QM BenchQuality engineering tools

Push vs pull: same line, same parts, different stock

Four stations in a row, each with random processing times; station 2 is slightly slower than the rest. On the left, work is released to a fixed schedule (push). On the right, a station may only work when there is room in the buffer after it (pull, kanban). Change the variability, the release interval and the kanban limit.

Line and rules

four stations with average times 1.0, 1.1, 0.9 and 1.0 minutes; station 2 is the bottleneck

Push: work released to the schedule

Finished in 8 h
414
Average WIP (pcs)
35.7
Average lead time (min)
36.2
Max WIP (pcs)
66

Pull: work only when the next buffer has room

Finished in 8 h
407
Average WIP (pcs)
7.9
Average lead time (min)
9.3
Max WIP (pcs)
10

Work in progress over the shift

pieces in the line, minute by minute
0.020.040.060.00123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479Minute of the shift
PushPull

Little’s law check

Throughput (pcs/min)Average WIPWIP ÷ throughputMeasured lead time
Push0.86335.741.4 min36.2 min
Pull0.8487.99.3 min9.3 min

Both lines make about the same number of parts, because the bottleneck sets the output. Push only adds stock and lead time. Lead time ≈ WIP ÷ throughput in both cases: the only way to shorten lead time at the same output is to limit work in progress.

Things to try

  1. Release every 1.00 min: faster than the bottleneck (1.1 min). Push stock grows all shift; pull stays flat with the same output.
  2. Release every 1.20 min: slower than the bottleneck. Now push has little stock too, but the line loses output.
  3. Kanban limit 1 and variability 80 %: pull starts to lose output. Raise the limit until output recovers: that is the buffer the variability needs.
  4. Variability 0 %: with perfectly even times a limit of 1 is enough.

Size real kanban loops with the kanban calculator, see batch versus flow in the one-piece flow simulator and read about Little’s law. All lean tools: lean hub.

Frequently asked questions

What is the difference between push and pull?

In a push system work is released to a schedule or forecast, whether or not the next step is ready. In a pull system a step only produces when the next step has used something and signalled it, for example with a kanban card, so stock between steps cannot grow beyond the limit.

Does pull produce less?

No, as long as the kanban limit is not too small. The output of a line is set by its bottleneck. Push adds stock in front of the bottleneck but no extra output; pull keeps the bottleneck busy with a small buffer and removes the rest.

What is Little’s law?

In a stable system, average lead time equals average work in progress divided by throughput. The simulator calculates both sides: measured lead time and WIP ÷ throughput agree.

Why does variability matter?

With random station times, a buffer protects the bottleneck from starving when the step before it is slow. Too small a kanban limit with high variability loses output; the right limit balances output and stock.