Key takeaways
Short answer: Push and pull are two answers to the same question: what tells the line to make the next unit? In a push system, a forecast-driven schedule does — you plan output in advance and push it through, building to plan. In a pull system, real downstream consumption does — nothing is made until a signal says it is needed, classically a kanban. Push optimises for plan adherence and machine loading; pull optimises for low inventory and responsiveness. Most plants blend the two. For the signal that drives pull, see andon vs kanban.
A push system starts with a forecast and a schedule. Planners decide in advance how much each step should produce, release those orders, and each step builds to its plan and pushes the result to the next — regardless of whether the next step is ready for it. Its strengths are predictability and utilisation: machines stay loaded, plans are explicit, and long-lead or seasonal items can be built ahead. Its weakness is that forecasts are always wrong to some degree, so push tends to accumulate inventory between steps and overproduce items that turn out not to be needed, tying up cash and floor space and hiding problems behind buffers.
A pull system inverts the trigger. Nothing is produced until a downstream step signals that it has consumed something and needs replenishment — most often a kanban card, bin, or electronic signal. Work is pulled through the system by actual demand rather than pushed by a forecast. The effect is much lower inventory, faster exposure of problems (there is no buffer to hide behind), and tight coupling between what the customer wants and what the line makes. The cost is fragility if it is poorly buffered: a sudden demand spike or an upstream stoppage can starve the line, because there is deliberately little inventory in reserve.
Push trades inventory for stability; pull trades buffer for responsiveness and cash. Push says keep everything loaded and absorb mismatches with stock; pull says hold almost no stock and stay tightly coupled to demand, accepting that the system is less forgiving. Neither is universally right. Pull is strongest where demand is reasonably steady, lead times are short, and changeovers are cheap. Push earns its place where demand is lumpy or seasonal, lead times are long, or you must build ahead for capacity reasons. The art is knowing which parts of your value stream fit which model.
A plant makes two products. Product A sells steadily, day in and day out. Run it on pull: a kanban loop keeps a small, capped buffer of finished A, and the line only rebuilds what packing consumes — inventory stays low and the line tracks real demand. Product B is seasonal, with a long supplier lead time on a key component. Forcing B onto pure pull would starve it the moment demand jumps. Run B on push: build to forecast during the off-season and hold the stock. Same plant, same lines, two scheduling models matched to two demand patterns. The hybrid beats dogma.
The decision is rarely plant-wide; it is per product family or even per part. Favour pull when demand is stable and frequent, lead times short, and the cost of a missed unit is low enough that a modest buffer covers it. Favour push when demand is volatile or seasonal, lead times long, or you must pre-build for a capacity crunch. Then size the buffers deliberately — a pull system with too little buffer is brittle, one with too much is just push wearing a kanban costume. Revisit the split as demand patterns and lead times change; it is not a one-time decision.
Scheduling model and OEE are more linked than they look. Push systems often post high machine utilisation that masks overproduction — the line is busy making units nobody needs yet, which flatters availability while creating waste. Pull keeps output matched to demand, so the OEE you measure reflects useful output rather than inventory build. Pull also exposes losses faster: with thin buffers, a downtime event on one machine quickly stops the next, making the six big losses impossible to ignore. That visibility is uncomfortable but it is exactly what drives improvement.
Fabrico measures the output that actually matters, not just whether machines were spinning. By tracking real production against demand-relevant targets and surfacing downtime and micro-stops with reason codes, it helps teams see where push is producing buffer rather than value and where a pull-starved line is being throttled by an upstream loss. That makes the push-versus-pull conversation evidence-based instead of philosophical. Book a demo to connect your scheduling model to live OEE.
Push production builds to a forecast and schedule, pushing output downstream whether or not it is needed yet. Pull production builds only in response to real downstream demand, usually signalled by kanban. Push optimises for plan adherence; pull optimises for low inventory and responsiveness.
No. Pull excels where demand is steady and lead times are short, but it can starve the line when demand spikes or lead times are long. Push is better for seasonal or long-lead items. Most plants use a hybrid matched to each product's demand pattern.
Most pull systems use kanban — a card, bin, or electronic signal that authorises replenishment as stock is consumed. The signal, not a forecast, tells the upstream step to build the next batch.
Overproduction and excess inventory. Because push builds to a forecast that is never perfectly accurate, it tends to accumulate stock between steps, tie up cash, and hide problems behind buffers.
Push can show high utilisation that masks overproduction, flattering availability while creating waste. Pull keeps output matched to demand and exposes losses faster because thin buffers make downtime propagate, making the six big losses visible.
Zaloguj
Blog
