Heijunka / Production Leveling
Heijunka levels production in volume and mix over the period, so operations run in a steady rhythm rather than in bursts — it makes the capacity requirement predictable and is the prerequisite for takt, flow and small inventories.
Consumption data per variant per period (at least 8-12 weeks) has been analysed
The changeover times allow the planned number of switches (SMED is typically completed first)
A small, deliberate finished-goods stock (or order buffer) can absorb the difference between level production and uneven sales
Separate genuine customer variation from self-created variation: campaigns, month-end closings, batching in order intake and “round number” orders often create more swing than the customers themselves. The self-created part is removed at the source — it shouldn’t be levelled, it should go.
Takt = available time ÷ average period demand. Decide the levelling period (day/week) and the target for EPEI — “every part every interval”: how often should each variant be produced? A shorter EPEI = less inventory and faster reaction, but more changeovers.
Distribute the variants in a repeated pattern over the period (e.g. A-B-A-C-A-B-A-D…) with the big runners most frequent. The pattern is visualised in a heijunka box or planning board with fixed time intervals — the plan becomes physical and visible, not a spreadsheet.
The difference between level production and uneven pull is absorbed by a sized finished-goods stock on the runners (controlled with kanban) — not by overtime and panic. The buffer’s size is a calculated decision that is revised regularly.
Plan against actual is tracked daily on the board: each break in the pattern has a cause (breakdown, material shortage, rush order), and the causes are the improvement list. As stability grows, the EPEI is shortened — smaller batches, less inventory.
How it goes wrong in practice — and this is where most implementations part ways.
More changeovers with long changeover times eat the capacity. The order is not optional: short changeover time first, frequent switches afterwards.
If anything can break the pattern, it doesn’t exist. Define one controlled exception path with approval — and measure how often it is used. If it is used daily, the pattern is wrongly sized.
Campaigns and quarter-end sprints recreate the variation production levels. Levelling is a cross-functional decision — without commercial backing it becomes a production-internal fight against its own order book.
The method only works when it becomes daily routines and leadership behaviour. The operations check shows in 4 minutes where your operations stand — and what should come first.