Pull systems: letting consumption authorise production instead of a forecast

What changes when consumption authorises the work

The decision about when to start a job moves from a planner with a screen to a physical event on the floor. Planning does not disappear; it moves to setting rates, deciding what is held in each loop and levelling the pattern of demand entering the plant. Two consequences surprise people. First, upstream operations will sometimes have nothing to do, and this is correct rather than a failure. Second, the schedule stops being a list of jobs with dates and becomes a rate the plant is expected to hold. Supervisors judged on keeping people busy will fight both until the measurement changes.

Three architectures and the conditions each one needs

Replenishment pull holds a controlled quantity of finished parts and remakes what is taken; it suits items that repeat often and are cheap enough to hold. Sequential lanes hold no stock of specific parts but cap how much can queue between operations and enforce order; they suit high variety where holding every variant is impossible. A constant work-in-progress arrangement caps total released work across a whole route and admits a new job only as one leaves; it suits mixed and low-volume work where neither of the others fits. Most plants use a combination, split by how regularly each part is consumed.

The preconditions people skip

Pull assumes the supplying process can replenish reliably within the time the loop covers. If that process breaks down unpredictably, takes a long time to change over, or produces variable quality, the loop empties and the downstream operation stops, which is visible and unpopular. So the honest sequence is to improve availability and changeover capability first, then reduce circulation. It also assumes people will accept idle time when no signal is present, that material is available at the front of the stream, and that nobody senior will authorise an unsignalled batch when a customer complains.

Where pull is the wrong answer

Do not force it on parts made once, on engineered products designed to order, or on items with a very long procurement lead time for their inputs. Demand that arrives in rare, enormous blocks defeats replenishment logic, as does a product with a short shelf life and a slow-moving variant. Extremely expensive components rarely justify holding a loop of them. In those cases scheduling against real orders is more sensible, and the useful borrowing from pull is the cap on released work rather than the stock loop. Mixing approaches is normal; pretending one arrangement covers everything is what causes failures.

Evidence that pull is real rather than decorative

Look at the number of times someone overrides the signal in a week, whether any production has been made without authorisation, how often loops run dry and why, and whether stock in a loop is ageing because the part is no longer consumed. Compare elapsed time through the affected section before and after, since a calmer-looking floor with the same lead time has just been tidied. Also watch the planning department: if planners are still issuing detailed job lists to the operations inside the loop, the old system is running underneath and the cards are decoration.

Frequently asked questions

Does moving to pull mean we can stop using our planning system?

No. The planning system still handles material procurement, longer-horizon capacity, customer order promising and financial records. What changes is that it stops issuing detailed start instructions to the operations governed by a loop, since those are triggered on the floor. Running both without deciding which one authorises work is a common and damaging halfway house, because supervisors receive two instructions and follow whichever is louder. Write down explicitly which operations are signal-driven and which remain scheduled.

Will a pull arrangement cope with a sudden spike in demand?

Only within the range the loops were sized for. A spike larger than that will drain them, and the correct response is a deliberate, temporary decision to increase circulation or to add capacity, taken visibly rather than by improvisation on the floor. This is why levelling the pattern of work entering the plant matters so much: pull mechanisms are stable when the demand they see is reasonably smooth, and they transmit chaos faithfully when it is not.

How do we introduce pull without stopping the plant?

Start with one loop between two adjacent operations for a part that runs often, keep the existing scheduling in place elsewhere, and deliberately hold more in circulation than you think necessary. Watch it for several weeks, record every time it fails and why, fix those causes, then reduce circulation. Expanding one loop at a time is slower than a plant-wide switch and far more likely to survive, because each step teaches the team what their own process cannot yet do.