Heijunka

395. Heijunka

Heijunka (平準化, “leveling”): a Toyota Production System technique for smoothing production by mixing products in a level cadence rather than running batches.

Heijunka makes takt time sustainable by absorbing demand fluctuations inside the production system.

395.1. Volume leveling vs mix leveling

Volume leveling: produce roughly constant total output each period (week, day, shift). Inventory absorbs demand variability over time.

Mix leveling: produce a mix of products each period, in repeating sequence, rather than running large batches of one product at a time.

Example: producing A, B, C in ratio 4:2:1 per day.

Without heijunka	With heijunka
AAAA AAAA BB BB C	A B A A C A B A
Long batches; A inventory grows then drains	Steady draw on all parts; minimal inventory swings

395.2. EPEI (Every Part Every Interval)

A key heijunka metric. EPEI = how often the line produces each product.

If $𝑛$ products are scheduled to run, total setup time is $\sum_{𝑖} 𝑠_{𝑖}$ , and available time per interval is $𝑇$ :

EPEI = 𝑇 - \sum_{𝑖} \frac{𝑠_{𝑖}}{production rate}

Smaller EPEI = more frequent mix changes = more flexibility, but more setups.

Reducing setup times (SMED — Single-Minute Exchange of Dies) is the lever for reducing EPEI.

395.3. Heijunka box (load-leveling box)

A physical (or virtual) board that schedules production:

Time slot  ──→
Product
A           [kanban][kanban][kanban][kanban]
B           [kanban]──────[kanban]──────
C           [kanban]──────────────────

Pull cards from left to right in each slot — produce the indicated product. Maintains the mix across time.

395.4. Benefits

Smooth upstream demand — suppliers see steady draws on each part
Low inventory swings — no boom-bust pattern from batch runs
Customer-aligned — production cadence tracks demand
Enables JIT — pull system needs level loading to function

395.5. Costs

More setups — each mix change is a changeover (SMED becomes critical)
Less efficient per setup vs long batches
Worker training — operators must adjust between products frequently
Tooling cost — quick-change fixtures, modular setups

395.6. When heijunka is harder

High product variety with infrequent demand (long-tail SKUs)
Long setup times not yet reduced via SMED
Custom / engineered-to-order — no repeating sequence possible
Volatile demand mix — sequencing must adapt frequently

In those cases, partial heijunka (level the high-volume products, run rest in batches) is a pragmatic compromise.

395.7. Implementation steps

Calculate average demand per product per period
Reduce setup times via SMED
Establish a pitch (small repeating time slot)
Sequence products in the heijunka box per the demand ratio
Pull production based on the box; replenish customer pulls