Push production schedules work from a forecast and sends it downstream on a plan, whether or not the next step is ready for it. Pull production makes or moves an item only when a downstream signal shows the previous one was consumed. Push is driven by a plan; pull is driven by real demand. That single difference, what authorizes the next unit, changes almost everything about how a plant carries inventory, exposes problems, and reacts to change.
This is a concept comparison, not a tutorial on cards and bins. If you want the mechanics of the signal itself, read kanban in manufacturing. Here we compare the two philosophies underneath it, because most plants run some of both and the interesting question is not "which is right" but "which belongs where." Pull is one of the core principles of lean manufacturing but a badly placed pull loop is worse than an honest push schedule.
What Is the Difference Between Push and Pull Production?
The difference is the trigger for making the next unit. In a push system the trigger is a schedule built from a forecast: the plan says build 400 of part A on Tuesday, so the cell builds 400 and moves them to the next step, whether that step needs them yet or not. In a pull system the trigger is consumption: the next step takes a container of parts, and that act of taking is the only authorization to make one more. Nothing gets built ahead of a real signal.
Push asks "what does the forecast say we should build?" Pull asks "what did the customer just take?" A forecast is always partly wrong, so push tends to build the wrong mix and cover the error with inventory. Consumption is always right, it already happened, so pull builds exactly what left the shelf, capped at a fixed maximum. The cost of that accuracy is that pull only works when replenishment is fast and reliable enough to keep up.
How Does a Push System Work, and Where Does It Break?
A push system plans production centrally and releases it on a schedule. The classic engine is MRP: take a demand forecast, explode it through the bill of materials net against on-hand inventory, and issue work orders with due dates. Each work center does its due-date work and pushes the output to the next center. It is a rational, plannable way to run a plant, and for long lead times and stable products it is often the right one.
Push breaks in three predictable ways. First, the forecast is wrong, it always is to some degree, so the plant builds a mix that does not match what customers actually order, and the gap shows up as finished goods that do not sell and stockouts on things that do. Second, push decouples steps with inventory, so a problem upstream is hidden by the buffer downstream and nobody feels it until the buffer runs out. Third, pushed material accumulates as work-in-process between steps, and that inventory is not free: it ties up cash, hides defects, and lengthens the effective lead time because a new order sits behind everything already in the queue. Overproduction is the first of the eight wastes for exactly this reason, it manufactures the other seven.
How Does a Pull System Work?
A pull system replaces the central schedule with local signals. Downstream consumption authorizes upstream replenishment, and the amount that can be in flight is capped by the number of signals, a form of visual management such as cards, bins, empty squares, or an electronic equivalent, in circulation. Because the cap is fixed, inventory cannot creep past it no matter how enthusiastic a work center gets. Taiichi Ohno built the idea at Toyota after watching American supermarkets restock a shelf only as customers emptied it: the shelf is a small, bounded buffer that the store refills to a set level, never past it.
Most real pull systems are not pure "make only on order." They use a small, deliberate buffer called a supermarket at the points where flow cannot be continuous, different cycle times, shared equipment, a changeover in between. The downstream step pulls finished parts from the supermarket; the pull from the supermarket is the signal that tells the upstream step to make one more and refill it. That is why pull is often described as replenishment: you are not building to a plan, you are keeping a bounded shelf topped up to the level real demand sets.
Which Is Better: Push or Pull?
Neither is better in the abstract; each wins on a different axis. Pull minimizes inventory and surfaces problems fast, but it needs short, reliable replenishment and reasonably level demand. Push copes with long lead times and lumpy demand, but it pays for that flexibility in inventory and slow problem detection. The honest comparison is a trade-off table, not a verdict.
| Push (make to forecast) | Pull (replenish what was consumed) | |
|---|---|---|
| Trigger to build | A schedule from a forecast | A downstream consumption signal |
| Inventory | Tends to build; capped only by discipline | Capped by the number of signals, by design |
| Responds to demand change by | Re-planning and re-releasing orders | Automatically, it only replaces what left |
| Problems are | Hidden behind buffers until they run out | Exposed fast when a loop starves |
| Needs | Good forecasts, MRP discipline | Short lead times, level demand, few defects |
| Struggles when | Forecast error is high or the mix churns | Demand is lumpy or lead times are long |
| Fits | Long-lead, custom, or highly seasonal items | Stable, repeating, high-runner parts |
A common trap is treating pull as automatically "lean" and push as automatically "wasteful." Pull applied to a part with wildly lumpy demand just creates a supermarket that is either always empty (stockouts) or sized so large it is push wearing a card. And push run with a disciplined master production schedule and honest inventory targets can be perfectly lean for the products it fits. The waste is not push or pull; it is using the wrong one for the demand pattern in front of you.
How Do You Decide Push vs Pull for Each Product?
Decide per part, not plant-wide, by comparing demand stability against replenishment lead time. A workable sequence:
- Sort parts by demand pattern. Run a simple runner-repeater-stranger split: runners sell steadily every day, repeaters sell in known cycles, strangers are sporadic or one-off. Pull loves runners; strangers usually belong on push.
- Measure replenishment lead time versus demand variation. If you can refill the buffer faster than demand can drain it, pull is available. If lead time is long relative to how fast demand swings, a pure pull loop will starve, push or a larger buffer is honest.
- Set the pull points where flow cannot be continuous. Continuous one-piece flow beats both push and pull; use a supermarket only where different cycle times, shared machines, or changeovers force a break. Map these on a value stream map.
- Size the buffer against variation, not comfort. Base the maximum on average demand during lead time plus a deliberate safety factor, the same logic as safety stock. Write the number down; do not let it drift.
- Level the schedule feeding the loop. Pull works far better when the upstream demand it sees is smoothed. Heijunka leveling the mix and volume, is what keeps a pull system from being whipsawed.
- Review the split on a cadence. Products migrate. A stranger that becomes a runner earns a pull loop; a runner that fades should lose its buffer before it becomes dead stock. Revisit at least quarterly.
Why Do Most Plants Run a Hybrid?
Because no real catalog has one demand pattern. A typical plant has a handful of high-runner parts that pull beautifully, a long tail of sporadic items that only make sense on push, and seasonal or custom work that has to be planned. Forcing one philosophy across all of them means either supermarkets full of slow movers or high-runners waiting on a re-plan. The hybrid is not indecision; it is the correct answer applied part by part.
Many plants also run a mixed model on purpose: push the long-lead components to a decoupling point using a forecast, then pull the final assembly and finishing from real orders. That is the same idea as the make-to-stock versus make-to-order decision, and it is often the most economical shape, buy time upstream where lead times are long, stay demand-driven downstream where the customer is watching. The constraint that governs the whole thing usually sits at one step; theory of constraints is how you decide where to hold the buffer.
By the Numbers
Pull is not a Toyota curiosity; it is one of the load-bearing principles of lean. The Lean Enterprise Institute lists "pull" among the core lean principles, defining it as producing only what the next step needs when it needs it, rather than pushing on a forecast (Lean Enterprise Institute, Continuous Improvement). Toyota credits this consumption-driven logic as the foundation of just-in-time production in the Toyota Production System, the system that made pull famous (Toyota Motor Corporation, Toyota Production System). The structural payoff is that inventory stops being a number you fight down every month and becomes a property of the system: cap the signals and you cap the stock. Where Harmony fits: pull only works if consumption signals are accurate and timely. Harmony captures real usage from the line and keeps it connected across your systems, so a pull signal reflects what actually happened on the floor rather than a stale count in a spreadsheet, see the platform.