Pull system or push system? Choose the best one for your production

A pull system triggers production based on actual consumption, while a push system triggers production based on a plan or forecast. This difference affects inventory levels, lead times, the number of work-in-process items between operations, and how the company responds to changes in orders.

After reading this article, you’ll know when a pull system works best, when a push system is preferable, and in what situations it’s worth combining both models. You’ll also learn about five criteria that will help you assess the workflow in your facility without having to overhaul your entire production process.

The pull system, also known as the demand-driven system, triggers the production or replenishment of materials upon receiving a signal indicating actual demand. The signal may come from a customer, an assembly line, another workstation, or a production warehouse.

The material moves toward the recipient, and information about the need is fed back to the previous process. When the assembly line retrieves a container of components, the supply station receives an instruction to replenish the specified quantity.

The Lean Enterprise Institute defines pull production as a control method in which a process further down the production stream signals its demand to the preceding stage. One of its goals is to reduce overproduction.

Hopp and Spearman propose a more technical approach: the fundamental characteristic of a pull system is a mechanism that limits the amount of work in progress (WIP). Simply responding to orders is not enough to classify a process as a pull system. Control over the number of items or orders in the flow is necessary.

Toyota applies a similar principle in its Just-in-Time system. Each process takes the required amount from the previous one at the exact moment it is needed.

However, “pull” does not mean production without inventory. It exists, but its size depends on the rate of consumption, the replenishment time, and the security measures in place.

A push system initiates production based on a forecast, sales plan, or a schedule. Orders are created in advance, before the next process or a customer indicates an actual need.

If the forecast calls for sales of 10,000 products next month, the company orders materials in advance and schedules the work of its departments. Semi-finished products move through the subsequent production steps according to the schedule.

Push often works in conjunction with an MRP system, which calculates demand based on:

This model allows you to secure materials and production capacity in advance. However, its effectiveness depends on the quality of the data. A forecasting error, a supplier delay, or a machine breakdown can quickly cause the plan to diverge from the actual situation on the shop floor.

In a push system, production is triggered by projected demand. In a pull system, production is determined by actual withdrawal, consumption, or orders, and the amount of WIP is controlled.

Push determines what will be needed according to the plan, while pull checks what has been used and needs to be replenished.

None of these systems is automatically better. The outcome depends on demand variability, replenishment time, process stability, the number of variants, and the cost of holding inventory.

Pull limits the number of orders that can be in process at any given time. A workstation should not produce another batch simply because the machine is idle. It needs a signal from the receiving process and available capacity within the established WIP limit.

Such a signal could be:

Kanban grants approval to produce or move a specific quantity of material. It is a tool of the pull system, not a synonym for it.

Consider a simple example. The assembly line uses an average of 200 components per day. The full replenishment cycle—measured from the time the replenishment signal is sent until the materials are delivered—takes one day. The company maintains a safety stock of 100 units.

In this simplified example, 300 units cover the average consumption during the replenishment period and the assumed safety stock. This is not yet the number of kanban cards or the target inventory level for the process.

When determining the number of signals, the following must also be taken into account:

A container that is too large increases inventory and delays the detection of discrepancies, while one that is too small increases the number of transport runs and alerts. The parameters must be calculated for a given throughput.

The most immediate effect stems from the reduction in overproduction. The plant does not start new orders without a signal of demand, so fewer work-in-progress items accumulate between operations.

A decrease in work in progress may result in:

A large backlog between machining and assembly delays feedback. If the assembly line detects a defect after a few days, the previous station may have already produced additional batches using the same parameters during that time. A shorter queue reduces the number of components susceptible to the same error.

The effect must be evaluated based on data from your own process. The following will provide the most information:

Implementations described in the literature often combine Kanban with reducing changeover times, standardizing work, rearranging workstations, and improving quality. The results of the entire program should not be attributed solely to the pull system.

The risk most often applies to processes that are unable to replenish materials within a predictable timeframe.

Frequent breakdowns, fluctuating cycle times, long changeover times, and inconsistent quality can quickly deplete a supermarket’s inventory. An insufficient buffer will lead to component shortages. Continuously increasing inventory will undermine the purpose of the pull system and once again mask disruptions.

Before implementation, check the following:

Pull quickly identifies areas where the process becomes unstable. The Kanban signal alone will not eliminate the cause of the delay. If materials are not consistently returning on time, you need to determine whether the source is a breakdown, a backlog of orders, a shortage of operators, quality issues, or internal transportation.

A “push” system is justified when production or the purchase of materials must begin well before a customer’s order is received.

This applies, among other things, to:

A manufacturer that imports specialized materials cannot wait to place an order for every single purchase over a three-month period. It must determine part of its demand in advance.

A similar situation arises ahead of a predictable sales season. Materials or standard products can be prepared in advance, since producing them only after orders have increased would result in delays.

The biggest weakness of the push system stems from initiating new orders without checking whether the next process is capable of handling them. The schedule may look fine, while on the shop floor, queues are growing, space is running out, and lead times are increasing.

It is possible to effectively combine pull and push in a single workflow. Part of the process proceeds as planned, and subsequent operations begin after an order is placed or inventory is depleted.

The boundary between these areas is defined by the customer order split point. Before that point, you can purchase materials or produce common components according to the plan. After that point, you’ll begin configuration, assembly, or packaging tailored to a specific need.

Modular production is one example. Standard components are manufactured in advance because they are used in many product variants. Final assembly begins once the customer has selected a configuration.

This arrangement shortens delivery times without requiring the company to maintain a ready stock of all possible versions. The most important decision, then, is the location of the distribution point.

The matrix below helps translate theory into a decision regarding a specific product family.

Do not treat a single response as conclusive. A predominance of pull-side signals may justify testing a suction loop. A split in the response between the two columns often indicates a hybrid model.

The relationship between lead time and the customer’s deadline requires special attention. If it takes six weeks to manufacture a product and the customer expects delivery in one week, part of the process must begin earlier. Final assembly can still be started based on the actual order.

The first test should involve a single loop between two well-defined processes. The easiest way to start is with a component that is used regularly and for which the path and replenishment time can be measured.

Data from the MES system can help measure actual consumption, replenishment time, WIP levels, and the causes of downtime. Without this information, control loop parameters are often set based on self-reported figures or averages, which do not reflect the variations that occur during normal operation.

After a period covering a representative number of replenishment cycles, compare the results with the data from before the pilot.Take into account typical fluctuations in demand, production changes, changeovers, and disruptions. Verify that inventory and lead time have decreased, and that the number of shortages and on-time delivery rates have remained at acceptable levels.

Only such a result provides the basis for including additional indices in the system.

It will be most helpful to determine what triggers a task at each stage and whether the next process is ready to receive its output.

If orders are opened primarily to keep the machines running, measure work in progress, queue lengths, and total lead time. If production from raw materials takes longer than the customer’s expected delivery time, determine which materials or modules need to be prepared in advance.

A well-designed system rarely relies on a single principle throughout the entire facility. Planning makes sense where it is necessary to stay ahead of demand. Replenishment based on consumption works well where production without a signal increases inventory and slows down the flow.

The safest first step is to analyze a single product family. Collect data on consumption, replenishment time, work-in-progress (WIP), and material shortages. Based on this, you can identify where to use the pull model, where the push model is needed, and the point where the two models should converge.

A pull system is a method of production control in which the production or replenishment of materials begins only after a signal indicating an actual demand from a customer or a subsequent process. The system also includes a mechanism that limits work in progress.

In manufacturing, the pull system links the work of earlier operations to consumption at later stages. The supplying process produces a specified quantity upon receiving a pull signal.

Pull responds to orders or demand and controls the level of work in progress (WIP). Push initiates production based on forecasts, sales plans, or schedules.

No. Pull determines how the flow is controlled, while kanban is a signal that triggers production or the replenishment of materials.

No. It helps control their size based on consumption, replenishment time, and the security measures in place.

Yes. The company can prepare materials or common components based on forecasts and begin final assembly once an order is received.

For a single product family or a single material loop. Measure consumption, total replenishment time, WIP levels, and the frequency of shortages.

Compare inventory levels, lead times, on-time delivery rates, the number of shortages, and urgent schedule changes. The measurement should cover a representative number of replenishment cycles.