CPG Scheduler
All posts

How Many Kanban Cards? Sizing the CPG Supermarket Without Starving or Flooding the Line

par levelsinventory planningschedulingstockoutschangeovers

The idling pack line nobody planned for

Picture two failures that look opposite but share one root cause. In the first, your pack-out line goes quiet at 10:40 a.m. because the bulk supermarket — the buffer of semi-finished syrup, dough, or batter between mixing and packing — ran dry. Operators stand around, OEE craters, and the shift misses its number. In the second, the same buffer is piled so deep that totes of work-in-process age past their internal hold time and get dumped. Cash and product, written off.

Both are the same problem: the number of kanban cards in that replenishment loop was set wrong. Turn the dial too low and you starve the line. Turn it too high and you drown it in working capital and spoilage. This post is about how to set that dial on purpose — with a formula to start, and a discipline to keep tightening it.

What a kanban loop actually is

Kanban is a demand-driven, visual pull signal from the Toyota Production System, developed under Taiichi Ohno. Instead of pushing product forward on a forecast, a kanban authorizes an upstream step to make or move something only when a downstream step has consumed it. That consumption-triggered replenishment is the mechanism that right-sizes inventory and exposes bottlenecks, because the cards physically cap how much can exist in the loop.

Two card types matter on a line. A Production (P) kanban authorizes a workstation to make a fixed amount — one container's worth. A Transportation (T) kanban authorizes moving a full container to the downstream process. In a two-card loop, every container carries a card; when the downstream step empties a container, the card is freed and travels back upstream as the signal to replenish. The loop closes when the empty returns.

This is a different scope than a whole-line WIP cap. CONWIP caps total work-in-process across an entire line with one card type. A kanban loop sizes one SKU's buffer between two specific steps — the supermarket that decouples making from packing. If you've read the CONWIP piece, think of this as zooming in on a single link in that chain.

The sizing formula

Here's the core arithmetic in operator terms. The number of cards is set from three inputs: average demand over the replenishment lead time, a small buffer for variability, and the container (bin) size.

                 (average demand during lead time) + safety buffer
# of cards  =  ---------------------------------------------------------
                            container quantity

Every card equals one container, so the number of cards is the amount of inventory you've authorized to sit in the loop.

A worked CPG example

Say a flavored-beverage pack line pulls bulk syrup from a mixing supermarket.

  • Average demand: the pack line consumes 1,200 liters of syrup per hour when running.
  • Replenishment lead time: this is the full round trip for a signal, not just mixing time. Mixing a batch takes 2 hours, the changeover to that flavor takes 1 hour, transport and queue add 1 hour. Total lead time = 4 hours.
  • Demand during lead time: 1,200 L/hr × 4 hr = 4,800 L.
  • Safety buffer: demand and lead time wobble, so add 25% = 1,200 L.
  • Container size: each tote holds 600 L.

Plugging in:

# cards = (4,800 + 1,200) / 600 = 10 cards

Ten kanban cards authorize ten totes — 6,000 liters — to circulate in that loop. That is your supermarket size, derived from floor measurements rather than a guess.

The discipline here is that every input is something you can measure on the floor: clock the consumption rate, time the full replenishment cycle (including changeover and queue), and read the tote capacity off the spec. No input is a vibe.

Reading the dial: what each input does

Lead time dominates — and changeover hides inside it. The single biggest lever on card count is usually replenishment lead time, and a surprising chunk of that is setup. If the flavor changeover in the example drops from 1 hour to 15 minutes, lead time falls from 4 hours to 3.25, demand-during-lead-time falls to 3,900 L, and you can pull a card out. This is why SMED changeover reduction isn't just a line-speed project — it directly shrinks the buffer you have to carry.

The buffer term tracks variability, not average volume. The safety portion exists to absorb demand spikes and lead-time stretch. The more erratic either one is, the fatter the buffer must be. That's the same logic behind the safety-stock combined formula, and the variability penalty is exactly what Kingman's equation predicts: high utilization plus high variability inflates queue time, which inflates lead time, which inflates card count. Reduce variability and the whole chain deflates.

Container size trades smoothness for card count. Bigger bins mean fewer cards but chunkier flow — the line moves in bigger, lumpier steps. Smaller bins smooth the pull and shorten the time between replenishment signals, but add handling and more cards to manage. There's no universal right answer; it's a flow-versus-handling trade you tune.

The governing principle, stated plainly in the kanban literature: the number of cards is the inventory lever. When supply response can't keep up with demand swings and stockouts threaten, a stock build may be appropriate, and it's achieved by placing more kanban in the system. Fewer cards means leaner WIP but higher starvation risk; more cards means more protection but more cash tied up and more shelf-life exposure.

The six rules that keep the loop honest

Toyota's six rules are what separate a real pull system from a pile of cards. Restated for a CPG floor:

  1. The downstream step pulls only what it has consumed — consumption triggers the signal, nothing else.
  2. The upstream step produces only the amount the kanban requests — no "while we're at it, let's run extra."
  3. Nothing is made or moved without a kanban — no card, no action.
  4. The kanban stays attached to the product — the card travels with the tote so the signal never gets lost.
  5. No defects are passed forward — a bad batch doesn't get a card to move downstream.
  6. Reduce the number of kanban over time — the count is meant to fall, not sit still.

Break rule 2 or rule 3 — run extra "to be safe," or move product without a card — and you've quietly converted your pull system back into push. The cards stop meaning anything, inventory balloons, and you're back to firefighting stockouts while sitting on excess. The discipline is the system.

The formula is a starting line, not a finish

Here's the part most how-to guides skip. The lean counterpoint, argued forcefully by the Lean Enterprise Institute's Michael Ballé, is that there is no magic across-the-board formula for supermarket sizing. Supermarkets exist only because of real, fixable gaps: demand surprises, insufficient equipment uptime, slow changeovers, and large batch sizes. Each kanban card is covering for one of those weaknesses.

So the right practice isn't to compute the count once and file it. It's a monthly, line-by-line review where you treat the buffer as living — challenge each one, ask which weakness it's covering, and aim to cut it roughly in half. The supermarket size becomes a driver for kaizen rather than a set-and-forget calculation. The formula gives you a defensible starting number; the monthly review is where the real work happens.

Shrinking the card count

If rule six says reduce cards over time, how? You attack the inputs.

  • Lift uptime. As Total Productive Maintenance raises OEE and schedulable run rate, replenishment becomes more reliable, the buffer covering breakdowns shrinks, and you can pull a card.
  • Compress changeovers. Every minute SMED removes from setup comes straight out of lead time.
  • Cut batch size. Smaller, more frequent runs mean the supermarket needs less depth to bridge the gap between replenishments.

There's a shelf-life dividend too. Fewer cards mean fresher WIP and less aged product sitting in the buffer, which lowers expiry write-offs — directly relevant if you run shelf-life-constrained schedules.

A simple cadence works: measure the current service level, remove one card, watch for starvation over a few cycles, then lock it in or restore it. One card at a time keeps the experiment safe and the learning clear.

Where kanban sizing fits the scheduler's toolkit

Kanban loops live at the decoupling point — the boundary between make-to-stock and make-to-order described in the push/pull decoupling-point piece. The supermarket is where upstream pull meets downstream flow.

The same logic extends upstream to supplier and inbound kanban for components and packaging — caps, labels, closures, films. The card carries the vendor, pack size, lead time, and safety factor, and it fires when a bin hits its reorder point, exactly like a shop-floor loop but pointed at a supplier.

And kanban sizing solves only half the problem. The card count sets the buffer — how much sits between steps. It doesn't set the rhythm — how often you make each SKU. For that, pair it with EPEI / the production wheel and heijunka leveling. Buffer and rhythm together make the pull system work.

Key takeaways

The four numbers you need before sizing a loop:

  1. Average consumption rate of the downstream step.
  2. Full replenishment lead time — make time + changeover + transport + queue.
  3. A buffer percentage that reflects demand and lead-time variability.
  4. Container (bin) size.

Red flags you have too few cards: recurring line starvation, expediting, operators hoarding product off the books to cover themselves.

Red flags you have too many cards: aging WIP, expiry write-offs, no visible pressure to improve — the buffer is hiding every weakness instead of exposing it.

The monthly kaizen question: which card can we remove next, and what weakness do we have to fix to make that safe?

The formula gives you a number. The six rules keep it honest. The monthly review — and the willingness to pull cards out — is what turns a static buffer into a system that gets leaner every quarter.

CPG Scheduler computes and continuously re-sizes kanban and par buffers from live run-rates and lead-time variability, so your card counts track the actual floor instead of a once-a-year spreadsheet. See how the agentic approach works or stop by /for-agents.

Sources

More from the journal

How Many Kanban Cards? Sizing the CPG Supermarket Without Starving or Flooding the Line · CPG Scheduler