Lean tools

Kanban

In French: Kanban

A kanban is a signal — card, bin or empty location — that authorises production or replenishment, and that puts the flow in pull rather than push.

What it is

Kanban means "label" or "signboard". In a pull system, nothing is produced until a signal has authorised it: the downstream station consumes, the signal travels back, the upstream station replenishes. Stock stops being decided by a forecast and becomes a consequence of the actual rate of consumption.

The most useful consequence is unintended: the number of kanbans in circulation caps the work in progress. It is no longer possible to produce ahead, so the problems that were absorbed by stock become visible. That is uncomfortable, and it is exactly the effect sought.

Beware of the homonym: the kanban of industrial production and the kanban board of software teams share the idea of limiting work in progress, but are not managed the same way.

The simplest form: two bins

The two-bin system is the minimal kanban, and often the only one needed. Two containers of the same reference at the workstation: you draw from the first, and when it is empty you send it back to the store — that empty bin is the signal — while you consume the second. No card, no software, no data entry.

It assumes only one thing: that the second bin covers the replenishment time. The whole sizing lies in that.

When kanban is not suitable

Pull flow assumes a fairly regular consumption and a stable catalogue of references. On make-to-order production, highly varied and with no recurrence, there is nothing to replenish in advance: kanban has no grip, and the answer lies elsewhere — changeover reduction, sequencing, continuous flow.

It also assumes upstream quality under control. A pull system has no stock to absorb a non-conforming batch: the defect immediately becomes a downstream stoppage. That is a benefit when one is ready to treat the cause, a blockage when one is not.

Sizing a loop, figure by figure

A reference consumed at 480 parts per day, replenished in 1.5 days, packed 60 parts per bin, with a safety factor of 20 % to absorb the variability of demand and replenishment.

N = (480 × 1.5 × 1.2) / 60 = 864 / 60 = 14.4, rounded up to the next whole number: 15 cards. Rounding is always upwards — fourteen cards would not cover the lead time, and the station would run out before the end of the replenishment.

The figure that counts is not 15, it is what it implies: the work in progress of this loop is now capped at 900 parts, i.e. 1.9 days of consumption. That cap is physical, not declarative — with no card available, upstream does not produce, even if it has the time and the inclination. That is the real contribution of kanban, and it is independent of the exact value of the calculation.

The heaviest parameter is the lead time, by far. By bringing replenishment down from 1.5 days to half a day — through an SMED project upstream, or simply a more frequent collection round — the calculation gives (480 × 0.5 × 1.2) / 60 = 4.8, i.e. 5 cards and 300 parts of work in progress. The stock is divided by three without touching the formula or negotiating anything: one has acted on what it measures.

The safety factor, on the other hand, is paid for in cash. Raising it from 20 % to 50 % "to be on the safe side" gives (480 × 1.5 × 1.5) / 60 = 18 cards: three more cards, 180 more parts of work in progress, and a loop gone silent — that is, incapable of signalling anything. The comfort bought is exactly what prevents seeing the disruptions one claimed to be covering.

In routine operation, tuning is done by removing a card. Going from 15 to 14 cuts the work in progress by 60 parts and tightens the loop until a problem appears — a late collection, a temperamental machine, an operator absent on Tuesdays. That problem is treated, then the next card is removed. That is the improvement mechanism of kanban, and it works only if it was decided in advance to treat what appears rather than put the card back.

Number of kanbans

N = D × L × (1 + α) / Q

D: average demand per unit of time · L: replenishment lead time · α: safety factor covering variability · Q: quantity per container. The formula is a starting point to be adjusted on the shop floor: a comfortable α makes the system silent, but takes away precisely its ability to reveal problems.

The common mistakes

Sizing the number of cards on peak demand and never revisiting it. The kanban then becomes stock with labels: it pulls nothing and reveals nothing any more. A kanban system is reduced progressively — it is that movement that brings out the real points of fragility.

Answering a stock-out by adding cards. It is the immediate reflex, it works, and it turns the system into disguised safety stock. The stock-out is a signal, not a breakdown: it says that the replenishment lead time exceeded what the loop covered. Adding a card masks the signal; reducing the lead time treats it. The question to ask before touching the number of cards is always the same — what lengthened the replenishment this time?

Letting the kanban and a work order released by the ERP coexist on the same reference. The two signals contradict each other, upstream produces on whichever arrives first, and the work in progress stops being capped by anything. On references switched to pull, the ERP must stop scheduling — otherwise there are two control systems and no rule to arbitrate between them.

On the shop floor — Logistics

In a components store, switching to a two-bin kanban eliminated stock-outs without increasing overall stock: it was not the quantity that was missing, it was the replenishment trigger that came too late.

The other tools illustrated by this trade: logistics and distribution.

Frequently asked questions

How do you calculate the number of kanbans?

You multiply the average demand per unit of time by the replenishment lead time, add a safety factor, then divide by the quantity held in one container, and round up to the next whole number. A consumption of 480 parts per day, a lead time of a day and a half, 20 % safety and bins of 60 parts give 15 cards. The result is a starting point, to be reduced afterwards on the shop floor.

What is the difference between pull flow and push flow?

In push flow, you produce on the basis of a forecast: each station makes what the schedule asks of it and passes it downstream, whether it is needed or not. In pull flow, nothing is produced without a signal sent by the downstream station that has just consumed. The practical difference lies in the work in progress: in push it results from the gaps between forecast and reality, in pull it is capped by the number of signals in circulation.

Is production kanban the same as the kanban of software teams?

They share one idea — limiting work in progress to reveal blockages — and nothing else in their implementation. Industrial kanban replenishes a physical reference consumed on a recurring basis, and is sized by a calculation from demand and lead time. The software kanban board tracks unique, non-repeatable tasks, where the work-in-progress limit is set per column and adjusted through use.