Flow
Moving to pull does not remove the inventory. It shows you where it was.
Push builds to a forecast and hides the surplus between steps. Pull builds to a signal and makes the same surplus visible. The second is better, and not for the reason usually given.
Pull is usually sold as the system with less inventory, and that is a slightly misleading way to describe it. A well run pull system does typically hold less. But the reason is not that pull is thrifty. It is that push lets inventory accumulate in places nobody is looking, and pull does not have those places.
The difference is visibility before it is quantity.
The two signals
Push releases work based on a plan. A forecast says the month needs nine hundred units, so nine hundred are started, and each step processes what arrives and passes it on as fast as it can.
Pull releases work based on consumption. A step takes something, and that act authorizes the step upstream to make one more. Nothing is made because a plan said so; it is made because something downstream was used.
The consequence is simple and it is the whole thing. Under push, the work released is set by the plan. Under pull, the work in the system is capped by the signals in circulation, and that cap is a number somebody chose.
Where push puts the surplus
Every step in a process has a different capacity, and under push the fast steps run at their own speed rather than at the pace of the line. What they produce has to go somewhere.
| Cutting, capacity for the month | 1,100 | |
| Forming, capacity for the month | 950 | |
| Assembly, capacity for the month | 700 | |
| Finishing, capacity for the month | 1,000 | |
| Shipped, and units left standing between steps | 700 and 200 |
Illustrative figures. Cutting and forming both ran to plan because they could. Assembly could not, so two hundred units are now sitting in front of it. Every step reports having met its own target and the line missed by two hundred.
The two hundred units are not a mistake anybody made. They are the arithmetic of three steps being told to make nine hundred when one of them could make seven hundred. Every step did what it was asked. The surplus is the difference between the plan and the constraint, converted into a pile, and it is sitting in a corner nobody is measured on.
Switch this line to pull and the two hundred do not appear. Assembly consumes at seven hundred a month, so forming is authorized to make seven hundred, so cutting makes seven hundred. The stock does not vanish, it never gets made.
What pull actually costs
This is the part that gets left out of the pitch, and skipping it is why pull implementations get abandoned.
Every step now runs at the pace of the constraint. Cutting is capable of eleven hundred and will produce seven hundred. Its utilization drops from a hundred percent to sixty four, and if anybody is measured on step utilization, the pull system will read as a catastrophe on their scorecard while the line ships exactly what it shipped before with two hundred fewer units of stock. That measurement conflict kills more pull rollouts than any technical problem.
Variability now stops the line instead of being absorbed. Under push the pile in front of assembly was also a buffer. When cutting had a bad morning, assembly kept working, because there were two hundred units to work on. Remove the pile and a problem upstream is felt downstream almost immediately. That is the system working as designed, and it is genuinely harder to run.
And the cap has to be chosen. Too few signals in circulation and the line starves on ordinary variation. Too many and you have rebuilt push with extra paperwork. The number depends on how variable the steps are, which is a thing to measure rather than guess.
What to do with it
Find the piles before changing anything. Walk the process and write down what is standing between each pair of steps. Those locations are your capacity imbalances, already measured, for free. Do it even if you never touch the release system.
Check what step utilization is worth to you. If people are measured on it, a pull system will look like a failure to them on the day it starts working. Change the measure first or expect the rollout to be reversed by people acting rationally.
And size the buffer deliberately rather than removing it. The pile was doing a job. A pull system replaces an accidental buffer with a chosen one, and choosing it requires knowing how much the upstream steps vary, which most operations have never measured.
The gain is not tidiness. It is that the imbalance stops being invisible.