster anywhere before it does not arrive sooner. It arrives at the same time and waits longer. Speeding up a step that is not the bottleneck produces inventory, not output.
I learned this by getting it wrong in public. In the run-up to the Exceptions Inbox launch the slow step was QA, and its queue grew every week. I did not look at the queue. I looked at design, which was quick, pleasant to work with and always had something ready to show, and I kept feeding it. We approved more screens and wrote more stories. All of it joined the pile in front of testing, and each addition made the pile harder to reason about. From where I sat the team had never looked more productive.
The useful question turned out to be embarrassingly physical: where does the work actually wait? Not who is busiest, since everyone upstream of a bottleneck is busy, but where things sit untouched. Once that was visible the rest followed in order. Stop adding to the pile. Make sure the slow step never does anything another step could do. Only then think about giving it more capacity.
An hour lost at the bottleneck is lost for the whole team. An hour saved anywhere else is a rounding error. This changed what I count as help: often the most useful thing a fast step can do is slow down on purpose, or lend its hands.
Where I might be wrong. The bottleneck moves. Relieve testing and the constraint turns up in review, or in my own approvals. A note that says QA is the bottleneck is true for a month.
I may also be flattering myself. It is possible the real constraint in February was the date, and the queue in front of testing was only where the date became visible.
🧱BUILT FROM
🌍WHY IT MATTERS
- Stand-up starts at the queue in front of testing, not with a round of the room.
- I do not start design work the slow step cannot receive within two weeks.
- Busy is no longer evidence of anything.
Tested on: the launch, February to March.
☑️NEXT STEPS
- Put the testing queue at the top of the board
- Stop approving screens until the queue is under a week
- Find the constraint of the merged team before Q3 planning
A factory novel in which a plant is found to be a queue, and an hour lost at the constraint is an hour lost for the whole plant.
The queue in front of the constraint is the stock in 💎 A backlog is a stock and only the outflow shrinks it. The busy upstream steps are the subject of 💎 Local efficiency can lower the output of the whole.
Throughput of a chain equals the throughput of its slowest link. Upstream steps that outpace it only add to the queue before it; downstream steps are starved whatever their capacity. Improvement anywhere but the constraint changes local numbers and not the total.
A single fixed bottleneck is a factory idea. In product work the constraint is often a decision nobody will take, and hunting for a slow step can be a way of not noticing a slow person with authority.
An issue written on the train back from Rotterdam, the way the launch project was run after the slip, and the five steps as a routine.