What is a kanban card?

A physical (or electronic) signal — typically a laminated card riding with a container of parts — that authorizes exactly one action: make one container of one part, or move one container from a store. It carries the part number, the container quantity, the producing process, and the store address. In a pull system, no station produces or moves anything without a card, so the number of cards in circulation caps the inventory in the system by construction.

How do you calculate the number of kanban cards?

The classic formula is K = demand per period × replenishment lead time × (1 + safety factor) ÷ container quantity, rounded up. For example: 120 parts a day, half a day to replenish, 10% safety, and 20 parts per container gives 3.3 — so 4 cards. The number is a starting point, not an answer: too few cards cause stockouts, too many just hold inventory, and mature systems remove cards deliberately, one at a time, to expose the next problem.

How is kanban different from a push or MRP schedule?

A push system decides what each station makes from a forecast, computed in advance — every station produces because the schedule says so, and forecast error accumulates as inventory between stages. In kanban, actual consumption is the schedule: a station only makes what was just withdrawn, so work-in-process is capped at the card count and lead time stays short. MRP still has a role for planning materials and capacity; kanban replaces it at the level of hour-by-hour execution for repetitive parts.

Does kanban work for software and office work?

The software 'kanban board' borrows the core idea — visualize work and cap work-in-process — and it genuinely helps, but it is a cousin, not the same system. Manufacturing kanban is a replenishment loop for standard, repeating items held in a store; a card is a production order for a specific part. A ticket column with a WIP limit caps concurrency but replenishes nothing. The shared lesson is the WIP cap: less work in process means shorter lead times, in code as on the floor.

When is kanban the wrong tool?

When there is nothing standard to replenish: one-off engineer-to-order products, wildly unstable demand that would need the card count resized every week, very expensive or short-shelf-life items where even one container of standing inventory is too much, or processes so unreliable the loop would need absurd card counts to survive. In those cases sequence work directly, level demand first (heijunka), or stabilize the process — then introduce cards where repetition exists.

▮ Lean Fundamentals · A Pull System You Can Run ▮ ▮

How Kanban
Actually Works

A kanban card spends its life attached to a container — and the moment the first part is used, it comes off, and that is the reorder. Watch both kinds of card circulate through a living three-stage line, break the rule and watch inventory pile up, then run the experiments yourself.

KAIZUMI · JULY 2026

Scroll

Every factory has to answer one question at every workstation, all day long: should I make the next one? A push system answers with a schedule, computed days in advance from a forecast. A pull system answers with a card. When a container of finished parts is withdrawn from a store, the card that was attached to it comes off and travels back to the process that made it — and only then does that process make one more.

That card is a kanban (Japanese for “signboard”), and the loop it travels is the mechanism behind just-in-time production. It sounds too simple to matter. The point of this guide is to let you watch it matter — and then break it, starve it, and resize it yourself.

The Card

Physically, a kanban is unglamorous: a laminated card (or a barcode label, or an empty bin) that rides with a container of parts. Everything it needs to say fits in five fields:

One card = permission to produce one container of one part. The card carries the entire production-control system on a piece of laminated paper.

Two kinds of card circulate in a full system, and both spend most of their lives attached to containers. A production kanban rides each full container in the producing process's store; it detaches the moment that container is claimed, and the detached card is the instruction to make a replacement. A withdrawal kanban rides each full container at the consuming process's lineside; it detaches the moment the operator uses the first part — that detach is the reorder point — and a material handler carries it back to the store to fetch the next container. Each production kanban circulates only within its own process loop; the withdrawal kanban link the loops together. That's better watched than read. Scroll.

The Loop, Running

The lineThree stages, running live: machining fills a parts supermarket, assembly — the pacemaker — fills a finished-goods store, and a customer buys on a steady pitch. Look closely: every container on a shelf wears a production kanban, and every container at assembly's lineside wears a teal withdrawal kanban. The blue cards in the heijunka box are today's schedule. Nothing here is scripted — it's a real simulation, and it never stops.

The pacemaker loopEach morning, production control loads the heijunka box with the day's leveled schedule; the box releases one card per pitch — that card is assembly's instruction to build one container, and it rides the finished container onto the shelf. When a customer order takes a container, the attached card detaches and flows back to production control, informing tomorrow's load. One-way box: scheduled in, released out. Cards never return to it mid-day.

Consumption is the triggerNow the parts loop. The moment assembly uses the first part from a lineside container, its teal withdrawal kanban drops into the line collection box — that detach is the reorder point. A handler collects the box on a timed route, places each card on a full container at the supermarket — knocking that container's amber production kanban into the store's own collection box — and conveys the carded container back lineside. Machining's handler returns the amber cards to machining: its instruction to produce.

Read the numbersNow the readouts. Every container in the system wears a card, so WIP can never exceed 13 containers — one per card in circulation: three amber and seven blue production kanban plus three teal withdrawal kanban. Cards cap inventory by construction, the way a fixed number of plates caps a buffet line. And because nothing waits long on a shelf, lead time stays short and flat.

Now break the ruleTake all the cards away and run the same line on a push schedule — a forecast that releases work to machining on its own clock, slightly faster than the customer actually buys. Each station now makes and moves parts because the schedule says so, not because anything was consumed.

Watch it pile upNobody is idle, every station looks productive — and the shelves quietly fill with containers no one asked for. WIP climbs without limit, and lead time balloons with it, because every new part joins the back of an ever-longer queue. This is overproduction: the waste that hides all the others.

Put the cards backReturn the kanban. The surplus containers drain off first — inventory with no card attached never authorizes anything — and the system settles back under its 13-container cap. Same machines, same demand. The only difference is who's allowed to say “make one more.”

If the pile-up felt familiar, it's the same physics as batch versus flow: inventory between stages is where time, cash, and defects hide. Kanban is how a line holds the pull discipline automatically — no expediter, no morning replan, no heroics. And the lead-time readout you just watched balloon is exactly the clock explained in cycle time vs takt time vs lead time: more WIP is more lead time.

Run the Experiments

A pull system's behavior lives in a few dials: how many cards circulate, how fast — and how evenly — the customer buys, and whether the stations hold up. This copy of the line is yours — starve it, flood it, break it.

Try it yourself

Mode

Kanban / parts loop3

Orders / min15

Orders

WIP9

Service level100%

Avg lead time—

Missed orders0

The slider sets the parts loop: that many production kanban at machining and withdrawal kanban at the line (assembly's pacemaker loop keeps a day's schedule of cards on top). Things to try: drop to 1 and watch orders start missing (the loop can't replenish fast enough) · raise it to 8 and watch inventory sit idle on the shelves · flip orders to Lumpy and watch the heijunka box keep releasing one card per pitch — the bursts hit the store, not the line · Break machining in pull mode and notice the line starves gracefully — missed orders, but no pile · do the same in push mode and watch the supermarket flood.

Notice what a breakdown does not do in pull mode: it doesn't bury the plant in half-finished work. The system starves — visibly, immediately, at the finished-goods shelf — which is uncomfortable and exactly the point. Kanban converts problems you'd discover in next month's inventory count into problems you can see from the aisle today.

How Many Cards?

You just felt the trade-off on the slider: too few cards and the loop can't replenish before the customer returns — stockouts; too many and containers sit idle — inventory. The classic sizing formula splits the difference:

K = demand × lead time × (1 + safety) ÷ container qty

Demand / day

Lead time (days)

Safety (%)

Parts / container

4kanban cards (3.3 rounded up)

A worked hypothetical, not a benchmark: 120 parts a day, half a day to replenish, 10% safety, 20 parts per container → 3.3, rounded up to 4 cards. Round up — a fractional card can't circulate. Then treat the number as a starting point and remove cards one at a time to surface problems.

The Rules

Kanban only works as a discipline. Toyota's six rules — paraphrased — are what keep the cards honest:

RULE 1

Downstream withdraws only what it needs

No grabbing extra “while you're there.” A withdrawal without a kanban is theft from the system's information.

RULE 2

Upstream produces only what was withdrawn

In the quantity and sequence the cards arrive. The card queue is the schedule.

RULE 3

Nothing moves or is made without a card

No card, no production, no conveyance. The moment exceptions are tolerated, the WIP cap you watched above silently disappears.

RULE 4

Never pass on a defect

A container on a store shelf is a promise. Send a defective container downstream and pull delivers the defect at full speed.

RULE 5

Level the demand on the system

Kanban assumes a reasonably steady beat — that's why production control loads the pacemaker's schedule into a heijunka box each day and releases it one card per pitch. Lumpy schedules whipsaw the loops; see heijunka.

RULE 6

Reduce the number of cards over time

Each card you remove lowers the water and exposes a rock — a breakdown, a long changeover, a quality problem. Fix it, then remove another.

When Kanban Is the Wrong Tool

Honest limits, because kanban has them. A store only makes sense for parts you'll make again.

Unstable or erratic demand. The card count is sized to a demand rate. If demand swings wildly, the loop is always either starving or drowning — recompute the formula above with a demand number that doubles month to month and watch K thrash. Level first (heijunka), or use kanban only for the stable runners and schedule the erratic movers conventionally.

One-off and engineer-to-order products. A kanban replenishes a standard item a standard store will hold. A product built once, to one customer's print, has nothing to replenish — sequence it directly instead.

Very expensive, bulky, or short-shelf-life items. Every card is deliberate standing inventory. For a part that costs a fortune, fills a bay, or expires, even one container on a shelf may be worse than making to order and accepting the lead time.

Deeply unreliable processes. Rule 6 cuts both ways: kanban exposes instability, it doesn't fix it. If a process breaks down half the time, a pull loop sized to survive it needs so many cards it stops teaching anything. Stabilize enough to run, then let the cards drive the rest of the improvement.

None of these are failures of the idea — they're the idea's edges. Inside those edges, a few laminated cards replace a planning department, cap your inventory by construction, and turn every problem into something you can see from the floor. That's a strong return on paper and a laminator.

Good to know

Frequently asked

What is a kanban card?: A physical (or electronic) signal — typically a laminated card riding with a container of parts — that authorizes exactly one action: make one container of one part, or move one container from a store. It carries the part number, the container quantity, the producing process, and the store address. In a pull system, no station produces or moves anything without a card, so the number of cards in circulation caps the inventory in the system by construction.
What is the difference between a production kanban and a withdrawal kanban?: Both spend most of their lives attached to containers. A production kanban rides each full container in the producing process's store; it detaches into a collection box the moment that container is withdrawn, and a material handler returns it to the producing process as its instruction to make one replacement — it circulates only within that process's loop. A withdrawal kanban rides each full container at the consuming process's point of use; when the operator uses the first part in the container, the card comes off into a collection box at the line — that consumption is the reorder point — and a handler carries it to the upstream store, places it on a full container, and brings that container back. Together they link stores and stations into loops, so one real customer withdrawal cascades upstream card by card.
How do you calculate the number of kanban cards?: The classic formula is K = demand per period × replenishment lead time × (1 + safety factor) ÷ container quantity, rounded up. For example: 120 parts a day, half a day to replenish, 10% safety, and 20 parts per container gives 3.3 — so 4 cards. The number is a starting point, not an answer: too few cards cause stockouts, too many just hold inventory, and mature systems remove cards deliberately, one at a time, to expose the next problem.
How is kanban different from a push or MRP schedule?: A push system decides what each station makes from a forecast, computed in advance — every station produces because the schedule says so, and forecast error accumulates as inventory between stages. In kanban, actual consumption is the schedule: a station only makes what was just withdrawn, so work-in-process is capped at the card count and lead time stays short. MRP still has a role for planning materials and capacity; kanban replaces it at the level of hour-by-hour execution for repetitive parts.
Does kanban work for software and office work?: The software 'kanban board' borrows the core idea — visualize work and cap work-in-process — and it genuinely helps, but it is a cousin, not the same system. Manufacturing kanban is a replenishment loop for standard, repeating items held in a store; a card is a production order for a specific part. A ticket column with a WIP limit caps concurrency but replenishes nothing. The shared lesson is the WIP cap: less work in process means shorter lead times, in code as on the floor.
When is kanban the wrong tool?: When there is nothing standard to replenish: one-off engineer-to-order products, wildly unstable demand that would need the card count resized every week, very expensive or short-shelf-life items where even one container of standing inventory is too much, or processes so unreliable the loop would need absurd card counts to survive. In those cases sequence work directly, level demand first (heijunka), or stabilize the process — then introduce cards where repetition exists.

Kanban Pull system Supermarket Heijunka: level the mix Batch vs. flow Kanban in manufacturing Kanban in a warehouse Kanban calculator

Matthew Savas

Founder of Kaizumi, an AI-powered Lean training platform. More about Matthew →

Updated July 2026 · The line, its stations, and every number on this page are a teaching simulation — cycle times, card counts, and the compressed “day” are illustrative, not benchmarks from any real plant. The mechanism follows Ohno, Toyota Production System (Productivity Press, 1988) and the two-card pull discipline as defined in the Lean Enterprise Institute's Lean Lexicon.