See the
Relevant
Traits
Better inventory management with intelligent demand segmentation
By Shawn Kaul
Y
ou’ve been on your lean journey for at least a year. You’ve
focused on the five Ss; implemented production cells; and whittled
down your setups by 50, 75, maybe
even 80 percent. Now is the time to
figure out how to manage your raw and
outside-processed inventory as they
come in via countless suppliers.
At this point, most of you probably
find yourselves with too many part
numbers to manage and quantities that
are all across the board. You manage
all stockkeeping units (SKUs) using
a single method, and you trust the
forecast—buy what it says you need
and attempt to predict any spikes or
anomalies in customer order patterns.
All the while, you negotiate the best
purchase price from suppliers through
volume. Sound familiar?
30 May/June 2010 | APICS magazine
There must be a better way. Perhaps
you could set up kanban for all part
numbers or vendor-managed inventory.
Maybe you could negotiate consignment agreements with your suppliers.
But, before beginning to work toward
either of these alternatives, it’s necessary
to consider demand segmentation.
Demand segmentation is the categorizing of different demand types
into groups that share similar characteristics. This graphical representation
shows the sales or consumption volume of products versus their demand
variability. Its premise is that all parts
or products are treated uniquely. Each
one has its own personal demand
characteristic. When setting up buy or
reorder strategies, products with high
volume and low demand variability
are treated differently than products
with low volume and high demand
variability.
Traditional product quantity—also
described as ABC analysis—fails to recognize that high volumes are not always
predictable and that low volumes can
be. In trying to classify certain products
for certain pull techniques, it is necessary to understand demand variation.
Demand segmentation analysis
To build a demand segmentation
analysis of product types, you first
need historical sales data, preferably
in 52-week buckets organized by SKU.
More is better here, especially if you’ve
had an increase or decrease in the past
or if you experience cyclical seasonality.
These data may be difficult to come by.
Consult your information technology
department, or contact your software
provider to assist you. It’s crucial that you
only look at pieces or units at this point,
not dollars. You will examine inventory
cost as a measure of success after implementing your new inventory levels.
Now refer to your historical sales
data to calculate the average and
standard deviation of a historical
52 weeks for every SKU. Use the average and standard deviation results to
calculate the coefficient of variation
(CV) for each SKU. The CV helps you
see the variability of your inventory
order patterns. The lower the CV, the
more consistent the order pattern.
It’s predictable.
For example, look at an item with
a low CV, such as a common screw
or fastener used in multiple-assembly
builds. You probably use close to
the same amount of the part each
month, so it might make sense to have
a vendor manage this item for you.
Or maybe you can negotiate a bulk
stocking program with your supplier.
Conversely, an item with a high CV
indicates there is little-to-no predictability in its order pattern, so it might
make more sense to purchase make-toorder (MTO) when needed.
Next, plot your data in a scatter chart
arranged by SKU. Put the total historical volume on the Y axis and the CV on
the X axis. (See Figure 1.)
The purpose of looking at order patterns this way is to give you some predictable data to develop a strategy for
purchasing inventoried SKUs. You now
have something other than a forecast to
use when making strategic decisions.
The CV gives your company a risk
factor to use when buying component
inventory.
A CV less than 1.0 lends itself to flow
and pull techniques. One less than 0.5
often can be handled with rate-based
replenishment methods. If a CV is
greater than 1.0, it might be considered
assemble-to-order. A very high CV
would represent those products that
you may begin to phase out of your
offerings altogether—or at least convey
to customers that these are custom
items requiring greater lead times,
which will give you ample time to order
components from suppliers rather than
holding these slow movers in inventory.
The bottom line is that you do not
want to hold inventory of any SKU with
a high CV. There’s a lot of risk in keeping these items in stock because you
can’t predict when they will sell. Sales
teams often have issues with supporting
this philosophy. After all, the best way
to serve customers is to have what they
want on the shelves, right? But remember: If you sell more of the high-CV
item, it eventually will have a lower
CV—and, at that point, it will qualify
for a revised stocking strategy.
The next piece of information most
professionals overlook when looking at
demand segmentation is days of inventory on hand (DIOH). If DIOH data
are difficult to retrieve, you still can
move forward with your analysis, but
it’s really helpful to chart DIOH along
with historical usage and CV as a way
to compare how well your organization is performing in buy strategies of
slow-moving SKUs. Your DIOH data
also will confirm for any doubters that
using only one buying method for all
product types regardless of CV is not
the best method.
Figure 1 shows DIOH data plotted on
the secondary Y axis. Some SKUs with
a CV greater than 5.0 have an excessive
amount of inventory days; whereas,
SKUs with a low CV might not have
enough inventories to meet demand
requirements.
Now, draw several diagonal lines
across the chart to help identify four
distinct categories. This step can help
your team come to a consensus on a
strategy to use going forward. Figure 2
shows the four distinct categories.
Rate-based components typically are
common in assemblies and include items
such as fasteners or hardware. Because
they are high volume and low CV, they
are predictable in their use. Also, these
items often are inexpensive and easy
to buy—perfect for a vendor-managed
inventory program, which frees up your
purchasing team to deal with more
troublesome day-to-day issues.
Kanban pull components usually
are common items, but have a lower
historical volume or a higher CV. These
SKUs are candidates for a kanban pull
strategy and can be ordered only when
they meet a trigger or reorder point.
MTO items require two data characteristics to be considered. Some
SKU quantities simply are too low to
realistically have a stocking program.
And some quantities exhibit too much
variability to hold. The best strategy
could be MTO. But remember: This is
only a guideline in establishing a risk
factor. There is flexibility here.
Custom, long-lead-time components
often bring about the worst ordering
habits and can really hurt inventory
turns. At this point, it is a good idea to
incorporate inventory dollars back into
your data in order to spotlight the need
for a change in the company’s paradigms.
What are some options for these
custom SKUs? You could write off some
of them, but be sure to consult your
finance department for guidelines. You
also can work with your sales professionals to explore options such as
offering sales liquidation promotions
or reduced-freight programs. These can
create a sales spike in assemblies and
Figure 1: Demand segmentation
APICS magazine | May/June 2010 31
Figure 2: The four categories of demand segmentation
ultimately reduce some excessive, slowmoving inventory. Regardless of what
you decide here, it will take some time.
Stick to your plan, and be patient.
Now that you’ve determined your
demand categories using historical data,
your team can begin to build a strategy
and policies to support its hypothesis.
You also can dive down one more level
to look at how to use the data to calculate safety stock for kanban pull items.
Calculating safety stock levels
Many professionals look at demand
fluctuation and assume there’s not
enough consistency to predict future
demand with any confidence. They
see no other way to set safety stock
levels than by using trial and error—or
maybe just throwing out a comfortable
percentage.
Using the data in Figures 1 and 2, it
would be best to use a stocking strategy
for these rate-based and kanban SKUs.
If your goal is to reduce inventory while
maintaining or improving service levels,
you must rely on historical demand
variability to predict shortfalls and thus
set your safety stock level according to
data, not opinion.
Consider establishing a service-level
factor based on the standard deviation
of your historical data. Figure 3 is a
positive Z-score chart commonly used
in statistics to measure the distance in
standard deviations of a sample from
the mean. In addition, it may be used to
determine service-level factor.
For instance, if you set a service-level
goal of 99.9 percent, the Z-score chart
shows this as a 3.09 service-level factor.
Multiply that by the historical standard
32 May/June 2010 | APICS magazine
deviation of each SKU to calculate the
safety stock quantity required to maintain
a 99.9 percent service level. It’s that easy.
Using a calculation based on normal customer demand variability, it’s possible to
achieve a 99.9 percent probability of not
being out of stock—if you maintain the
calculated safety stock quantity on hand.
When using a service-level factor and
standard deviation to calculate safety
stock, remember the following. First,
this formula is not recommended for
all SKUs and should never be used for
mid- to high-CV part numbers.
The higher the CV, the greater the
variability—meaning, more inventory will
have to be calculated to maintain a higher
service level. For example, an SKU with a
CV of 1.5 (relatively high volume) would
be a good candidate for a kanban pull system. Yet, with a service-level factor of 99.9
percent, the recommended safety stock
quantity may be much greater than what
realistically can be held in inventory. Don’t
be afraid to relax your service-level factor
on an SKU to 97 or 95 percent—or until
you reach an acceptable balance between
the service level and inventory on hand.
The timing to recalculate your CV
greatly depends on your business volatility. Companies that experience high cyclical demand or have had recent increases
or decreases in their business should
recalculate CV a minimum of twice a year.
You can use more than 52 weeks of data
to capture such volatility, as well. Either
way, this is not a formula to calculate once
and then forget about. It must be managed
like everything else, but it’s not necessary
Figure 3: Positive Z-score chart
Positive Z-scores
Z
0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
3
0
.5
.5398
.5793
.6179
.6554
.6915
.7257
.758
.7881
.8159
.8413
.8643
.8849
.9032
.9192
.9332
.9452
.9554
.9641
.9713
.9772
.9821
.9861
.9893
.9918
.9938
.9953
.9965
.9974
.9981
.9987
.01
.504
.5438
.5832
.6217
.6591
.695
.7291
.7611
.791
.8186
.8438
.8665
.8869
.9049
.9207
.9345
.9463
.9564
.9649
.9719
.9778
.9826
.9864
.9896
.992
.994
.9955
.9966
.9975
.9982
.9987
.02
.508
.5478
.5871
.6255
.6628
.6985
.7324
.7642
.7939
.8212
.8461
.8686
.8888
.9066
.9222
.9357
.9474
.9573
.9656
.9726
.9783
.983
.9868
.9898
.9922
.9941
.9956
.9967
.9976
.9982
.9987
.03
.512
.5517
.591
.6293
.6664
.7019
.7357
.7673
.7967
.8238
.8485
.8708
.8907
.9082
.9236
.937
.9484
.9582
.9664
.9732
.9788
.9834
.9871
.9901
.9925
.9943
.9957
.9968
.9977
.9983
.9988
.04
.516
.5557
.5948
.6331
.67
.7054
.7389
.7704
.7995
.8264
.8508
.8729
.8925
.9099
.9251
.9382
.9495
.9591
.9671
.9738
.9793
.9838
.9875
.9904
.9927
.9945
.9959
.9969
.9977
.9984
.9988
.05
.5199
.5596
.5987
.6368
.6736
.7088
.7422
.7734
.8023
.8289
.8531
.8749
.8944
.9115
.9265
.9394
.9505
.9599
.9678
.9744
.9798
.9842
.9878
.9906
.9929
.9946
.996
.997
.9978
.9984
.9989
.06
.5239
.5636
.6026
.6406
.6772
.7123
.7454
.7764
.8051
.8315
.8554
.877
.8962
.9131
.9279
.9406
.9515
.9608
.9686
.975
.9803
.9846
.9881
.9909
.9931
.9948
.9961
.9971
.9979
.9985
.9989
.07
.5279
.5675
.6064
..6443
.6808
.7157
.7486
.7794
.8078
.834
.8577
.879
.898
.9147
.9292
.9418
.9525
.9616
.9693
.9756
.9808
.985
.9884
.9911
.9932
.9949
.9962
.9972
.9979
.9985
.9989
.08
.5319
.5714
.6103
.648
.6844
.719
.7517
.7823
.8106
.8365
.8599
.881
.8997
.9162
.9306
.9429
.9535
.9625
.9699
.9761
.9812
.9854
.9887
.9913
.9934
.9951
.9963
.9973
.998
.9986
.999
.09
.5359
.5753
.6141
.6517
.6879
.7224
.7549
.7852
.8133
.8389
.8621
.883
.9015
.9177
.9319
.9441
.9545
.9633
.9706
.9767
.9817
.9857
.989
.9916
.9936
.9952
.9964
.9974
.9981
.9986
.999
to look at day to day, as you may do with
current purchasing strategies.
A real-life example
Joe Wilson, director of operation planning for Eldorado Stone in San Marcos,
California, is responsible for the planning and inventory logistics of six plants
and two distribution warehouses. His
company historically had approached
inventory management challenges from
an analytical standpoint, using standard
elementary statistics to estimate stocking targets. Wilson turned to a consultant in order to validate his company’s
inventory management strategy.
“Demand segmentation allowed us to
stratify our product offering in a more
thoughtful way and zero in on the true
variability in our sales,” Wilson says.
“Our purchasing and inventory managers [gained] insight on the importance
of having our inventory levels tied to
actual demand history.”
He adds that the benefits of calculating Eldorado Stone’s safety stock using
translates into a very high cost in any
supply chain. Each additional inventory
turn achieved per year can free up cash
for new product development, expanded
marketing and sales, modernization,
reengineering, expansion, acquisitions,
and debt reduction. The more inventory
turns achieved, the more savings gained.
A committed change in your company’s inventory management strategy—
one that is based on statistically sound
and proven data—will enable you to
improve your working capital position
and service levels. Even if digging into
inventory management feels like the next
daunting battle, don’t hold back. Demand
segmentation is a powerful lean tool that
will provide you with a fresh perspective.
statistics included sales team members
being better able to meet the needs of
customers and stocking the right product
on shelves. With a few minor tweaks,
employees constructed an inventory
strategy that enabled them to reduce
inventories an additional 60 percent
from previous levels.
Seeing results
In the short term, your inventory
management team will begin to realize
the benefits of using historical demand
data to help bring about positive and
permanent solutions in inventory
management. They will feel that they
are accomplishing meaningful results
in an otherwise temporary fix and the
never-ending battle of shortages and
planning mishaps. Communication
quickly spreads across the purchasing, planning, and scheduling teams,
promoting a greater good for your
company and customers.
The long-term benefits are endless:
Obsolete and excess inventory quickly
APICS
Shawn Kaul is president and chief
executive officer for LSI Consulting Group
LLC. He may be contacted at skaul@
lsiconsultinggroup.com.
To comment on this article, send a message
to [email protected].
extra
APICS Extra Live:
intelligent inventory Management
Presented by: shawn Kaul
Date:
June 3, 2010
Time:
1:00 p.m.-2:00 p.m. ct
Attend APICS Extra Live to gain deeper insight into the May/June
APICS magazine article by Shawn Kaul, which demonstrates how
to improve inventory management practices through intelligent
demand segmentation.
In this APICS Extra Live, you will deepen your understanding
of demand segmentation and learn how to achieve results by
defining demand segmentation categories that match your
inventory to your business model. In addition, you will gain
insights into managing abnormal conditions and building
confidence with data.
RegisteR online at apics.oRg/extRa.
APICS magazine | May/June 2010 33