A method of regulating production in which downstream activities make known what they need to upstream activities. Pull production aims to prevent overproduction and is one of the three key components of a complete Just-in-Time production system. 

In pull production, a downstream activity (either within the same facility or in a separate facility) provides information to the previous (upstream) activity, often via a kanban map, about what part or material is needed, how much is needed, and when and where it is needed. The upstream supply process in the value stream produces something only when a need is expressed from the downstream customer process. This is the opposite of push production. 

There are three main types of pull production systems: 

Supermarket pull system 

The most basic and most widely used type, also called fill-up, replenishment or type-A pull system. In a supermarket fill-up system, each process has a store - a supermarket - in which there is a certain amount of each product being produced. Each process produces purely to replenish what is consumed from its supermarket. The moment material is withdrawn from the supermarket by the downstream customer process in the value stream, a kanban or other type of information carrier is sent to the upstream supply process to take product. This authorizes the upstream process to replenish what has been consumed. 

Each process is responsible for replenishing its supermarket. In this way, it is relatively easy to manage the shop floor and opportunities for kaizen are relatively easy to see. The disadvantage of a supermarket system is having to have inventory, which is sometimes not feasible if the number of types of parts is large. 

Sequential pull system 

A sequential pull system - also known as a type-B pull system - can be used when there are too many types of parts, not all of which can be stocked in a supermarket. Products are essentially made to order while minimizing inventory for the entire system. 

In a sequential system, the planning department must determine the right mix and numbers of products to be produced. This can be done by placing production canban cards in a heijunka box, often at the beginning of each shift. These production instructions are sent to the first process all the way upstream in the value stream. This is often done in the form of a "sequence list," also called a sequential tablet . Each subsequent process produces purely the items it has been supplied by the previous process, always adhering to FIFO of individual products. 

A sequential system creates pressure to maintain short and predictable lead times. This system is effective only if one understands the pattern of customer orders. If orders are difficult to predict, production lead times must either be very short (shorter than the order lead time) or an adequate store of finished goods must be provided. 

A sequential system can only be maintained with strong management, and improving it on the shop floor can be quite challenging. 

Mixed supermarket and sequential pull system 

Supermarket and sequential pull systems can be used together in a mixed system - also known as a type-C pull system. A mixed system may be in place when an 80/20 rule applies, where a small percentage of part numbers (e.g., 20%) accounts for most (e.g., 80%) of the daily production volume. An analysis is often performed to segment types of parts by volume 

into high (A), medium (B), low (C) and rare orders (D). Type D includes, for example, custom or service parts. For these low-demand items, a special type-D canban can be created that does not represent a specific type of part, but an amount of capacity. Then the production order for the type-D items is determined using the method used by the planning department for types of parts within the sequential pull system. 

With such a mixed system, both supermarket and sequential systems can be applied selectively and reap the benefits of both systems, even in environments where demand is complex and varied. The two systems can be applied side-by-side throughout the value stream, or they can be used for a particular type of part at different locations within that number's individual value stream. 

With a mixed system, it can be more difficult to keep work in balance and to identify abnormal conditions. It can also be more difficult to manage and execute kaizen. Therefore, it takes discipline to make a blended system work well. (Based on Smalley 2017.)