Heijunka (Production Levelling): A Practical How-To Guide
In Japanese, heijunka means to make flat or level. In Lean manufacturing, it means spreading production volume and product mix evenly across a planning period rather than building in batches driven by the order in which customer orders happen to arrive. It sounds simple — almost too simple — yet most organisations that claim to practise Lean never actually achieve it, because they have not addressed the real problem first: the way demand volatility propagates through every upstream process.
By 2022 most manufacturers and service delivery teams had lived through two years of demand chaos. Heijunka is not a cure for a broken supply chain, but it is the discipline that makes a stabilised supply chain run as efficiently as it possibly can. Understanding it properly — and applying it — is the difference between a Lean programme that delivers sustainable gains and one that just moves the piles around.
Why demand volatility destroys Lean
When production follows raw customer-order patterns, three wastes appear almost automatically:
Overproduction — teams race to fill a spike, building far more than can be consumed before the next trough.
Excess inventory — the buffer that accumulates because nobody trusts demand will stay flat.
Overtime and undertime — people are stretched thin one week and underutilised the next, making capacity planning a permanent crisis.
The Toyota Production System addressed this by inserting a levelling step between actual customer orders and the production schedule. The production schedule sees a smooth, averaged demand signal; the swings are absorbed upstream, in finished-goods inventory or by managing the order-release pace. That absorbing layer is what heijunka creates.
How heijunka works in practice
The physical tool most associated with heijunka is the heijunka box (also called a production-levelling board). It is a grid: columns represent time slots called pitch intervals — typically a fraction of a shift — and rows represent product types or work categories. Kanban cards, one per unit of work, are loaded into the slots in a sequence that repeats the product mix in every pitch interval rather than running all of Product A before starting Product B.
Calculate your takt time. Takt time is available production time divided by customer demand rate. It sets the heartbeat of the system. Everything downstream of heijunka is designed around this rhythm.
Define the pitch interval. The pitch is the smallest meaningful batch you can release. A common choice is the pack-out quantity — the number of units that fills one standard container — multiplied by takt time. Shorter pitches mean faster feedback; longer pitches mean less changeover pressure.
Design a mixed-model sequence. Rather than running a full day of Product A followed by a full day of Product B, sequence them in proportion to demand within each pitch. If A outsells B two-to-one, the repeating cycle might be A–A–B. The goal is every product, every pitch — or as close as changeover time allows.
Load the heijunka box. Place cards for the pitch's planned mix into the corresponding column. A material handler or pacemaker process pulls cards on the pitch interval and releases that exact mix to production.
Respond to deviation, not to forecasts. When actual demand deviates from the averaged plan, adjust the supermarket inventory that absorbs the swing — do not blow up the production schedule. The stability of the schedule is the point.
Heijunka beyond the factory floor
The principles translate directly to service environments. In case processing — tax assessments, permit applications, insurance claims — the equivalent of a heijunka box is a controlled queue with a fixed release cadence. Rather than processing files in the order they arrive (first-in, first-out raw), a levelling step groups them by type and releases a proportional mix to the processing team in each time block. The result is stable workloads, more predictable cycle times, and the ability to staff to a real average rather than to a panic peak.
The preconditions are the same: you need a buffer (a controlled intake queue), a defined pitch, and the discipline to not bypass the levelling step when a large batch arrives. That last point is where most service organisations fail. When the large file drops, the temptation is to drop everything and process it. Heijunka says: put it in the queue. Pull on the cadence. Trust the system.
Heijunka also has real limits. It works best when setup or changeover times are short, when demand is genuinely forecastable over the levelling horizon, and when you have the buffer inventory or queue capacity to absorb the gap between actual orders and the levelled schedule. If your changeovers take half a shift, mixed-model sequencing may cost more than it saves. Reduce changeover time first — that is a separate improvement project, often using SMED — then introduce levelling.
If your organisation is ready to move beyond firefighting and build the stable operating rhythm that makes every other Lean tool work, XNM's strategic advisory can help you design and implement a production levelling system that fits your context.