Heijunka is production leveling: scheduling a fixed period so that volume and product mix are spread evenly across it, instead of building each product in large batches. The line produces small quantities of every product frequently, which smooths the load on people, machines, and suppliers while still matching total demand.
Heijunka exists because customers do not order evenly, but factories run badly when the schedule whipsaws. A plant that chases raw order patterns runs flat out one week and half-idle the next, which creates unevenness (mura) and overburden (muri), and those two breed waste everywhere. Leveling breaks the whipsaw: total demand still gets built, but in a smooth, repeating rhythm. In the Toyota Production System, heijunka is the primary countermeasure to mura and it is the foundation that takt time kanban and stable standard work all sit on (Toyota UK, TPS glossary). It is a pillar concept in lean manufacturing.
What Does Leveling by Volume and by Mix Mean?
Heijunka levels two things, and you need both:
- Leveling by volume means producing at a steady rate based on average demand over the planning period, rather than matching each day's order swings. If customers take 500 units a week in a jagged pattern, the leveled line builds 100 a day, every day, and a small finished-goods buffer absorbs the daily noise.
- Leveling by mix means distributing different products across each day and shift instead of dedicating long runs to one item. Rather than running Monday and Tuesday on product A, Wednesday on B, and Thursday on C, a mix-leveled line cycles A-A-B-A-B-C repeatedly through every day.
Volume leveling stabilizes how much the plant builds; mix leveling stabilizes what upstream processes and suppliers are asked for hour by hour. A line that builds every product every day pulls components evenly, so no upstream process sees a famine followed by a spike.
How Does a Heijunka Box Work?
A heijunka box is the physical scheduling tool that holds the leveled plan. It is a grid of slots, usually a wall-mounted rack: each row is a product or product family, and each column is a short, fixed time interval called a pitch, often 15 to 30 minutes. Production kanban cards sit in the slots. A material handler pulls the cards column by column at the pitch interval and delivers them to the pacemaker process, which builds exactly what the cards say, in that order. The box makes the leveled mix physical and visible: anyone can look at it and see what the line should be building this pitch, what comes next, and whether withdrawal is running ahead of or behind plan. It is visual management for the schedule itself, and it pairs naturally with a kanban pull system, since the cards in the box are the kanban.
What Is EPEI and Why Does It Matter?
EPEI, "every part every interval," measures how leveled your plant actually is. It is the time it takes to cycle through every product in the family once: a line running every SKU every day has an EPEI of one day, while a line that needs three weeks to get through all its SKUs has an EPEI of three weeks. EPEI is governed mostly by changeover time: the interval must be long enough to fit all the changeovers required to run every product once, plus the run time itself. That makes EPEI the honest scoreboard for SMED work: cut changeover times and EPEI shrinks, batches shrink with it, and the schedule can level further. Chasing a smaller EPEI is usually a better framing than "reduce batch sizes," because it points directly at the constraint (changeover cost) instead of at a symptom.
How Do You Implement Heijunka? A 6-Step Sequence
- Pick the product family and calculate takt. Level one pacemaker line first. Use average demand over the planning period, typically two to four weeks, to set takt time.
- Set the leveled volume. Commit the line to a steady daily rate at that takt, and size a small finished-goods buffer to absorb daily order noise for make-to-stock items.
- Build the mix pattern. From each product's share of demand, derive a repeating sequence (for demand of 50 percent A, 33 percent B, 17 percent C: A-B-A-C-A-B). Keep it boring and repeatable.
- Check EPEI against changeover reality. If the pattern demands more changeovers than current setup times allow, run SMED first or start with a coarser pattern and tighten it as changeovers improve.
- Install the heijunka box and pitch. Choose a pitch (often one pack-out quantity's worth of takt), load kanban cards to the pattern, and assign the material-handler route that pulls cards each pitch.
- Track adherence and level further. Measure schedule adherence by pitch, not by day. Misses point at the next problem to fix: a changeover, a supply gap, an unstable cycle. Re-level as demand shifts.
Does Leveling Make You Less Responsive?
It feels like it should: you are deliberately not chasing today's orders. In practice, leveling usually makes plants more responsive, for two reasons. First, mix leveling means every product is in production every day or every few days, so a surprise order for product C waits hours for the next C slot, not a week for the next C campaign. Second, the plant's inventory position improves. When production runs in big batches, finished goods pile up ahead of actual demand. U.S. Census Bureau M3 survey data shows manufacturers' inventories-to-shipments ratios have run around 1.4 to 1.5 in recent years (U.S. Census Bureau, M3 Survey), roughly a month and a half of shipments sitting as inventory across the sector. Leveling attacks the batch-driven share of that: smaller, more frequent runs mean less cash parked in the warehouse and fresher stock when demand shifts. The trade is explicit and managed: a small, sized buffer at the end of the line buys smoothness everywhere upstream.
Where leveling genuinely strains is high-mix, low-volume plants with volatile demand. There, the answer is usually to level what is stable (runners), handle true strangers as a planned capacity slice, and keep the pattern honest rather than pretending every SKU can ride the wheel.
What Does Heijunka Look Like With Live Data?
The classic heijunka box is cardboard and kanban, and it works. What it cannot do is see: whether the line actually hit each pitch, which changeovers ran long, and how demand is drifting against the pattern. Plants that put live factory visibility and AI production scheduling over their existing systems get the same leveling logic with a feedback loop attached: pitch-by-pitch adherence tracked automatically from the floor, and constraint-aware schedules recalculated when demand or downtime moves, instead of waiting for the next planning cycle. The principle stays Toyota's; the clipboard becomes a live signal. No rip-and-replace required to get there, which is exactly how Harmony deploys on running plants.