Flow before pull: why the order of these two principles decides the outcome

Flow is an argument about queues, not about speed

Trace one order through a plant and record where the material sat. In almost every case the great majority of elapsed time is spent waiting rather than being worked on. That finding reframes the improvement question: making a machine cycle faster attacks the small portion, whereas reducing the queues attacks the large one. It also explains why plants that invest heavily in faster equipment frequently see no change in delivery performance. Flow asks a different question of engineers, namely what would have to be true for a part to move to its next operation immediately, and the answers are rarely about machine speed.

Why flow has to be attempted before pull is installed

A signalling system transmits a stoppage rather than absorbing it. If the supplying operation is unreliable, the downstream one stops as soon as the loop empties, and the plant experiences the improvement as a loss of output. Everything that used to be hidden by a pile of parts becomes an immediate delivery problem. So the productive sequence is to make processes capable of holding a rate first: attack the chronic breakdowns, get changeovers short enough that small runs are affordable, stop defects being passed on. Only then does removing the buffers become an exercise in improvement rather than an exercise in stopping.

The uncomfortable implications management has to accept

Taking flow seriously means smaller batches, more changeovers and lower utilisation of every resource that is not the constraint. Machines will stand idle by design. Traditional cost reporting reads all of this as deterioration, because unit costs rise when overhead is absorbed over fewer pieces and equipment appears underused. Unless the plant manager and the finance function agree in advance how performance will be judged during the transition, the reporting will win and the change will be reversed. That conversation is a precondition, not a detail to be handled once the numbers start looking strange.

Where continuous flow is physically impossible

Some operations cannot be broken into small pieces without destroying their economics or their process. Heat treatment, curing, plating, painting, fermentation, kiln firing and large shared presses all impose a batch character. The realistic approach is to decouple around them: hold a controlled buffer either side, run the shared asset on a repeating pattern that downstream operations can plan against, and pursue flow in the segments where it is achievable. Pretending otherwise leads teams to attack the one operation where the constraint is genuinely physical, while leaving the surrounding queues untouched.

Testing whether the principles are being applied at all

Pick a real order, follow it, and record the clock time at each handover, including nights and weekends. Add up the time anything was actually being done to it. Repeat the exercise on a comparable order after the changes. If the proportion of waiting has not fallen, nothing structural happened regardless of what was installed. A second test is behavioural: ask a supervisor what they do when the next operation is not ready. If the answer is to keep producing and stack the output somewhere, the plant is still running on the old principle whatever the boards say.

Frequently asked questions

Can a job shop with high variety apply flow principles at all?

Yes, though not by arranging dedicated lines. The useful moves are capping how much work is released so queues stop growing, grouping equipment into cells for families of parts that share a routing, sequencing work consistently at shared resources, and reducing setup so small orders are affordable. The principle survives high variety; the standard textbook layout does not. Expect the biggest single gain to come from release discipline rather than from any physical rearrangement.

Is it wrong to hold any buffer stock if we are pursuing flow?

No. Buffers placed deliberately, sized consciously and reviewed regularly are a legitimate engineering choice, particularly around a batch process, a shared asset or an unreliable external supply. What flow objects to is stock that accumulates because nobody decided anything, and stock whose level is never questioned. The discipline is to know why each buffer exists, who owns it, and what would have to improve for it to be reduced.

Our output fell when we reduced batch sizes. Did we do something wrong?

Probably the changeover work was not tackled first. Smaller batches mean more setups, and if each setup still takes a long time the plant simply loses productive hours at the resource that limits output. The remedy is to protect the constraint from extra setups while attacking setup time on the machines that have spare capacity, then reduce batches there. Reducing batch size uniformly across a plant, without regard to where capacity is tight, reliably costs output.