Production kanban: designing a signal the floor will actually obey

The card is permission, not a job list

Treated properly, a returning card is the only thing that permits a supplying operation to run a part. That reverses the usual instinct, which is to keep a machine busy whenever it is set and material is available. The first serious test comes when a supervisor sees an idle machine and an obvious opportunity to build ahead. If that extra run is authorised, the loop quantity is no longer meaningful and everyone learns that the rule is advisory. Making the rule stick requires management to state plainly that a stopped machine with no signal is the correct condition, and to stop measuring the area in a way that punishes it.

Choosing the carrier: card, container, floor square or electronic call

The physical form should match the distance and the environment. Cards are flexible and get lost, soaked or pocketed. A dedicated container is its own signal and cannot be misplaced, which suits line-side parts where an empty bin triggers replenishment. A marked floor square works where product is bulky and visible. Electronic signalling survives distance and separate buildings but removes the visual cue that lets a passing supervisor see a loop is in trouble. Many plants end up mixing forms, which is fine provided each loop has one unambiguous trigger rather than a card system shadowed by informal phone calls.

Setting the loop quantity without pretending to precision

The quantity in circulation has to cover consumption during replenishment, plus an allowance for how erratic that consumption and that replenishment actually are. Container size and minimum practical run length constrain it further. Formulas exist, but they are only as good as the replenishment time you feed them, and most plants do not know theirs. A more honest approach is to start deliberately generous, measure how often the loop runs dry and how long stock sits, then withdraw circulation in small steps and watch what breaks. Each withdrawal is an experiment, and the problem it exposes is the thing worth fixing.

Exceptions are where these systems quietly die

Rush orders, engineering changes, quality holds, trial builds, seasonal ramps and supplier failures all arrive eventually. If there is no defined way to handle them, people improvise: extra cards appear, quantities are written over, a batch is made without a signal because the customer was shouting. Within months the loop bears no relation to its design and nobody trusts it. Decide in advance who may inject an exception, how it is marked so it is visible, and how the loop returns to normal afterwards. Reviewing exceptions is also good diagnostics, because a loop needing constant overrides is telling you its design is wrong.

Keeping the loop honest over time

Loops decay through ordinary neglect. Parts are superseded and their cards stay in circulation. A container specification changes and the quantity printed on the card no longer matches what fits. Volumes shift with the product mix and nobody revisits how much is circulating. Someone has to own a periodic count of cards against the register, removal of dead parts, and a review whenever a design change or a volume change lands. Give that ownership to a named person in production control rather than to the improvement team, because it is maintenance of a control system, not a project.

Frequently asked questions

Can a kanban loop work for parts we make only occasionally?

Poorly. A signal that authorises replenishment assumes the part will be consumed again reasonably soon, so holding circulation for a rarely used item ties up stock and space for something that may never repeat. The usual split is to run loops for the items that repeat regularly and consume a large share of capacity, and to schedule the long tail conventionally against actual orders. Reviewing which parts belong in which group as the mix changes is part of maintaining the system.

Is an electronic signal better than physical cards?

It depends on what you are trading away. Electronic triggers travel between buildings, cannot be lost in a pocket, and update records automatically, which matters where the supplying process is remote. The cost is visibility: a board full of cards shows anyone walking past how the loop is behaving, whereas an electronic queue is visible only to whoever opens the screen. Where the supplying process is nearby, physical signals usually teach the discipline better, at least while people are learning it.

What do we do when a loop keeps running out?

Resist the reflex to add circulation immediately, because that hides the cause and permanently raises stock. Look first at why replenishment is late: breakdowns at the supplying process, changeovers being deferred, the signal not being collected often enough, or a jump in consumption that nobody flagged. Add circulation as a temporary measure if shipments are at risk, mark it as temporary with an owner and a review, and remove it once the underlying problem has been dealt with.