Kanban pull system

Goal

The Kanban pull system decides when a production happens and in what quantity: nothing is made upstream before a consumption signal has come back from downstream, and the quantity produced is the one that was withdrawn. "Kanban" is the Japanese word for a sign or a board. In production vocabulary the word names the signal itself, then the whole system that circulates it.

The problem it addresses is overproduction. In a workshop driven by a forecast schedule, each station pushes its output towards the next one whether that station has immediate use for it or not. Ohno ranks overproduction first among the wastes because it masks all the others: work-in-process inventory between two stations absorbs a breakdown, a scrapped part or an imbalance in load. The pull rule removes the possibility of producing ahead: with no signal, the upstream station stops.

The deliverable is a set of sized loops, one per part number and per pair of stations. Kanban says who authorises a production; the fall in inventory follows from that.

The technique belongs to the just-in-time pillar of the Toyota Production System, whose other pillar Ohno names jidoka, the automatic stopping of a machine on a defect. Just-in-time states the goal, which is to supply the right part at the right moment in the right quantity; kanban is the mechanism that reaches it.

Usage

When to use it

• Repeated consumption of standard items: the loop rebuilds itself, with no scheduling item by item.
• Work-in-process inventory piling up between two stations: the number of containers in circulation becomes its physical ceiling.
• A short and measurable replenishment lead time: the quantity tied up in the loop stays small.
• External supply under a framework contract: the empty container is the order, with no call-off and no purchase order.
• Consumables held by the people who use them: the signal is visible at the station, with no data entry and no stock count.

When not to use it

• Sporadic demand or one-off manufacture: there is no regular consumption to rebuild, so plan each job on order.
• An unstable supply lead time: the quantity that would cover the spread of lead times costs more than the rule saves, so drive that item on a forecast with an accepted safety stock.
• An item at end of life or whose revision level changes often: the contents of the loop go obsolete before they are consumed, so go back to launching on order.

Description

The idea that came from the supermarket

Ohno traces the mechanism to an American supermarket, which restocks its shelves at the rate customers take from them, without anticipating what they will take. He transposes it: the upstream station is the store, the downstream station is the customer. The customer goes to the store for the part they need, at the moment they need it, in the quantity they need; the store puts the quantity just taken straight back on the shelf. Those two propositions carry the two rules of the technique: downstream withdraws only what it consumes and upstream produces only what has been withdrawn. The other operating rules of the Toyota system, among them production levelling and the ban on passing a defective part on, frame how kanban works without belonging to it.

Toyota generalised the mechanism in its own plants from the early 1950s to full use around 1962, before extending it to its suppliers: the rule is stated in two propositions, holding it in a workshop is built station by station.

Push flow, pull flow

Pull reverses what triggers production. Parts move downstream under both arrangements. Under push, a forecast or a schedule decides upstream to produce, and the output goes down the flow independently of what the next station needs; the gaps between the schedule and actual consumption settle as work-in-process inventory at every interface. Under pull, consumption observed downstream releases the permission to produce upstream; the forecast schedule keeps its role for capacity, staffing and long-lead supply, but it no longer commands the launch.

The two-bin system

The smallest form that works calls for two containers and one label per item. The two bins holding the same part number sit at the line. The operator draws from bin A until it is empty, then puts the empty bin in the agreed place, a shelf edge or a return area, before starting bin B. The empty bin placed there is the replenishment order: the internal store or the supplier under framework contract takes it away and brings it back full while bin B empties in its turn. The bin that comes back is placed behind the one in service.

The signal and the authorisation are one and the same physical object: there is nothing to key in and nothing to forget. Three rules are enough to hold it. The second bin is not opened until the first has been put at the return point. A partly full bin is never sent back, since that would trigger a wrong replenishment quantity. Nothing is ever drawn from the bin under replenishment, the one left on the return area.

The bin label carries everything the replenishment calls for. In a card loop that same information sits on the kanban, which comes off the container as it is broached and travels back to the supplying station, where it becomes the production order for the quantity written on it.

Sizing the loop

In a card loop the capacity of the container is given by the standard pack and the question is how many containers to circulate. The working formula relates consumption over the replenishment lead time to that capacity:

N = (D × L × (1 + α)) / Q, where D is average consumption per unit of time, L the replenishment lead time in the same unit, α a safety margin and Q the quantity per container.

The two-bin form takes the problem from the other end: the number of containers is two by construction and it is the capacity that is calculated. While the bin that has left is being replenished, the station has only the second one, which must therefore cover consumption over the lead time on its own. The quantity per bin is then Q = D × L × (1 + α), rounded to the standard pack. As long as containers are few, it is the N − 1 bins left at the station that cover the lead time, that is (N − 1) × Q ≥ D × L × (1 + α); the general formula instead assumes continuous replenishment, hence enough containers in movement.

The lead time L is measured from the moment the signal is placed to the moment the full container is available at the station: it includes the pickup, the transport, the wait in the queue at the supplying station and the changeover, all of them absent from the machining time one would be tempted to put there. Consumption D is read off the history, separating items with steady demand from those whose demand arrives in bursts, for which the loop is not the right answer. For the margin α, 10 to 20% is the common choice in a stable chain, more when the supplier is far away. The quantity tied up grows with L: a long lead time makes the loop expensive without ruling it out. Reducing the contents of the loop means reducing L.

The pull rule and the kanban board

The word kanban names two distinct objects that are often confused. The first is the control rule born on the shop floor: a consumption signal authorises a production. The second is the kanban board and its WIP limits, a surface that spreads a team's tasks across columns according to their state of progress, each column capping the number of tasks admitted at the same time. Work in progress, which the tools call WIP, is work started and not finished.

The column cap transposes the pull rule to knowledge work: a free slot downstream is the empty bin, and it alone authorises the team to start one more task. Portfolio Kanban applies the same cap one level higher, on the initiatives a portfolio holds open at the same time. The rule and the board do not imply one another. The task board carries the same columns with no cap: it makes the state of the work visible, each person starting a task at the moment they judge right. The two-bin system, conversely, applies the rule with no display surface at all.

What makes the technique fail

The comfort stock beside the station

A box of parts kept under the bench so as never to run short is enough to neutralise the loop: the bin no longer empties, the signal no longer leaves and the store discovers the shortage on the day the box runs out. The hidden stock is the normal reaction of a workshop that has lived through shortages. It is treated by correcting the sizing that caused them; banning the box leaves the cause in place.

The production order kept alive in parallel

The planner goes on issuing orders out of material requirements planning, the ERP's MRP, while the cards circulate. The upstream station then receives two authorisations for the same part and produces on the more generous one. Kanban becomes a display and the flow turns back into push. On items driven by a loop, MRP no longer issues the launch order.

The empty container waiting to be collected

A weekly collection round adds several days to the real lead time without anyone recalculating a loop sized on the old one. The shortages that follow are blamed on the technique, when what they measure is a gap between the L of the calculation and the L of the shop floor. Collection frequency is a parameter of the loop in the same way as the quantity.

Cards taken out to bring stock down

Removing kanban reduces work in process by construction, which makes it the most tempting move in a stock reduction plan. It changes neither the lead time nor the variability of demand: the loop drops below the floor those two values impose, and the line stops. Taking a card out is prepared by the reduction in lead time that permits it, then verified over a few weeks before the next one.

AI considerations

A model works the consumption history of an ERP to maintain the loops. It sorts part numbers by how regular their demand is and names those whose profile has turned sporadic. It recalculates the quantities per bin on the consumption of recent months and flags the gaps with the values printed on the labels, tedious work nobody takes up item by item. On a history of movements it reconstructs the observed time between a signal being placed and the full container being made available, then compares that distribution with the lead time used in the calculation. It simulates the effect of taking a card out on the risk of a stockout, which puts numbers under a decision usually taken by feel.

Two things cannot be delegated. The replenishment lead time and the safety margin describe one particular workshop, with its transport hazards, its changeovers and the reliability of its supplier; a model ordered to fix them produces a plausible value with no origin, which is paid for in shortages. A model able to forecast a station's consumption invites replenishment before the signal comes back, which restores the push flow the technique removes. Forecasting serves to size the loop; the trigger stays with the signal. Consumption histories also carry the volumes of end customers and the prices of suppliers: they are anonymised before any transfer to an external service.

Examples

A subcontractor in the Jura Arc assembling precision sub-assemblies holds every one of its fastener and spring part numbers in a two-bin loop at the line. The artefact the technique produces is the loop label.

The store receives no written request for this part number. The empty bin placed on the return area is the whole of the order: its label says what, how many and for which station. The two bins tie up CHF 98.00 of material, seven days of consumption for a three-day lead time. That doubling is the price of the two-bin form, where each bin covers the lead time on its own.

The store then moves from a daily collection to a collection every two days and the lead time rises to four working days. From then on the bin in service has to cover 40 × 4 × 1.15 = 184 pieces, outside the pack of 140. The workshop keeps the pack and puts a third bin in the loop: two bins stay at the station while the third leaves for replenishment, giving 280 pieces available against a need of 184.

The same logic holds a stock of medical consumables in a care unit, two labelled bins per item in the cupboard, the nurse placing the empty bin on top. Nothing is manufactured there.

Visualisations

The figure sets the two arrangements against each other on two levels. Goods travel from left to right in both cases. The trigger arrow points downstream under push, from scheduling to each station; it points upstream under pull, from the consuming station to its supplier. The reversal of that second arrow is what the figure has to make visible.

The two-bin loop is drawn as a closed cycle: the bin in service at the station, the empty bin on the return area, the trip to the supplying store, the return of the full bin. The event that closes the loop, the bin running empty, carries its label with the quantity, which ties the drawing to the sizing calculation.

Cost

Tooling

Bins and laminated labels are the complete kit of an internal loop. A return area visible from the station and a standard pack are enough to circulate the signal, with no licence and no network. This is where to start even in a well-equipped plant: the physical loop is corrected by changing a label.

Kanban cards with a barcode or a QR code add traceability when the supplying store is far from the station. Scanning the kanban on departure and on arrival gives the real lead time per movement, the measurement the whole sizing depends on.

The kanban modules of ERP systems, among them SAP S/4HANA which carries the function under that name, hold electronic loops: the station declares the container empty, the system issues the replenishment order and keeps the history. They suit loops between sites or with external suppliers. The name invites confusion: in many management tools a "kanban view" is a display in columns with no pull rule. The risk with the real modules is that they rebuild classic scheduling behind the vocabulary: the order then leaves again from material requirements planning.

Supplier portals extend the loop beyond the company under a framework contract: the empty container triggers a delivery on a contract already negotiated, with no order to issue part by part.

A spreadsheet remains the tool of the sizing. It carries the consumption read off, the lead time measured, the margin and the quantity obtained for each part number, which keeps the periodic recalculation sustainable.

Sources

• Taiichi Ohno, Toyota Production System: Beyond Large-Scale Production, Productivity Press, 1988, ch. 1 "Starting from Need": the two pillars of the Toyota Production System, just-in-time and jidoka, together with the rank of overproduction among the wastes. The section "An Idea from the U.S. Supermarket" of the same chapter carries the origin of the mechanism, the supermarket analogy and the two rules of withdrawal by the downstream station and production of the withdrawn quantity alone.
• Lean Enterprise Institute, Lexicon: Kanban: the definition of kanban as a signalling device carrying the authorisation and the instructions to produce or to withdraw in a pull system.
• Lean Enterprise Institute, Lexicon: Pull Production: the opposition between push flow and pull flow, together with the three forms of application, among them supermarket pull, to which the two-bin system belongs.
• American Society for Quality, Quality Glossary: Kanban: the formulation of the card cycle, which returns to its source and there counts as a replenishment order.