7.

Managing Supply Chain Inventory Flows

COMPONENT RISK POOLING

Risk pooling,1 the portfolios effect, and safety stock aggregation all refer to
the same idea, which we will call risk pooling. Risk pooling is the
phenomenon whereby combining demand streams reduces the amount of
safety stock because the sum of random variables has lower levels of
relative uncertainty than the aggregate amount of uncertainty of the
individual random variables. This is true as long as the correlation of the
random variables is less than one. The implication is that less safety stock is
required. For example, if two different demand markets are fulfilled from
two different distribution centers, less safety stock would be required if only
one distribution center served both markets, other things being equal, as
long as the correlation of demand is less than one.
Suppose the demand for an SKU has a standard deviation of 4 in the
market served by Distribution Center 1, and a standard deviation of 3 in the
market served by Distribution Center 2, and that the demand is not
correlated. Suppose that both distribution centers have a one day lead time
and that they multiply the standard deviation of demand during lead time
by 3 to get safety stock. Then currently Distribution Center 1 has 4 × 3 = 12
units of safety stock and Distribution Center 2 has 3 × 3 = 9 units of safety
stock, which is 12 + 9 = 21 units in aggregate. Now, if the two distribution
centers are combined, the new standard deviation is  , so the
safety stock is 5 × 3 = 15 units, which is almost a 30 percent reduction in
the safety stock required.
Now, if the demands are correlated, a modification is required.
X1 = random variable of the demand for SKU 1
X2 = random variable of the demand for SKU 2
Var = variance
ρ = correlation
Then the standard deviation of the sum of the two random variables is

Using the preceding example and assuming that the correlation between
the demands is 0.1, then
And 5.3 × 3 = 15.9, which is still less than 21. In fact, it will only be the same
if ρ(X1, X2) = 1. Recall that ρ(X1, X2)   (–1, 1). In the previous example, if
ρ(X1, X2) = –1, then

Also, in the previous example, if ρ(X1, X2) = 1, then

And 7 × 3 = 21 units, the same amount of inventory as would be required in
aggregate if they remained separate distribution centers.
Risk pooling can occur by combining distribution centers, but it can also
occur in other ways. If two products are exact substitutes, perhaps two
brands of yellow number 2 pencils in a retail store, elimination of one of the
SKUs has the same effect, assuming the same service level target remains.
This concept can also be helpful with forecasting. You can forecast in
aggregate more accurately than you can at a disaggregate level. For
example, you can generally forecast more accurately for one SKU at 100
stores than you can forecast at a single store. You can also typically forecast
in larger time buckets than you can in shorter time buckets. For example, it
is generally true that you can forecast more accurately for a given SKU at a
monthly level than you can at a daily level. The concept of risk pooling
occurs in many areas of supply chain management and in many other
disciplines as well, such as finance. In finance, they call it the portfolio
effect.
As we mentioned earlier, risk pooling goes by the names of the portfolio
effect and safety stock aggregation, but it is also sometimes called
the square root law. It is really not a law, so we do not refer to it as such in
this book. In addition, the square root law assumes that there is no
correlation between the demand random variables, and that is not usually
the case.

BULLWHIP
Bullwhip2 is the amplification of uncertainty in demand as it moves up the
supply chain. For example, the uncertainty of point of sale (POS) would be
less than the uncertainty of orders faced by the retail distribution center,
which would be less than the uncertainty of orders faced by the supplier.
This is one of the reasons why it has become popular for retailers to share
POS data with suppliers—it gives the suppliers a better estimate of demand
than orders do since orders have more noise than POS. Nevertheless, there
is another side to the issue and that is that the order data has information
relevant to the forecasting—namely, the rhythm of the replenishment
process per se. That is, the replenishment system itself many times has a
rhythm that can be incorporated into a forecast. We discuss this in more
detail later in the chapter. First, we go back to the bullwhip phenomenon
per se.
In this section we discuss some of the ways bullwhip is generated.3 One way
bullwhip is generated is through order batching, where individual orders
are in larger increments than sales. For example, shoppers might come into
a store and purchase Cheerios every day of the week. But the store might
only order from the distribution center twice per week. That order batching
hides some of the detailed information about demand from the perspective
of the retail distribution center. The retail distribution center then might be
receiving orders from each of the 20 stores, which in turn places orders on
the supplier once every week, further reducing the amount of demand
information in the order data. Let’s look at an extreme example to illustrate
the idea. Suppose on Day 1, sales at Store 1 for Cheerios go up tenfold per
day, and then on Day 3, Store 1 places an order on the distribution center
that is ten times the size of its typical order. One week later, on Day 10, the
distribution center places an order with the supplier that is much larger
than the size of its typical order. So, clearly, order batching causes a delay
in information regarding demand; in this example the delay is ten days.
Order batching can hide trends and other patterns, but it also simply delays
information about changes in demand. Clearly order batching increases the
noise in the demand information, which leads to stockouts and the need for
excess inventory. In addition it can result in the need for expedited
transportation, increasing transportation costs.
Bullwhip can also be created by errors in forecasts that are used to create
orders; such errors might be introduced through human error in
estimation, through the forecasting method per se, or both. For example, if
a trend forecasting method, such as second order exponential smoothing,
also known as Holt’s models, is used when, in fact, there is no trend, and if
a high smoothing constant is used in the model, there will probably be over
forecasting in the creation of some orders and under forecasting in others.
This is especially magnified as the time horizon of the forecast increases. If
a high smoothingconstant is used to update the estimate of the trend
component of the forecast, most recent change in sales will be represented
more persuasively in the trend component. Then if the forecast goes out
further into the future, as it would with larger order batching, the error of
the false trend, be it upward or downward, is magnified.
There are a number of ways of measuring bullwhip, but we discuss the most
straightforward method, the ratio of the variances. Which ratios depends
on where you want to measure bullwhip. For example, if you want to
measure the bullwhip generated by a node, you could measure the ratio of
the variance of orders to the variance of sales. If you want to measure from
one echelon to another, then you could, for example, measure the variance
of POS at all retail stores owned by a retailer and the variance of all orders
from all of the retail stores.
Since the primary cost of bullwhip is excess inventory (that is, excess safety
stock), increased stockouts, and/or increased transportation costs for a
given SKU, measurement of bullwhip is most meaningful at the SKU level.
A number of publications talk about bullwhip at the industry level, and that
is probably worthwhile for economics, but for inventory management, it is
not meaningful. Similarly, bullwhip at the monthly level, quarterly level, or
annual level is not meaningful for SKUs that are replenished daily and
weekly. In general, bullwhip measurement is most meaningful from a
supply chain management perspective when it is at the SKU level and in
time intervals that match replenishment cycles.
Using the discrete event simulation tool developed in Chapter 5, “Discrete
Event Simulation of Inventory Processes,” you can investigate how a single
node can generate or reduce bullwhip. Using discrete event simulation you
can investigate various replenishment processes and parameters to see the
impact on bullwhip. For example, using the model developed in Chapter 5,
you could set the variance of demand high, the order quantity close to the
mean of demand, and the safety stock high, and you would discover how a
replenishment system can actually reduce bullwhip or create production
smoothing. Production smoothing occurs when the variance of orders is
less than the variance of demand. Earlier we mentioned that although you
can get a clearer understanding of demand by looking at POS data, there is
useful information in the order data. Consider the example just described
where demand variance is high, the store order quantity is set to the mean
of demand, and the safety stock is high, then the store will order the same
amount nearly every day while demand will have a lot of variance. If the
supplying entity were to set safety stock based on point of sale data, it
would have too much safety stock since the actual uncertainty in the
amount ordered is low.4
Figure 7-1 compares the ideas of risk pooling and bullwhip.

Figure 7-1 Bullwhip and risk pooling

In Figure 7-1, the bottom row represents aggregate POS for four different
retailers and the circle around them represents all of that POS being
aggregated. Suppose that all four of these retailers’ distribution centers are
supplied by a single supplier distribution center for this particular SKU.
Then the level of analysis of bullwhip appropriate is the aggregation of
these orders. The middle row represents retailers processing the demand
and turning the demand into orders. This could involve several levels of
ordering, possibly including stores ordering from retail distribution centers,
and then the retail distribution centers ordering from the suppliers.
In Figure 7-1, it says, for example, “Aggregate Orders for Retailer 1.” This
might include orders from one distribution center or 100 distribution
centers. So, although these orders are coming from four retailers, the
number of ship-to locations is probably much greater.
Measures of bullwhip compare the level of variability of the top row, in
aggregate representing orders from retailers, to the bottom row.
Remember, this analysis is for a single SKU since the supplier has to hold
inventory for a single SKU. Risk pooling could be used to analyze the
incremental cost associated with splitting the distribution center into two
different distribution centers. If the supplier did that and had it set up to
where one distribution center serves Retailers 1 and 2, and another
distribution center serves Retailers 3 and 4, then two bullwhip calculations
would be needed instead of one. Of course this would require the supplier
to hold more safety stock to achieve the same level of service, unless at least
one of the new distribution centers was so much closer to the retailers’
distribution centers that the lead time decreased enough to make up the
difference.

INVENTORY POSTPONEMENT
Benetton made clothing in Italy and sold a great deal of products in stores
in the United States, primarily located in shopping malls. Benetton was
well-known for having a particular color theme, and it was recognized as
being very good at recognizing color trends in the market.
Color trends are some of the most fickle of clothing styles. It is difficult to
judge which colors will be popular for a particular season. Benetton’s
approach was to offer solid colors in its clothing lines seeking to include the
trendy colors. Invariably, however, at the end of a season Benetton would
have excess inventory of certain colors, which prompted markdowns to
move the products out. Markdowns are sometimes set at a price below the
cost of the product just to sell or eliminate the inventory. Like many in the
clothing industry, Benetton noticed that it would stockout of popular colors
and have to mark down the remaining colors because no one wanted them.
Traditionally Benetton’s clothing production process involved taking
bleached white yarn, then dyeing it to a particular color specification,
followed by knitting the garment, say, a sweater. Finally, the manufacturer
would ship the product from Italy to the United States.
The dyeing process was capital intensive and fast, while the knitting and
assembling process was labor intensive and long. The question for Benetton
became, “How do we deal with the problem of excess stock of colors at the
end of a season?” Obvious answers might be, “Try to speed up the process
by assembling faster,” or, “Find a machine that knits faster.”
Benetton decided to switch the process around and knit and assemble the
garment and then dye it. This was iconoclastic. But there was no reason not
to with the solid color style. Postponement was beneficial because as the
season approached, Benetton was able to more accurately predict color
trends, which dictated sales. Benetton, in this case, was able to produce in-
demand colors and reduce the excess inventory at the end of the season by
minimizing the production of the unwanted colors.
In general, the delay of production or distribution is referred to
as postponement.5 Another definition states that postponement6 is the
number of stages of production and distribution that are delayed before
receipt of a customer order. Postponement allows a company to make
product customization more cost effective. It can also be used to delay
transportation and warehousing costs.
Postponement can allow for more centralized holding of inventory, and
therefore, possibly increase cycle stock at the centralized location, while
reducing it in the forward inventory holding locations. It can also result in
less safety stock in end item inventory.
Speculation is akin to the opposite of postponement. Speculation involves
conducting various value added activities well in advance of demand,
including shipping the product early to locations where it is anticipated that
demand will occur, customizing products prior to receiving orders, and so
on. Speculation allows a company to be first or to fulfill the early demand.
So, although postponement can facilitate cost savings, it must be weighed
against the benefits of speculation. However, it is possible to implement a
mixed strategy, including using speculation on some of the production and
postponement on the remainder of the production. Furthermore, if you
view speculation and postponement as a continuum, you can imagine a
portfolio strategy where different portions of expected demand are
produced and distributed on different points of the continuum. It would
seem that the marketing function of an organization would push for
speculation, and the supply chain organization would push for
postponement. It is the responsibility of supply chain management to find
areas of production and distribution where postponement can be
implemented without jeopardizing marketing and sales opportunities.

MERGE-IN-TRANSIT
Merge-in-transit7 is a method of bringing together components or items
that have disparate origins but a common destination. For example,
suppose a company bought office equipment from three sources in
Southwest Michigan and that all together they cube out a 53-foot trailer for
delivery to Phoenix, Arizona. Rather than paying for three different
truckloads from Southwest Michigan to Phoenix, the three shipments could
be combined in Southwest Michigan and then a single truckload to
Phoenix. This is a type of shipment consolidation. However, merge-in-
transit can also be assembly-in-transit where components come from
multiple sources but need to be delivered as a single unit. Merge-in-transit
can reduce transportation costs as has been explained. It can also reduce
inventory costs for the receiving company because everything arrives
together and ready for use. Without merge-in-transit, it is possible that the
components will arrive and then cannot be used until the other components
arrive and are assembled, thus increasing the inventory holding cost.8

VENDOR MANAGED INVENTORY

Vendor managed inventory9 (VMI) is an inventory management process
whereby the supplier makes the decision about the replenishment timing,
quantity, or both. For VMI to work, the supplier must have, at a minimum,
visibility to the customer’s inventory position. However, it would be better
if the supplier using VMI had visibility to point of sale data as well since
inventory management requires forecasting. The key idea behind VMI is
that the supplier has visibility to other customers’ demand as well so the
supplier can make decisions about shipping orders, taking this into
account, thus smoothing its own demand. By smoothing its own demand
the supplier will need to hold less safety stock and will have fewer
stockouts. Stockouts at the supplier result in lead time uncertainty from the
customer’s perspective, which can result in customers holding more safety
stock or experiencing more stockouts or both. To implement VMI, not only
does the customer need to share inventory position and demand data, but
the customer also needs to work with the supplier to understand its own
performance targets such as inventory targets and fill rate targets.
Customers must also be careful to understand how these metrics are to be
calculated and updated. Customers are always wary of this because it is
possible for the supplier to make a decision that will make their competitor
better off at their expense. Another concern is that if a supplier is having a
month where sales are down, the supplier could place a lot of orders, thus
improving its sales. Of course, if metrics are set forth and agreed to, this
would be detected. In addition to the possibility of better inventory
management, customers might like VMI because it transfers some of the
labor costs from the customer to the supplier.

CONSIGNMENT
Vendor managed inventory is sometimes confused with the concept
of consignment. Consignment occurs when the supplier owns the inventory
in the customer’s facility until the customer sells the inventory. Vendor
managed inventory and consignment are two separate decisions, because
VMI can be implemented with or without consignment. Customers like
consignment because it takes away much of their risk and can also help
them with cash flow. On the other hand, because it reduces their risk of not
selling, the customer might not try as hard to sell the inventory.10 Whether
or not a customer is better off with consignment is not as straightforward as
it might appear on the surface. Suppose the terms of sale with the supplier
are net 30 days if there is no consignment, and with consignment, the
customer has to transfer the funds the same day. If the customer is turning
the inventory every week, the customer is better off without consignment in
terms of cash to cash cycle time. Without consignment, the customer
receives the inventory on Day 1 and sells it on Day 7 but does not have to
pay the supplier until Day 30. This means that the customer gets the cash
23 days before he has to pay the supplier. With consignment, the customer
gets the cash and pays the supplier on the same day in this example. The
point is that when analyzing the benefits and costs of consignment, you
must consider (1) how it affects incentives, (2) terms of sale with
consignment and without consignment, and (3) rate of inventory turns.

REVERSE CONSIGNMENT
Reverse consignment11 occurs when the customer buys the product and
owns it but does not want the supplier to ship it until a later date. This can
happen when the (1) customer is at capacity in terms of inventory storage,
(2) there is a shortage of the product in the market and the customer does
not yet know which distribution center or location will need the inventory,
and (3) there is a special deal given, a promotion, that the customer wants
to take advantage of even though the customer doesn’t need the inventory
yet. In addition, there are a few other reasons, but in some ways this can be
considered a form of postponement because you are essentially delaying the
movement of the inventory.

COLLABORATIVE PLANNING, FORECASTING,

AND REPLENISHMENT
Collaborative planning, forecasting, and replenishment 12 (CPFR) is an
inventory replenishment process that involves collaboration between a
customer (usually a retailer) and a supplier that is implemented in many
different ways, but the goal is to come to an agreement regarding the
forecasts and replenishment objectives. A number of guidelines have been
proposed over the years regarding how the CPFR process should
work.13,14 We do not go into those here, but instead discuss the key concepts
behind the process. One of the fundamental concepts behind CPFR is that
retailers and suppliers have different types of information, which together
can result in better forecasting and replenishment decisions. For example,
suppliers have visibility to the sales of all retailers, but retailers themselves
know their own shoppers better and the specific markets they are in. A
supplier might be able to see a trend developing across the country more
quickly than some retailers, especially regional retailers. However, a retailer
is aware of changes it is making in terms of assortment, remodeling,
changes in local demographics, road construction, and so on.

PUSH VERSUS PULL

A push inventory system forecasts demand and sends inventory based on
the forecast, whereas a pull inventory system sends inventory based on
inventory being used or purchased. Is a (Q,ROP) inventory replenishment
system push or pull? When enough inventory has been taken away, it sends
inventory; that is, the ROP is the trigger to send inventory when enough has
been removed. In this sense, it is a pull system. However, the reorder point
is usually based on a forecast, which makes it a push system. Consider a
situation where the reorder point is ROP1 but at time t the forecast
increases so that the reorder point increases to ROP2. Now, an order will be
triggered even if no inventory has been removed. So technically, a (Q,ROP)
replenishment process is a hybrid; the same argument can be made about
the (T,OUL) replenishment process. Both of these inventory replenishment
processes fall on the pull side of the push versus pull continuum in
comparison to a material requirements planning (MRP) system or a
distribution requirements planning (DRP) system.15 In these systems,
inventory is pushed based on a forecast into the future planning horizon.
A kanban16 system, developed in Japan, is the epitome of a pull system and
is typically used within a production environment. There are many kanban
system designs, but one of the simplest is a two-bin kanban system.
Imagine on the factory floor, at a workstation, you have a bin full of
components used in the production of some product. When the bin is
empty, the empty bin is taken to the inventory storage area where another
bin is full. The empty bin is left in the inventory storage area, and the full
bin is taken to the workstation on the factory floor. Instead of moving a bin,
many times a kanban card is moved to the inventory storage area. The
kanban card includes information about how much to move, where it goes,
when it is due, and so on. This allows for additional inventory prior to the
bin being completely empty. In this regard, it is a (Q,ROP) process. The size
of the bin determines Q, and either the size of the bin or the information on
the kanban card determines the reorder point. In the two-bin kanban
system described earlier, Q and ROP are equal to the size of the bin. Many
unique aspects of the kanban system are visual in nature; you can see when
something needs to be replenished.
CHANNEL SEPARATION
Channel separation17 is the notion that marketing channels and physical
distribution channels do not need to be the same. For example, Firm A can
buy from Firm B and then sell to Firm C, but the product can flow from
Firm B to Firm C. That is channel separation because the marketing
channel is different from the physical distribution channel: In this example
the marketing channel flows from Firm B to Firm A and then to Firm C, but
the physical distribution channel flows from Firm B to Firm C. This reduces
transportation costs, labor costs, and inventory costs in many cases.
Channel separation can be thwarted by lack of trust. For example, if Firm A
does not trust Firm B, because Firm A thinks that Firm B might cut Firm A
out of the relationship with Firm C, then channel separation will probably
not occur. Sometimes Firm A will help Firm B set up operations so that
Firm B can print labels so that Firm C is not even aware that product is
coming directly from Firm B. This type of arrangement is often referred to
as drop shipping. For example, Amazon.com offers drop shipping services
for firms so that all they have to do is marketing; Amazon.com handles the
holding and shipping of the inventory.

INVENTORY PLACEMENT OPTIMIZATION

Inventory placement optimization has to do with where inventory, and
especially safety stock, is held in the supply chain. As discussed earlier, a
node can order near demand quantities frequently, with high safety stock
levels to shield the higher levels in the supply chain from having to hold as
much safety stock. However, it is often the case that it is less expensive to
hold inventory at higher echelons in the supply chain. At the extreme, retail
locations tend to be in higher rent locations, near population centers,
whereas distribution centers and factories tend to be located in rural areas
away from the population where rent is low. Similarly, in retail facilities
there are high opportunity costs for the shelf space because there is a fixed
amount of space and when safety stock is held on the shelf, there is less
space for additional assortment depth and/or breadth. Nevertheless, for
some SKUs with very high stockout costs, in some cases it makes sense to
hold the preponderance of the inventory at the retail store. But finding the
optimal placement of inventory, including safety stock, cycle stock, and so
on, requires careful analysis of transportation costs, inventory holding
costs, stockout costs, and must take into account various inventory
replenishment process options, including cross docking.
THE GLOBAL SUPPLY CHAIN IMPACT
There are two categorically different challenges with global inventory
management: (1) inventory management for global sourcing and (2)
inventory management for selling in foreign markets. These are
fundamentally different and have different challenges. Inventory
management for global sourcing is a challenge because of the long lead
times, customs clearance, and communication challenges. Inventory
management for selling in foreign markets is a challenge due to differences
in legal systems, lack of infrastructure, differences in consumer and
shopper behavior, and taxation rules and complexities. We begin with
inventory management for global sourcing.
In managing inventory for global sourcing it is important to take into
account various transportation options and their impact on cycle stock,
safety stock, and in-transit stock, because the differences can be dramatic.
The difference between air and inland water/ocean can be three to four
weeks. For example, going from the inland water/ocean combination to air,
if the difference is four weeks, then in-transit stock could be reduced by
around 90 percent. In addition, for a given level of demand uncertainty, the
reduction in safety stock could be around 75 percent.18 Of course the
transportation costs are also dramatically different, but the point is that
good inventory management for global sourcing must carefully consider
transportation options and not just assume one method is better than
another. In addition, it is possible that in some circumstances that typically
ocean should be used with the allowance for emergency shipments using air
or that some base percentage of demand should be covered with ocean
carriage while the most uncertain portion of demand is covered with air
carriage. All of these options must be considered with respect to all of the
costs and customer service targets. Many times production and labor costs
are the key drivers in sourcing decisions when in fact other factors should
be weighed such as inventory holding costs, transportation costs, stockout
costs, as well as taxation and regulatory compliance costs. It is also
important to notice where costs are added to the inventory. For example, a
garment sourced from a country in Asia might have the preponderance of
value added in that country, whereas ocean carriage may be a small fraction
of the total cost of the garment. At the same time, the duty on the garment
could be around the same magnitude as that of the production cost in Asia.
Consequently, it might make sense to use a foreign trade zone (FTZ). If
product is brought into the United States through an FTZ, it is possible that
duty can be delayed until product leaves the FTZ. Details like this can affect
the optimal inventory positioning in the supply chain.
When products are imported, they must go through customs, and the
products must be classified according to the Harmonized Tariff Schedule of
the United States (HTSUS). The origin of the product and the product
classification have a significant impact on whether a tariff is imposed and
the magnitude of the tariff. Consequently, this has an impact on the
inventory investment and the inventory cost after the product is imported,
which has an impact on optimal product placement and logistics network
design. If you have a silk tie, where the silk comes from China, the product
is sewn and assembled in Bangladesh, and is designed and sold from an
Italian company, what is the country of origin? Generally, from a cost
perspective you would like the country of origin to be one with the lowest
tariff, but from a marketing perspective, you might want it to be from the
country with the most caché. Determining country of origin is important
and requires expertise from someone such as a customs broker. If a product
has a questionable country of origin or a debatable product classification, it
can cause shipments to be held up in customs, depending on the product
and the country where importation is occurring.
In managing inventory for sale and distribution in foreign markets it is
difficult to make many generalizations. Developing countries on the same
continent or developed countries on the same continent can vary
significantly in terms of infrastructure, shopper preferences, homogeneity
of product assortment, transportation capacity and competition, warehouse
availability, land costs, labor costs, labor regulations, and value added tax
(VAT) rules.
As a company moves operations to a foreign country to sell or distribute its
product, forecasting can be difficult because of the fact that the country has
different cultural events and holidays. For example, a retailer running
operations in a foreign country would need to understand holidays, not
only for shopping behaviors but also for hourly labor behaviors. In some
countries, hourly labor travel to their hometowns for certain holidays. For a
retailer, this has implications for forecasting demand since some cities have
high densities of transient labor and therefore may face a reduced demand
in those areas when people leave for a holiday, whereas the areas where the
labor is going to for the holidays may face increased demand. Incorporating
these differences into the forecasting and replenishment processes can
appear easy on the surface, but in reality it can be challenging, especially
early in a company’s foray into a foreign market.
Managing retail inventory in foreign markets can be challenging, especially
when coming from a homogeneous market. For example, while there are
many different demographics in the United States, the overall grocery and
general merchandise assortment is relatively homogeneous, especially in
comparison to a country like China. Throughout China there are many
different cuisines and raw materials that go into making those cuisines.
Consequently, the retail assortments in China are more heterogeneous than
those in the United States. In China, the sources of the grocery products are
more regional as well. This leaves fewer products for centralized
distribution. So on the one hand you have more homogeneous assortments
and more regional sourcing. All of this leads to the need for more inventory,
and lower transportation efficiency.
In addition, the economics of using distribution centers in various countries
differs widely due to land costs, labor costs, and labor regulations. These
variables clearly affect the optimal level of automation in a distribution
center, which in turn affects the fixed versus variables costs, the payback,
and the ROI. In some countries where land is relatively expensive you find
distribution centers that are multilevel and highly automated, whereas in
countries with low land costs you find sprawling distribution centers with
large yards. If the optimal solution is to have a multilevel, highly automated
distribution center, the distribution center will probably be smaller than it
would otherwise be. In a retail distribution setting this may result in a
higher number of products being shipped by direct store delivery (DSD).

RETAIL AND CONSUMER PRODUCTS

INVENTORY MANAGEMENT
There are three primary methods of store replenishment: (1) direct store
delivery, (2) from retail distribution center to store, and (3) from supplier,
cross docked through the distribution center to the store. With direct store
delivery the supplier delivers the product to the store and many times
actually puts the product on the shelf. If you have 100 stores and receive
deliveries from one supplier 300 days per year that is 100 stores × 300 days
= 30,000 invoices and receiving documents. That is just for one supplier.
Whereas, if you received one truckload from the supplier each week at a
distribution center that serves 100 stores, then that is just 52 invoices and
receiving documents. This reduces the transaction costs, which can be
significant due to invoice match failures—the invoice or the receiving
document or the order do not match. When a match failure occurs, the
accounts payable department has to investigate the failure. From a supply
chain perspective, transportation costs are higher than if the product was
just put on the truck from the distribution center with all of the other
products going to the store. However, there are benefits to DSD. Consider
fresh bread. The bakeries are local and bread goes out of date quickly so
DSD makes sense. Other benefits of DSD include the value knowledgeable
DSD delivery professionals bring to keeping the shelves looking attractive.

Cross Docking
Cross docking in the retail setting has a different meaning than it does in
less-than-truckload (LTL). In retail it means that when product from a
supplier arrives at the distribution center it is divided up based on specific
store orders and then staged for trucks heading to specific stores. If the
allocation to the specific stores is based on the original orders from the
store, at the time the order was placed with the supplier, then cross docking
requires more inventory to be held in the store to hit the same service levels
as would be the case with the product being stored in the distribution
centers. The reason for this is that the lead time to the store with cross
docking is the lead time from the supplier to the distribution center and
then through the distribution center and from the distribution center to the
store. Whereas if it is held in the distribution center, the relevant lead time
to the store is just the lead time from the distribution center to the store.
There is a means of overcoming this problem, namely, post receipt
allocation. Post receipt allocation means that once the product arrives at
the distribution center for cross docking, the product is then allocated to
the stores; it is not allocated to the stores based on the requirements at the
time the order was placed with the supplier.19 So without post receipt
allocation, there is a clear trade-off between more inventory in the stores
with cross docking, and more inventory in the distribution center without
cross docking, just holding the inventory in the distribution center. With
post receipt allocation, orders can be generated using the distribution
echelon inventory position and later using a heuristic for allocation to the
stores.

Assortment
Assortment decisions are typically based on demand and space availability,
but these decisions have a significant impact on inventory management and
forecasting as well. Assortment depth in a category has to do with the
number of different SKUs of a given number of brands in the category,
whereas assortment breadth is the number of different brands the category
carries. In a fixed amount of space, as the assortment increases, the
inventory holding capacity per SKU decreases. This increases the expected
number of units out of stock per replenishment cycle. However, oddly
enough, it can also increase the inventory holding cost.20 The reason for this
is when SKU n is added it tends to have lower volume than SKU n-1. That
is, retailers tend to start with the highest volume SKUs in their markets and
then add additional SKUs in order of decreasing volume. This is not always
true, but it is true in many cases. As SKUs continue to be added, eventually
stockouts of the top sellers increase and the average inventory in the
category increases. Of course the additional need to replenish the shelf for
the fastest moving SKUs can be addressed through the addition of store
labor and simply moving product more often from the backroom to the
shelf. This can be difficult during the busiest shopping times, when it is
needed most. There is a limit to how much additional labor can solve the
shelf replenishment problem.21 As more SKUs are added, it can increase
market share for the retailer, bringing in additional customers that perhaps
would have shopped elsewhere. This is one of the reasons why assortment
decisions cannot be made in isolation. But this effect must be traded off
against the additional stockouts that might occur at the shelf as a result.

New Item Introductions

New item introductions also cause similar challenges from an inventory
management perspective. When a new item is introduced to a category, it
creates uncertainty in inventory management and often takes space from a
top selling SKU. The top-selling SKUs typically have the most facings and
are sometimes the only ones with multiple facings. In those cases, the only
choice in terms of making space for the new item is from the top selling
items. This assumes that none of the existing SKUs in the assortment are
deleted. However, in addition to taking space from top selling SKUs, new
item introductions have highly uncertain demand, and it is difficult to make
forecasts for new items, because of a lack of historical data. The majority of
sales forecasting methods rely on historical sales. Most of the time new
item demand is forecasted simply using the judgment of salespeople and
merchandising managers. Another approach to new item forecasting is
using historical data of similar items, but this requires knowledge of which
item is similar. Even if it is similar, that doesn’t imply that the future sales
will be similar. Another source of uncertainty that new items create is
uncertainty about substitution. That is, these new items may absorb some
of the demand from existing items, making it more difficult to forecast their
demand.
Pallet, Case Pack, Inner Pack, and Units
The most common retail store replenishment quantities are pallet, case
pack, inner pack, and individual units. Replenishing in pallet quantities is
common in the retail club business and also occurs in other retail formats
in a more limited way, such as for special retail pallet display promotions.
Case pack replenishment is perhaps the most common, especially in
grocery and fast moving consumer goods. Replenishing in inner packs22 and
individual units requires a break pack process at the distribution center.
That is, instead of, for example, sending a case of 24 to a store that sells one
per week (24 weeks of supply), the distribution center might break down
the case and send individual units to each store (one week of supply). This
decreases inventory holding costs at the stores but increases labor costs at
the distribution center. Regarding case pack quantities, they are often
determined by the need for the cases to fit on a pallet and/or the need to
cube out a truck. One might think that suppliers would have many different
case pack quantities for different retailers and volumes, but there is actually
very little variety. So, for one store, a case pack might represent a half year
of supply while at another retail store it might represent a week of supply.

Retail Shelf Layout

The layout of the retail shelf and space allocation for products on the shelf
are important aspects of inventory management. How product is positioned
on the shelf can affect sales.23 Some products sell better on the ends of the
aisles or on the ends of the category, whereas others sell better in the
middle. Some items are better positioned high on the shelf and others lower
on the shelf. Some items sell better when they are next other specific items
than when they are not. For some items, having more inventory of it on the
shelf increases sales.24 This is especially true for impulse items and less true
for destination items. Some research indicates that shoppers have trouble
distinguishing assortment variety versus space allocation in a category.25 As
you can see, there are many things to consider in laying out a retail shelf,
especially from a demand perspective. All these decisions, where the item is
placed, the SKU it is placed next to, the amount of inventory on the shelf,
and so on, affect the demand, and therefore, should affect the forecast,
reorder point, and reorder quantity or the order up to level.
Another challenge with shelf layout from an inventory management
perspective is that the demand for items changes by day of the week in
some cases. For example, more people shop at the grocery store on the
weekend. Furthermore, some items sell more on certain days than on
others, even throughout the work week. This means that the optimal space
allocation one day might be different from the optimal space allocation for
another day of the week, but retailers change their shelf layout infrequently.
Demand also varies by time of day. At some retail store locations, the
retailers face heavy after work shopping patterns. Again, the amount of
inventory allocated to the shelf might be sufficient for the average day of
the week but might be insufficient for Saturday morning, for example. In an
extreme example, suppose that the top selling cinnamon roll sells 90
percent of the units on Saturday morning between 6:00 a.m. and noon. If
the space was allocated based on average weekly demand, it would have
enough demand for about 14 percent of weekly sales. This means that on
average, the store would sell out of those cinnamon rolls before 7:00 a.m.
On the other hand, if the store gave enough space for 90 percent of the
sales, then most of the week that space would be underutilized because not
all items have such uneven sales. This is a trade-off that must be considered
in retail inventory management. Optimization can be used to model this
trade-off and maximize profit.26
Retail shelf space allocation for seasonal merchandise is also challenging,
because space must be allocated for the seasonal merchandise and some of
that space may come from the inventory of some staple items. Some
retailers maintain space for seasonal merchandise, but the amount of
seasonal merchandise needed changes by season. For example, in the
canned goods category, during the Thanksgiving season, a certain brand of
canned peas is in high demand in the south part of the United States. This
can be managed in multiple ways, including special promotional displays
and taking space from some existing items in the canned goods category.
This will affect forecasting and inventory management not only for this
SKU of canned peas but also for many other SKUs in the category.
Generally, people might not leave the store and go to another store looking
for a certain SKU of canned peas, but in the South during a particular
season for a particular brand of peas, many shoppers will take that path.
This change in the cost of a stockout must be considered in the inventory
allocation decisions.

ENDNOTES
1. Zinn, Walter, Michael Levy, and Donald J. Bowersox. “Measuring the
Effect of Inventory Centralization/Decentralization on Aggregate Safety
Stock: The Square Root Law’ Revisited.” Journal of Business Logistics 10.1
(1989): 1-14.
2. Lee, Hau. L., Venkata Padmanabhan, and Seungjin Whang. “Information
Distortion in a Supply Chain: The Bullwhip Effect.” Management
Science 43 (4) (1997): 546-559.
3. Ibid.
4. This may not be wise since it is usually more expensive to hold inventory
at the store than upstream from the store.
5. Waller, Matthew A., Pratibha A. Dabholkar, and Julie J. Gentry.
“Postponement, Product Customization, and Market-Oriented Supply
Chain Management.” Journal of Business Logistics 21.2 (2000): 133-160.
6. Ibid.
7. Croxton, Keely L., Bernard Gendron, and Thomas L. Magnanti. “Models
and Methods for Merge-in-Transit Operations.” Transportation
Science 37.1 (2003): 1-22.
8. This assumes the consignee owns the goods once they are received.
9. Waller, Matt, M. Eric Johnson, and Tom Davis. “Vendor-Managed
Inventory in the Retail Supply Chain.” Journal of Business Logistics 20
(1999): 183-204.
10. Known as moral hazard.
11. Lee, Hau Leung, and Seungjin Whang. “The Whose, Where and How of
Inventory Control Design.”Supply Chain Management Review 12.8
(2008): 22-29.
12. Stank, Theodore P., Patricia J. Daugherty, and Chad W. Autry.
“Collaborative Planning: Supporting Automatic Replenishment
Programs.” Supply Chain Management: An International Journal 4.2
(1999): 75-85.
13. Kahn, Kenneth B., Elliot N. Maltz, and John T. Mentzer. “Demand
Collaboration: Effects on Knowledge Creation, Relationships, and Supply
Chain Performance.” Journal of Business Logistics27.2 (2006): 191-221.
14. McCarthy, Teresa M., and Susan L. Golicic. “Implementing
Collaborative Forecasting to Improve Supply Chain
Performance.” International Journal of Physical Distribution and
Logistics Management 32.6 (2002): 431-454.
15. MRP and DRP were discussed in Chapter 6.
16. Schonberger, Richard J. Japanese Manufacturing Techniques: Nine
Hidden Lessons in Simplicity. SimonandSchuster. com, 1982.
17. Hutt, Michael D., and Thomas W. Speh. “Realigning Industrial
Marketing Channels.” Industrial Marketing Management 12.3 (1983): 171-
177.
18. For a rough estimate of the reduction in safety stock you could use the
following formula. Let L1 be the current lead time and L2 be the proposed
lead time, then a rough estimate of the percentage change in safety stock is

given by  .
19. Waller, Matthew A., C. Richard Cassady, and John Ozment. “Impact of
Cross-Docking on Inventory in a Decentralized Retail Supply
Chain.” Transportation Research Part E: Logistics and Transportation
Review 42.5 (2006): 359-382.
20. Stassen, Robert E., and Matthew A. Waller. “Logistics and Assortment
Depth in the Retail Supply Chain: Evidence from Grocery
Categories.” Journal of Business Logistics 23.1 (2002): 125-143.
21. Eroglu, Cuneyt, Brent D. Williams, and Matthew A. Waller. “The
Backroom Effect in Retail Operations.” Production and Operations
Management (2012). Waller, Matthew A., et al. “Marketing at the Retail
Shelf: An Examination of Moderating Effects of Logistics on SKU Market
Share.” Journal of the Academy of Marketing Science 38.1 (2010): 105-117.
22. Inner packs are within a case and are multiple units bound together by
plastic or some other method. The inner packs are not for sale but for
distribution at stores. The stores then must break open the inner pack
before displaying the product on the shelf.
23. Dreze, Xavier, Stephen J. Hoch, and Mary E. Purk. “Shelf Management
and Space Elasticity.”Journal of Retailing 70.4 (1995): 301-326.
24. Urban, Timothy L. “An Inventory-Theoretic Approach to Product
Assortment and Shelf-Space Allocation.” Journal of Retailing 74.1 (1998):
15-35.
25. Broniarczyk, Susan M., Wayne D. Hoyer, and Leigh McAlister.
“Consumers’ Perceptions of the Assortment Offered in a Grocery Category:
The Impact of Item Reduction.” Journal of Marketing Research (1998):
166-176.
26. Dulaney, Earl F., and Matthew A. Waller. “System, Method and Article
of Manufacture to Optimize Inventory and Merchandising Shelf Space
Utilization.” U.S. Patent No. 6,341,269. 22 Jan. 2002.