Heijunka - Leveling by Volume & Mix

by eLe Abllla on !uly 16 2007

Heijunka is one oI those highly useIul Japanese words like mura and muri (see below)-
that have had a hard time Iinding their way into common English usage. (By contrast
muda, kaizen, and kanban seem to have made themselves quite at home.) Yet heijunka
which means leveling is truly critical to creating a lean production system, because it`s
the key to achieving stability.
Perhaps part oI the problem is that when Toyota uses the term heijunka it means two
diIIerent, but related, things. One is the leveling oI production by volume. The other is
leveling production by product type or mix. Let`s take them in turn.
Suppose that a producer has a highly variable level oI orders Ior a Iamily oI products
Ilowing through the same production process. Some weeks customers request 600, some
weeks they request 400, and-within the week demand also varies Irom day to day. One
way to run production which sounds attractive given today`s widespread notion that
producers should be totally and instantly responsive to customers is to continuously vary
output. II the customer wants 600 during a week, produce 600. II the customer requests
120 during a day, produce 120.
Correct? Years ago, Toyota reached the counterintuitive conclusion that this is a bad idea.
Its reasoning was that no production system can be continuously responsive to gyrating
orders without suIIering Irom mura (unevenness in productivity and quality), and muri
(overburden oI machines, managers, and production associates). And mura and muri
together create muda (waste).
Toyota managers concluded that it was much better to calculate long-term demand Ior the
product in question, and to run the production process very smoothly at the level oI long-
term demand. Thus iI demand averages 500 per week and 100 per day, run the process at
the leveled rate oI 500 per week and 100 per day.
To do this while never disappointing the customer-whose demand is variable around the
average-Toyota calculates a standard inventory oI Iinished goods at the end oI the
production process Ior make-to-stock items. (For make-to-order items, it establishes a
standard inventory oI parts just beIore the point oI customization.) The size oI this
inventory is proportional to the degree oI variability (the amplitude) oI customer demand,
the stability oI the production process, and the Irequency oI shipments.
Suppose that demand varies between 400 and 600 units per week, Ior example, while
averaging 500 over the long term. Suppose also that the production process is completely
stable (highly unlikely, but convenient Ior purposes oI this example!), and that goods are
shipped once per week. The minimum standard inventory needed would be 100 units at the
start oI the week and 600 at the end oI the week at the point oI shipment. It would then
revert to 100 units at the beginning oI the next week. And so on.
Obviously, iI the variability in demand is such that several weeks oI 600-unit demand can
occur in a row, the producer would need to decide just how much larger standard inventory
should be so that customer demand could always be met, yet production levels would not
need to vary. Some sort oI statistical analysis oI standard deviations would be required. At
some point-Ior example iI there is signiIicant seasonality in demand-it might be better to
adjust production than to carry larger standard inventories, but Toyota tries very hard to
avoid Irequent changes in output.
By carrying what seems to be extra inventory at the end oI the production process (or at the
point-oI-customization in the case oI make-to-order items), Toyota has Iound that it can
smooth production all the way upstream, and reduce inventories at every juncture between
Ilow and pull along the entire stream. Costs are reduced and total inventories in the value
stream are smaller as well. This is because a seawall oI inventory has been built in the
Iinished-goods area or at the point-oI-customization to protect the production process Irom
giant demand waves and troughs unrelated to the average level oI the sea (that is, the
average level oI demand.)
With heijunka Ior overall demand in place, lean producers are ready to turn to the second
aspect oI heijunka, which is leveling output by product mix. Most value streams will
produce a variety oI products, and the producer Iaces important choices about the
production sequence and mix. Traditionally, when mass producers have Iocused on
utilizing available machine and labor hours, managers have thought it best to group
products by type and to make large batches oI each type. Otherwise, valuable production
time is lost on changeovers Irom one product type to the next. (And remember that the
resulting inventory is counted as an asset so the Iull costs oI this approach are hidden.)
Toyota years ago came to a very diIIerent conclusion. The Iirm Iocused on reducing the
time and cost oI changeovers so that much smaller batches-ideally lots oI one-could be
produced without a severe cost penalty, either due to lost production time or signiIicant
quality problems. Doing this meant that demand Ior parts upstream could also be leveled,
with the beneIit oI reducing throughput/lead time and total inventories along the entire
value stream.
Suppose Ior example that a shirt company oIIers Models A, B, C, and D to the public, and
that weekly demand averages Iive oI Model A, three oI Model B, and two each oI Models
C and D. A mass producer, seeking economies oI scale and wishing to minimize
changeovers between products, would probably build these products in the Iollowing
weekly sequence:
AAAAABBBCCDD
A lean producer, mindIul oI the beneIits oI smoothing output by mix as well as volume,
would strive to build in the repeating sequence:
AABCDAABCDAB
The two approaches are contrasted in the illustration: Heijunka by Product Type.

A typical heijunka box has horizontal rows Ior each member oI a product Iamily, and
vertical columns Ior identical time intervals oI production. Production control kanban are
placed in the slots created, in proportion to the number oI items to be built (Source: Lean
Lexicon)
For lean managers who accept the notion that leveling by volume and mix produces
beneIits throughout the value stream, the problem remains oI how to control production so
that true heijunka is consistently achieved. Toyota came up with a simple answer many
years ago in the Iorm oI the heijunka box (or leveling box.)

A typical heijunka box (see the illustration) has horizontal rows Ior each member oI a
product Iamily, in this case Iive. It has vertical columns Ior identical time intervals oI
production, in this case twenty minutes. Production control kanban are placed in the slots
created, in proportion to the number oI items to be built oI a given product type during a
time interval.
In this example, the shiIt starts at 7 am and kanban are withdrawn bv a material handler
everv twentv minutes Ior distribution to the pacemaker point along the value stream. (In a
lean production system oI this type, there is only one pacemaker point along the value
stream where production instructions are introduced. From that point back up the stream,
parts are replenished at each break in continuous Ilow by means oI simple pull loops Irom
upstream parts supermarkets.) In the Iirst twenty minutes, the value stream will produce
one kanban oI Type A, two kanban oI Type B, one kanban oI Type C, and one kanban oI
type D.
Whereas the slots represent the timing oI material and inIormation Ilow, the kanban in the
slots each represent one pitch oI production Ior one product type. (Pitch is takt time
multiplied by pack-out quantity. This concept is important because it represents the
minimum amount oI material that can be moved Irom one operation to the next, and the
number oI items called Ior by a kanban are sized to this amount.)
In the case oI Product A, the pitch is 20 min and there is one kanban in the slot Ior each
time interval. However, the pitch Ior Product B is 10 min, so there are two kanban in each
slot. Product C has a pitch oI 40 min, so there are kanban in every other slot. Products D
and E share a production process with a pitch oI 20 min and a ratio oI demand Ior Product
D versus Product E oI 2:1. ThereIore, there is a kanban Ior Product D in the Iirst two
intervals oI the shiIt, and a kanban Ior Product E in the third interval, and so on in the same
sequence.
Used as illustrated, the heijunka box consistently levels demand by short time increments,
twenty minutes in this case. This is in contrast to the mass production practice oI releasing
a shiIt, or a day`s, or a week`s worth oI work to the production Iloor. Similarly, the
heijunka box consistently levels demand by mix. For example, it ensures that Product D
and Product E are produced in a steady ratio in small batch sizes.
Production process stability introduced by leveling makes it vastly easier to introduce lean
techniques ranging Irom standard work to continuous Ilow cells. As the mura (unevenness)
and muri (overburden) introduced by traditional production control recede, muda declines
as well.
Recently my colleague Ian Glenday and I have been thinking about ways to extend the
basic heijunka concept to types oI production Toyota does not conduct, speciIically in the
wide range oI industries where technologies require the production oI large batches oI
material. Examples are easy to Iind in the Iood industry (e.g., bottling oI soIt drinks), in
pharmaceuticals, in paint and other coatings, and in chemicals. And as you think about it,
you will see that batches are produced in many other activities as well-Ior example, large
batches oI patients waiting Ior a test procedure in a hospital that is designed to process all
oI the samples at one time in a large machine.
While it would be nice to think that technologies will soon be available so that production
oI soda or pills or paint can be reduced to the small increments handled by traditional
heijunka methods, the reality is diIIerent. The large integrated processes required Ior these
activities, consisting oI many steps that must be tightly linked and which are hard to
changeover quickly, are likely to be with us Ior a long time to come.
It would also be nice to think that dedicated product Iamily value streams can be created to
serve only a small number oI products: Ior example, Ior the Iive part numbers that
constitutes the complete product range running through the heijunka box shown in the
diagram published with this article. But the reality Ior many batch processes is that dozens
or even hundreds oI part numbers must go through a common set oI equipment, and this
situation will be with us Iar into the Iuture, no matter how clever lean thinkers are in
creating 'right-sized` processes.
In observing these large batch production systems in action, the Iirst thing you will note is
that things rarely go as planned. The system encounters technical problems, necessary
inputs Irom an upstream process are missing, or the customer suddenly demands something
totally diIIerent Ior immediate delivery.
The second thing to note-as a direct result oI the Iirst-is that manuIacturing managers spend
much oI their time rushing around to change the production schedule. These changes then
cause schedule changes in all oI the upstream processes. Because nothing is stable, mura,
muri, and muda are everywhere.
What we have discovered is that in these types oI processes a diIIerent approach is required
to break the mindset that believes it`s impossible to get away Irom constant short-term plan
changes, level production, and move steadily toward stable Ilow. The starting point is to
recognize that only a Iew products in any system account Ior most oI the production
volume (typically 6 oI products account Ior 50 oI the volume). It`s then not a big leap
to see that you can quickly establish a stable production schedule Ior these products,
making a given amount oI each oI these products (based on a standard inventory
calculation) during every production cycle. For the moment, the other products will
continue to be scheduled in the traditional manner.
II a shirt company oIIers Models A, B, C, and D, and weekly demand averages Iive oI
Model A, three oI Model B, and two each oI Models C and D, a mass producer, seeking
economies oI scale and wishing to minimize changeovers between products, would
probably build these products in the weekly sequence illustrated. A lean producer seeking
to smooth output by mix as well as volume would strive to build in the repeating sequence
shown. (Source: Lean Lexicon)
By conducting a sieve analysis to identiIy these products and put them on a repetitive
production schedule that is unnecessary to change Ior extended periods, it`s possible to
stabilize the great bulk oI production. And by accomplishing this stabilization, it is also
possible to gain a major beneIit in the Iorm oI what we call 'economies oI repetition. It
turns out that iI the same setup sequence is used every time Ior runs oI the same length Ior
each product, the production team is able to reduce setup times, deal with maintenance and
materials issues, and adhere to standard work to an extent that is simply impossible when
the production sequence is constantly changing, and run lengths Ior each product are also
constantly varying. This situation also provides the ideal conditions to begin linking all the
production steps Ior these products to create a true end-to-end Ilow through the plant.
Indeed, once end-to-end Ilow is established, it`s also possible to pull materials Irom
suppliers Ior these products in the same regular and Irequent quantities.
Then, as time goes on and progress is made in reducing changeover times, it is possible to
incorporate more and more products into the steady sequence oI 'each product each cycle,
until practically the entire output is leveled. In addition, the Irequency oI the production
cycle also increases Irom 'every product every month to twice a month, to weekly, and
then daily. And, as the production team`s experience grows, it`s also possible to vary the
volumes and the cycle to more closely match the pattern oI demand. This is about as close
as you can get in process industries to leveled production and one-piece Ilow.
As we see the state oI play in production activities across the industrial landscape, we`ve
made great progress in eliminating muda in the Iorm oI unnecessary steps and unnecessary
handoIIs; Ior example, the elimination oI intermediate inventories and extra handling,
which disappear when machines are relocated in cells. At the same time, we`ve made very
little progress in eliminating mura and muri, which most Toyota veterans see as the bigger
problem.
This situation explains why it is so important Ior production proIessionals to embrace
heijunka, so that every process can be leveled by volume and mix. Doing so creates a
diIIerent world Ior employees-who are no longer overburdened; Ior customers-who get
better products on the date promised; and Ior manuIacturers-who get to keep the money
saved when muda, mura, and muri are all reduced.