Heijunka: Why Most Plants Can't Actually Do It

Heijunka is one of the Toyota Production System concepts that almost every plant manager in mid-market manufacturing has heard of, can define correctly when asked, and does not actually practice. The gap between citation and implementation is wider for Heijunka than for almost any other lean concept I am familiar with — wider than for Kanban, wider than for 5S, wider than for standardised work. There is a structural reason for that gap, and it is the part most articles on the topic skip. This article is about the gap, the reason, and what it would actually take to close it in a plant that wants to.

I have been around manufacturing in one form or another since 1989, and I have visited a great many plants in that time. The pattern around Heijunka has been consistent for as long as I have been paying attention. Plants describe themselves as "running a Heijunka system" when they mean something quite different: producing in alternating campaigns of one product family, then another, with changeover events spaced hours or days apart. That is not Heijunka. That is campaign production with rotation. The distinction matters because the benefits of Heijunka — the ones that make it worth the effort in the first place — only appear when the leveling is fine-grained enough to genuinely smooth demand within a shift, not across a week.

What Heijunka actually is

Heijunka is the Japanese word for leveling, and in the Toyota Production System it refers to the practice of producing a smoothed mix of products in small, repeating sequences, rather than in large batches of a single variant. The intent is to absorb demand variability inside the production system instead of pushing it onto upstream suppliers, downstream customers, or downstream inventory. A Heijunka-leveled line that has to deliver four product variants in roughly equal weekly volumes will produce them in a repeating sequence — A-B-C-D, A-B-C-D — with the cycle of that sequence measured in minutes or at most a small number of hours, not in shifts or days.

The benefits, when Heijunka is genuinely practised, are well established: smaller finished-goods buffers, more even loading on upstream supply chains, faster detection of quality problems (because every variant is being made all the time, not in a campaign two weeks ago), more predictable workload for operators, and better resilience to demand shifts. Toyota did not develop the technique because it sounded elegant. They developed it because the alternative — campaign production with large batches — produces a long list of downstream costs that only become visible when you start trying to eliminate them.

Why mid-market plants do not actually practise it

The reason most plants that cite Heijunka do not actually practise it is not lack of intent and not lack of training. It is that the practice has structural preconditions which most plants have never met, and those preconditions are expensive and slow to build. The plants that say they "do Heijunka" almost always mean they have a Heijunka board hanging in the planning office. The board is a useful artifact. It is not the practice.

The three preconditions that have to be true first

Looking back over the plants I have seen attempt Heijunka — and the much smaller number that have made it stick — the difference between the two groups is not better lean training, more committed leadership, or a more sophisticated sequencing algorithm. It is whether three things are true about the production system before Heijunka is attempted. None of the three is about Heijunka itself. They are all about what has to exist underneath it:

1. Setup time below the production interval of the smallest variant. This is the SMED precondition, and it is the one that is most often missing. The lean canon names ten minutes as the target for single-minute changeover, and that target is not arbitrary — it is roughly the threshold at which mixed-model leveling becomes economically possible at the granularity Heijunka requires. If your changeover is forty minutes and your shortest variant production interval is twenty minutes, no leveling logic in the world will make Heijunka work. The work that has to happen first is SMED, and SMED is months of patient analysis, fixture redesign, and operator practice. There is no software shortcut.
2. Reliable, real-time demand signal that updates faster than the leveling cycle. Heijunka levels production against actual customer pull, not against a frozen monthly forecast. If the demand input that drives the leveling is updated weekly, the leveling cycle cannot be tighter than weekly. Most plants attempting Heijunka are leveling against a forecast that updates on a monthly planning cadence, which means the "leveling" is leveling against an imagined demand pattern that has already drifted from reality by the time it reaches the line. The fix is integration with the live order book, not better forecasting.
3. Equipment availability that does not break the sequence. Heijunka assumes the line will execute the planned sequence reliably enough that disruptions are the exception, not the rule. In a plant where unplanned downtime regularly removes one of the variants from the achievable mix, the Heijunka plan is fiction within the first hour of the shift, and the line falls back into reactive mode — which usually means whatever variant is easiest to make next, which usually means batching by variant. The precondition here is whatever combination of TPM, machine reliability, and rapid recovery the plant needs to keep the sequence executable. This is also not a software problem in the first instance.

Plants that try to practise Heijunka without these three preconditions in place do not get partial Heijunka. They get the appearance of Heijunka — the board, the planning meeting, the vocabulary — and the substance of campaign production. The combination is worse than honest campaign production, because it carries the overhead of the leveling discipline without the benefits the discipline is supposed to produce.

What to do when the preconditions are not yet met

The honest answer to "should we implement Heijunka" in a plant where the preconditions are not yet met is "no — first do the work the preconditions require." That work is usually some combination of SMED programmes targeting the longest changeovers, ERP-MES integration that gets the live demand signal onto the line, and reliability work on whichever equipment is the most frequent source of unplanned sequence disruption. None of this is glamorous. All of it produces value independently of whether Heijunka is ever attempted, which is what makes it the right next step for most plants.

The plants I have seen successfully transition into genuine Heijunka have almost always done these three workstreams first, often over a period of one to three years, and then the Heijunka implementation itself was comparatively quick — because by the time it was attempted, the system was actually capable of executing it. The order matters. Trying to sequence the practice before the preconditions are in place is the most common reason Heijunka initiatives in mid-market manufacturing quietly stop being mentioned after the first few months.

Where Heijunka does work — and where the leveling lives

Heijunka is most successful in environments that combine three characteristics: a stable product family with recurring demand across a manageable number of variants; a production process whose technology naturally supports rapid changeover (assembly lines, electronics SMT, packaging operations); and a customer relationship in which the leveled output matches what the customer actually wants, rather than the customer expecting custom batches on demand. Where those three characteristics are present, Heijunka pays for itself reasonably quickly. Where they are absent — high-mix low-volume project work, processes with structurally long changeovers like heat treatment or curing, customers who expect surge response — Heijunka is the wrong tool, and the right answer is to be honest about that and use a different scheduling logic.

The single most damaging pattern I have seen is plants attempting to apply Heijunka uniformly across all production areas because it is the lean-canonical answer, when half their production is structurally unsuitable for it. The areas where it fits should use it; the areas where it doesn't should use whatever logic actually fits — campaigns, make-to-order, batch sequencing optimised for tooling life. There is no rule in the Toyota Production System that says one scheduling logic has to apply everywhere.

In the SYMESTIC product set, the modules that touch Heijunka most directly are Production Planning and Production Control — the planning module is where the leveled sequence is computed against the live demand signal pulled from the ERP order book, and the control module is where the sequence becomes the executable instruction on the line. Process Data closes the loop by capturing actual cycle and changeover times against the planned sequence, so the leveling logic can be re-tuned against reality rather than against assumptions. None of this addresses the deeper preconditions — SMED, reliable demand integration, equipment availability — which have to be built independently and usually take longer than the software work. But once those preconditions are in place, the technical layer of running an honest Heijunka system is largely a question of having planning, control, and process-data modules that can talk to each other in real time. That has been a solved problem for some years now, and is not where Heijunka initiatives fail. They fail earlier, on the shopfloor, on changeovers and reliability — which is where the work that matters has to happen.

About the author

Uwe Kobbert

Founder & CEO of SYMESTIC GmbH. In manufacturing since 1989 — first at SAS, then STERIA. Founded SYMESTIC in 1995 in Dossenheim. Today: Cloud-native MES platform with 15,000+ connected machines across 18 countries on four continents. Bootstrapped, no external investors. Dipl.-Ing. Communications Engineering / Electronics. · LinkedIn