Getting more from the bottleneck

Improve has two steps, both from Goldratt's Theory of Constraints. The first is to get more out of the bottleneck with what you already have. The redesign in lesson 1 was the biggest part of it. This lesson covers the rest, and how you'll know whether any of it worked.

An hour lost there is lost everywhere

Module 1 showed that an hour saved anywhere but the bottleneck is a mirage. The other side of that idea is just as important. An hour lost at the bottleneck is an hour lost for the whole organisation, because nothing downstream can make it up and nothing upstream can get round it.

If the estimator spends twenty minutes chasing a missing specification, no amount of speed in sales gets that quote out sooner. That makes the bottleneck's time the most valuable time you have, even though it's rarely the best paid. Most organisations treat it like everyone else's: they fill it with meetings, interruptions and work that came in half-finished.

Four ways to get more from it

Each one is cheap, and none of them asks the people at the bottleneck to work harder.

1. Never let it idle. Keep a small buffer of ready work in front of it, so it never waits for the next item. Cover its breaks and absences. Keep its core hours clear of meetings that aren't its work. A day of work waiting is usually enough; much more than that is a pile, which is the next lesson.
2. Only good work reaches it. Check quality before the bottleneck, not at it. An incomplete request caught upstream costs a few minutes of someone else's time. Caught at the bottleneck, it costs the hours you can least afford. Write down what “complete” means for work arriving there, and send anything short of it back. With AI speeding up the steps before it, this matters more, because unchecked AI output lands here. The output standard you set in module 2 is the rule.
3. Take everything else off its plate. That's your redesign canvas: everything eliminated, simplified or moved to people with spare capacity.
4. Aim AI at the bottleneck step itself. This is the one place AI's speed is never a mirage. An approved tool on the bottleneck's own work, with the person there checking what it produces, adds output to the whole organisation.

Estimate what each move gives back below. Moves three and four come from your redesign canvas. With your prediction from module 4, the builder turns the hours into output and shows how much of your predicted rise the plan covers.

How you'll know: a check every two weeks

During the first loop, the bottleneck lead checks the handoffs into and out of the bottleneck every two weeks. After the first loop, it's monthly. The check is short and always looks at the same things:

• output at the bottleneck, in the aim's unit
• how long work takes to get through it
• rework at the bottleneck and at the step after it
• whether work arrives complete, and whether the queue is under its cap
• whether something else is now what everything waits on

Expect the early dip you wrote into your prediction. Changing how work is done slows it down before it speeds it up, so don't read the first checks as failure until the measurement window has closed.

Improve's three exits

When the window closes, the bottleneck lead re-measures against the prediction, and Improve ends one of three ways. This is what makes Improve a move with an end, rather than a programme that drifts.