Lean Foundations — Lesson 6: Pull systems and kanban: produce only what the customer takes

Lean Foundations · 4 min read · 27.10.2026

Sixth lesson of the Lean Basics series. In it, we break down the difference between push and pull production, kanban as a signaling system, and calculating the number of cards. We'll also cover where kanban fails. After this lesson, you'll be able to set up a pilot pull loop on a single product.

Push or pull: who issues the production order

In push production, a monthly plan pushes material forward. Each operation passes finished pieces downstream, even when the next workstation doesn't need them right now. During one audit, we counted 9 days of semi-finished parts between milling and assembly. Nobody had ordered them. They reached the shop floor because the plan pushed them there.

You know the consequence from the first lesson: overproduction, the worst form of waste (muda). It breeds the remaining six — inventory, waiting, transport, defects. In a pull system, the order isn't issued by the plan but by demand. Assembly withdraws from the supermarket — a rack with fixed locations next to the line — exactly as much as it consumes. Only then is replenishment triggered. Nothing is produced in advance “just in case.” It works like self-service: the shelf is restocked only after the customer takes from it.

Pull doesn't mean zero inventory. It means inventory with a hard cap and a clear replenishment signal. Open the VSM from lesson three. The operations with the highest inventory between them are the first candidates for pull — that's where it pays to start.

Kanban doesn't decide how much to produce. It decides when to stop.

Kanban: the card that replaces the order

Kanban means “visible card” in Japanese. The card travels with the container of material. When assembly empties the container, the card isn't thrown away — it returns to the preceding operation as the only authorized production order. We distinguish production kanban, which authorizes production, and withdrawal kanban, which authorizes moving material from the warehouse to the supermarket.

The signal doesn't have to be paper. An empty bin, an andon light, or a message in the ERP — what matters is the rule, not the medium. On one assembly line, we switched material supply to kanban with pickup every two hours. Parts inventory at the line dropped from 2 days to 3 hours.

• Nothing is produced without a kanban.
• Exactly as many pieces are produced as there are returned cards. Not a single piece more.
• A kanban always travels with material, never on its own.
• Only defect-free parts go into the supermarket; defective ones are dealt with at the source.

How many kanbans to put into circulation

The number of kanbans is the inventory cap. It's calculated using the formula:

Number of kanbans = (daily consumption × replenishment lead time × (1 + safety buffer)) ÷ number of pieces per container

Example: a line consumes 480 pcs per day. Replenishment, including waiting for the card, takes 0.5 days. The buffer for breakdowns is 25%, and a container holds 80 pcs. Calculation: (480 × 0.5 × 1.25) ÷ 80 = 3.75. We round up to 4 kanbans — you never round down; kanban is a cap, not a minimum. The inventory cap is therefore 320 pcs, roughly 0.7 days. Before implementation, six days' worth of the same part sat at that very workstation.

We recalculate the number of cards after a month. If a part repeatedly runs out, we don't automatically add kanbans. We look for the root cause: machine downtime, long changeovers, defect rate. Every extra kanban is a gauge of an unsolved problem.

Where kanban fails

In our projects, we see four recurring mistakes:

• Kanban on unstable demand. One-off make-to-order production stays in push. Pull only makes sense for products with regular demand.
• Long changeovers. When a changeover takes three hours, large batches are a reasonable response. That's why you first shorten changeovers as described in lesson four — otherwise the buffer in the kanban eats up all the savings.
• Lost cards. A kanban system without audits rots. Once a week, walk the loop, count the cards, and compare them against the system.
• Piloting everywhere at once. We start with one or two product families for 4 to 6 weeks. We only tackle the whole shop floor once the pilot runs without interventions.

This week's exercise. Pick one part with regular consumption in assembly and set up a two-bin system. Two identical bins at the line, each with a card. When the first one empties, its card goes out as the replenishment signal while you work from the second. For two weeks, record how many times the part ran out and what the average inventory was — the number of full bins at the end of the shift is enough for an estimate. The goal isn't perfection, but the first proof that pull works on your shop floor too.

A pull system relies on reliable machines. An unplanned press breakdown burns through the buffer faster than kanban can save. How to keep equipment predictable is covered in the next lesson on TPM and autonomous maintenance. You'll find the complete overview of the series and lessons in our e-learning offer.

Quiz — test your knowledge