3.

Inventory Control

This chapter focuses on explaining inventory control 1 processes and the

effect of uncertainty in these processes. In inventory control, we have
uncertainty in demand, lead time, and sometimes the review interval itself.
When we place an order, if demand spikes, we might stockout before the
inventory arrives. On the other hand, demand might be steady while lead
time takes longer than expected, possibly resulting in a stockout. 2 The fact
is, there is uncertainty in both demand and lead time and that has a
significant impact on the overall performance of an inventory control
system. In addition to uncertainty in demand and lead time, there is
uncertainty in execution of tasks involved in the inventory process, but we
do not address this particular type of uncertainty until Chapter 5, “Discrete
Event Simulation of Inventory Processes.”

UNCERTAINTY IN INVENTORY PROCESSES

Many variables affect the actual sales of a given SKU: weather, number of
shoppers in a store, stockouts of substitute products, advertisements,
promotions, changing demographics, traffic congestion, social media, price,
placement, assortment depth and breadth, parking lot expansions, road
construction, news reports, and many others. Similarly, there are a plethora
of drivers of lead time, including distance, order receiving processes, order
picking processes, availability of product, order staging processes, carrier
reliability, mode of transportation, and many others. So when you combine
all these into the demand during the lead time, you have many sources of
uncertainty.
Consider a retail distribution center ordering a particular SKU of laundry
detergent from a supplier.Table 3-1 shows the demand during lead time for
60 orders:3
Table 3-1 Demand During Lead Time for a Retailer Distribution Center

Assume the distribution center is serving 200 stores. The columns are
DDLT or demand during lead time. After the distribution center places an
order with this specific supplier, it keeps track of how many units are
ordered from the stores and adds them together until the order is received
and available for use. So, the uncertainty in the demand during lead time
in Table 3-1 represents a combination of demand uncertainty and lead time
uncertainty.4 Figure 3-1 is a graphical representation of Table 3-1.
Figure 3-1 Demand during lead time for a distribution center

In Figure 3-1, the horizontal axis is the order number, and the vertical axis
is demand during lead time. There is no pattern in the demand during lead
time so the variability is due to randomness.

Representing Uncertainty with Empirical Distributions

Figure 3-2 is a histogram of demand during lead time, where the horizontal
axis is the bin number and the vertical axis is the frequency. For example,
there were 2 orders where demand during lead time was 30 units or less, 10
orders where demand during lead time was greater than 30 but less than or
equal to 35, and so on. This could be converted into an empirical
probability distribution that could be used to represent the demand during
lead time.
Figure 3-2 Histogram of demand during lead time

Table 3-2 illustrates how the histogram could be converted to an empirical

probability distribution. For each bin, the frequency of observation is
divided by the total number of observations. For example, for the bin
representing 30 or less, there were 2 observations so 2/60 = 0.03. You
could even use a bin for each actual level of DDLT observed.

Table 3-2 Empirical Distribution of Demand During Lead Time

In that case, the histogram is Figure 3-3.

Figure 3-3 Histogram of demand during lead time

Comparing Figures 3-3 and 3-2 shows that Figure 3-2 looks more like a

normal distribution. We only have 60 observations, so it naturally doesn’t
look very close to a normal distribution, but it certainly looks more like one
than Figure 3-3. Table 3-3 shows this histogram as an empirical probability
distribution.
Table 3-3 Empirical Probability Distribution

Representing Uncertainty with Normal Distributions

Table 3-3 could be used as the probability distribution to represent the
demand during lead time. Table 3-3 is more granular and represents what
actually happened. However, the fact that we had one observation of 55
units, no observation of 56 units, and two observations of 57 units means
that if we were to use this distribution to represent demand during lead
time, there would be no chance of 56 units. That doesn’t seem to be a
reasonable assumption. So, on the other hand, we could simplify the
distribution to that of Table 3-2. The challenge here is to select the
appropriate intervals for the bin ranges. Different interval selections result
in different empirical probability distributions. These can be useful for
some purposes, but many times they are difficult to work with and are
therefore approximated by a normal distribution or some other continuous
distribution. In this case, we would approximate this with a normal
distribution with a mean of 49 units and a standard deviation of 12 units,
since this is the mean and standard deviation of the observations in Table
3-1.
The problem with the normal distribution is that it can have negative
values, which don’t make sense for demand during lead time. So, you
probably should not use the normal approximation if the probability of
negative values is greater than 0.01. To check this in Excel, use the function
=NORMDIST(0,49,12,1) which returns a value of 0.00002, much less than
1 percent. The “0” in the argument means less than zero, “49” is the mean,
“12” is the standard deviation, and “1” is a cumulative distribution. This can
be read as the following: The probability that a normal distribution with a
mean of 49 and a standard deviation of 12 will have a value less than 0 is
0.00002 or 2 chances in 100,000. If it is greater than 0.01, then the gamma
distribution is an alternative. We discuss this distribution later in the
chapter.

INVENTORY REPLENISHMENT PROCESSES

In this chapter we use many graphs to represent inventory control systems.
As a supply chain manager you want to get to the point where you can draw
such graphs to illustrate your ideas or to ask questions. When two people
have strong capabilities in drawing such graphs, they can communicate
more effectively about subtle ideas and concepts. Unfortunately, based on
the authors’ experience, this is a rare capability. When you consider how
critical inventory management is to successful supply chain management,
this inability in supply chain managers is probably hurting the efficiency of
supply chains as well as stifling innovation. When people are vague about
their descriptions of a replenishment process, a process can seem really
innovative and compelling. However, some of these processes, after careful
and rigorous thought, are found to have fatal flaws. We hope to help you
improve your ability to rigorously discuss various aspects of replenishment
processes using graphs of such processes. This will greatly improve your
ability to communicate, create, and innovate. As a caveat, some of the
discussion around the graphs may become tedious, but we believe this is
necessary to gain a sufficient proficiency in communicating with graphs of
inventory processes.
Fundamentals of Inventory Replenishment Processes
With inventory models there are two things to consider: (1) continuous
review versus periodic review, and (2) continuous levels of inventory versus
discrete levels of inventory. In a continuous review5system, the inventory
level is continuously monitored, and as soon as a reorder point (ROP) is
reached, an order can be placed. In a periodic review system, orders can
only be placed at certain points in time. Inventory control systems can
assume continuous levels of inventory as in gallons of fuel or discrete levels
of inventory such as cases of candy bars. We begin by looking at continuous
review with continuous inventory levels and then cover all of the others
later. We begin with a (Q,ROP) system shown in Figure 3-4.

Figure 3-4 Graphical representation of a continuous review, continuous

inventory, (Q,ROP) inventory system

The horizontal axis is time and the vertical axis is number of units. Starting
from time = 0, we see the on-hand stock decreasing. The slope of this line is
the rate at which inventory is being depleted or the negative rate of
demand. For example, the slope of the line might be – 2 units per day, so
the rate of demand would be 2 units per day. In the graph you see that the
slope of the line does not change, implying that the rate of demand is
constant. We are beginning the discussion with the unrealistic assumption
of constant rate of demand. It is useful for our purposes because once we
allow for uncertain demand, everything becomes more complicated. Later
we drop our unrealistic assumption. As you continue to follow the On Hand
line down, you see that it eventually hits a dotted line labeled Reorder
Point. The reorder point in a (Q,ROP) system is the number of units of
inventory position where the order should be placed.

Inventory Position
Recall that in this process orders are not triggered by a point in time but
rather by a number of units of inventory position. Recall that inventory
position is the on hand plus on order minus back-orders. Since there are no
back-orders and no outstanding orders (by assumption), on hand is equal
to inventory position. So, on hand has hit the reorder point. Notice that the
inventory position jumps way up at the reorder point. That is because now
the inventory position is on hand plus on order. The amount on order is Q
in the (Q,ROP) process. So, the dotted line is inventory position, and the
solid line is on hand. Also, notice where Lead Time is written below the x-
axis. The lead time is between when the order is placed and when the
inventory is received and available for use.
Notice that during the lead time the inventory position is the same slope as
the on hand but shifted up by Q units.6 Also notice that at the end of the
lead time, the inventory position and the on hand come back together. That
is because, in this example, there are no longer any outstanding orders. The
bracket ( { ) to the left of the point in time when the order is received
represents the amount received, or the order quantity, Q. Notice that there
are two other intervals of time noted below the x-axis, the Time Between
Orders and the Time Between Replenishments. Since we have constant
demand, and constant lead time, the time between replenishments and the
time between orders is the same. Also, since we have constant and known
demand and lead time, there is no need for safety stock, so you will notice
that as soon as on hand hits zero, the replenishment arrives and is available
for use.
Figure 3-5 is similar to Figure 3-4 in that it is continuous review,
continuous inventory, known and constant demand and lead time, and a
(Q,ROP) inventory process. In this case, we now have numbers associated
with it. Again, the horizontal axis is time and the vertical axis is units. In
this example, the lead time is four days (L=4 days), and the sales per day is
10 units per day. So this means that the sales during the lead time is 10
units per day times 4 days or 40 units. Therefore, to keep from running out
of stock, we must place an order when the inventory position is 40. Hence,
the reorder point is 40 units (ROP = 40). In this example, the order
quantity is 100 (Q=100 units). So (Q,ROP) = (100 units, 40 units). On the
graph, the reorder point is shown as a red dashed line. Let’s start at the
point in time designated as t=0, where the order is placed. We call this
order number one. As you can see, the inventory position goes up by 100
units. So now the inventory position is 100 units + 40 units = 140 units.
Now, at t=4 days, the order is received and the on hand and inventory
position go up to 100. Now, to calculate the number of days between
replenishments T we simply divide the order quantity by the average sales
per day, or T = Q/d = 100 units per order / 10 units per day = 10 days
between replenishments. So our time between replenishments is 10 days
and our lead time is 4 days, that is, L = 0.4T. The ratio of lead time to time
between replenishments is an important variable we consider later in this
chapter.

Figure 3-5 Numerical example of a graphical representation of a continuous

review, continuous inventory, (Q,ROP) inventory system

Figure 3-6 is a replenishment process identical to the one we described

in Figure 3-5 except that the lead time is 10 days instead of 4 days. Now in
this situation the lead time is equal to the time between replenishments,
and the ratio of lead time to time between replenishments is one. In this
situation, notice that the inventory position is always 100 units (Q) above
the on hand. Let’s see why. Look at t = 0 days when an order is received. At
the same time the order is received, an order has to be placed because it
takes just as long to receive an order as it does to go through the amount of
inventory received in a replenishment.
Figure 3-6 Numerical example of graphical representation of a continuous
review, continuous inventory, (Q,ROP) inventory system where lead time is 10

Figure 3-7 is once again the same as Figure 3-5 and Figure 3-6 except that
the lead time is 14 days, resulting in a lead time to time between
replenishments ratio of 1.4. Since the lead time is greater than the time
between replenishments, the inventory position is always greater than the
on hand. Notice that at t = 0 days, an order is placed when another order is
already outstanding. At t = 4 days, the previous order is received, and at t =
14 days the order that was placed at t = 0 is received. In this case the lead
time and the demand are known and constant, but when they are uncertain
and variable, having a lead time to time between replenishments ratio that
is greater than 1 is difficult to manage. The reason for this is that you are
placing orders while there are outstanding orders you have not received.
Figure 3-7 Numerical example of graphical representation of a continuous
review, continuous inventory, (Q,ROP) inventory system where lead time is 14

DEMAND DURING LEAD TIME

Demand during lead time can grow if the rate of demand grows or if the
lead time grows or both. Similarly, demand during lead time can be
uncertain because demand is uncertain or because lead time is uncertain or
both. In practice, demand and lead time are uncertain. Let’s consider that
situation now.
We are still assuming a (Q,ROP) inventory system but we are now assuming
that demand during lead time is uncertain. Figure 3-8 is a probability
distribution of demand during lead time.
Figure 3-8 Probability distribution of demand during lead time

In Figure 3-8 the horizontal axis is demand during lead time and the
vertical axis (not shown) is the probability density. The area under the
curve is the probability. The total area under the curve is 1. Recall that
probabilities range from 0 to 1. The expected, or forecast, demand during
the lead time is designated on the probability distribution. Now, if you set
the reorder point at the expected demand during the lead time, then based
on how this probability distribution is drawn, you would stockout during
the lead time quite often. So, you might set the ROP higher, as designated
on Figure 3-8. The protection period in-stock (PPIS) metric is the area
under the curve up to ROP. They way this is drawn it looks like that is
around 90 percent of the area, so the PPIS = 0.90. Recall that you cannot
run out of stock in a (Q,ROP) continuous review inventory system at any
time except the lead time. So, for this process, PPIS is the probability of
staying in stock during the lead time, which in this case is the protection
period. Notice that the reorder point is made up of the expected demand
during the lead time plus the safety stock. Recall that safety stock is the
expected number of units on hand when the replenishment arrives and is
available for use. Safety stock is used to guard against stockouts. If you
increase safety stock, you increase PPIS.
If we used a normal distribution to represent the expected demand during
the lead time, we would need to estimate the mean and the standard
deviation of demand during the lead time. We could do this a number of
ways. We could look back over time and take the mean and standard
deviation of demand during the lead time. If we were to do that, we could
use the Excel function =NORMINV(PPIS,MEAN,STANDARD DEVIATION)
to get the ROP that has a target PPIS, given the mean and standard
deviation of demand during the lead time. We could also use forecasting
models. For example, we could forecast demand per day and then calculate
the standard deviation of forecast error. We would then need to calculate
the mean and standard deviation of lead time. Then to calculate the
expected demand during the lead time we would multiply the forecast of
demand per day by the average lead time. This would give us the expected
demand during the lead time. Next to calculate the standard deviation of
demand during the lead time we could use the following formula:7

Where
σ DDLT = standard deviation of demand during lead time
 = average lead time
 = standard deviation of forecast error
 = demand forecast per period
σL = standard deviation of lead time
Notice that we are using the demand forecast per period as well as the
standard deviation of forecast error. One benefit to using the standard
deviation of forecast error is that if you can forecast it, you do not really
need safety stock. If you are using safety stock to guard against uncertainty
and you can forecast demand accurately, even when it has a lot of
variability, you don’t need as much safety stock as might be implied by the
standard deviation of demand.
Table 3-4 gives forecast errors for forecasts of demand for 60 days and the
lead times associated with 60 orders.
Table 3-4 Forecast Errors for Forecasts of Demand

If you calculate the standard deviation of forecast error you will find that it
is 8 units per day; if you calculate the average and standard deviation of
lead time, you will find that it is 1 day and 0.18 days, respectively. Suppose
the forecast is currently 49 units per day, then the standard deviation of
demand during the lead time is

This gives 12 units per day. Now, the forecasted demand during lead time is
the forecast of 49 units per day, times the expected lead time of one day, so
the forecasted demand during the lead time is 49. We could then use the
Excel function =NORMINV(PPIS,MEAN,STANDARD DEVIATION) to get
the ROP. So, if we want a PPIS = 0.95, then =NORMINV(0.95,49,12)
returns 69 units. That is, if we have a (Q,ROP) process and we order when
the inventory position is 69 units, then we will be in stock during the
protection period 95 percent of the time.
You do not need to use the formula

if you observe demand during lead time for each order, as we did in Table
3-1.
In that case, we can directly use the average and standard deviation we
calculated from the Table 3-1. Recall that for Table 3-1, the average was 49
and the standard deviation was 12. So, if we want a PPIS = 0.95, then
=NORMINV(0.95,49,12) returns 69 units. That is, if we have a (Q,ROP)
process and we order when the inventory position is 69 units, we will be in
stock during the protection period 95 percent of the time.
Figure 3-9 shows a graph of on-hand inventory with a probability
distribution of demand during lead time superimposed upon it.
Figure 3-9 Graph of on-hand inventory with a probability distribution of
demand during lead time superimposed upon it

As we move the reorder point up on the graph, more and more of the
distribution of demand during lead time is encompassed. Looking at the
probability distribution, the PPIS is the area under probability distribution
curve from the ROP to the horizontal axis.
Figure 3-10 is another example that illustrates the expected demand during
lead time as well as the safety stock.
Figure 3-10 Expected demand during lead time and safety stock

You can see from this graph that if the ROP were kept constant and the
expected demand during the lead time increased, it would effectively reduce
safety stock, assuming the amount of uncertainty remained the same or
increased.
Figure 3-11 illustrates the (T,OUL) replenishment process. The horizontal
axis is time and the vertical axis is units, in this case, bars of soap. The
dashed line across is the order up to level (OUL); q1 is the first order, and
q2 is the second order; the dotted line is the point in time where a review is
made; the bold dotted line is when a replenishment is received and
available for use; the bold double-headed arrows represent the lead time;
and the regular double-headed arrows represent the time between reviews.
Regarding the graph lines, the solid squiggly line represents on-hand
inventory; and the dash-dotted graphed line represents inventory position.

Figure 3-11 (T,OUL) replenishment process

Starting half way through the review interval you can see that there is no
outstanding order and that on hand is decreasing. The squiggly is not
straight because it shows that demand is varying over time. When we get to
the first review point, q1 is ordered. You can see that inventory position is
greater than on-hand inventory during the lead time. At the end of the lead
time the inventory position and on hand come back together and on hand is
increased by q1. This process continues. You can see that in the
second replenishment cycle the on-hand inventory is decreasing at a lower
rate than in the first replenishment cycle. That is because demand is not as
high. Consequently, less is ordered at the second review. As you can see, q2
< q1.
Figure 3-12 illustrates the point that, unlike the (Q,ROP) process, in the
(T,OUL) process you can run out of stock at any time during T and L. 8

Figure 3-12 With the (T,OUL) process you can run out of stock at any time
during T and L.

Looking at Figure 3-12 when q2 is ordered, we see that sales picked up

dramatically, causing a stockout very quickly. Then at the end of L, we
received the order of q2. At that point we wished we would have ordered
much more, but it is too late. We again stockout quickly after receiving q2
because demand is still high. We remain out of stock through the next
review, until the end of the next lead time. This demonstrates that the
protection period for the (T,OUL) process is T+L. Hence, when we set
safety stock we must do it with respect to the entire protection period.
Next, to estimate the standard deviation of demand during the lead time
plus the review interval we could use the following formula:

Where
σDDLT = standard deviation of demand during lead time
 = average review interval
 = average lead time
 = standard deviation of forecast error
 = demand forecast per period
σT = standard deviation of review interval
σL = standard deviation of lead time
We could then use the Microsoft Excel function =NORMINV(PPIS,MEAN
of T+L,STANDARD DEVIATION over T+L) to get the OUL.
We have been assuming a normal distribution; however, one could also
assume a gamma distribution. The advantage of the gamma distribution is
that it starts at zero, whereas the normal distribution goes from negative
infinity to positive infinity.
There are two parameters you need for the gamma distribution, alpha (α)
and beta (β).

If you are calculating ROP, the estimates of alpha and beta should be based
on L. If you are calculating OUL, the estimates of alpha and beta should be
based on T+L. You could then use the Excel function
=GAMMAINV(PPIS,ALPHA,BETA) to get ROP or OUL. For slower moving
SKUs, the gamma distribution is probably better than the normal
distribution for technical reasons we do not explore here.
Returning to the earlier example of the (Q,ROP) process if we use the
gamma distribution we have the following:
This returns 70 units. Recall the normal returned 69 units. However now
let’s consider a mean of 1 and a standard deviation of 5. In this case the
normal distribution gives an ROP of 5, and the gamma gives an ROP of 9.
Worse, this normal distribution would have more than 40 percent of the
observations below zero, whereas the gamma has no observations below
zero.
You might also want to use a discrete probability distribution such as the
Poisson. The problem is that in Excel, there is not an inverse function so
you have to create a table. Suppose you have a Poisson distribution with a
mean demand during lead time of 0.5 units. Then to calculate your PPIS in
Excel for a given level of demand during lead time DDLT you would use
=POISSON(DDLT,mean,1). So in this example, if you wanted to know the
PPIS for a DDLT of 2, then =POISSON(DDLT,0.5,1) would return 0.986.
You could create a table like Table 3-5.

Table 3-5 Cumulative Poisson Distribution

From the table you could find the PPIS from the Cumulative Poisson
column and then find the corresponding demand during lead time in the
Demand column to find the ROP.
You probably noticed that =POISSON(DDLT,mean,1) does not require
standard deviation. The reason for this is that the Poisson distribution has a
variance that is equal to the mean. Also, in =POISSON(DDLT,mean,1), the
last argument, “1”, tells Excel that we are looking for the cumulative
Poisson distribution. If we put a zero in instead, it would return the
probability of that observation, not the cumulative probability. This is
shown in Table 3-6.
Table 3-6 Probability Mass

So, if we set ROP to 2, we will have a PPIS of 0.986. The probability of

selling 2 during the lead time is found in the last column, 0.076.
In reality, many hybrid replenishment processes exist. Let’s first consider a
hybrid to the (T,OUL) replenishment process. Suppose we took an existing
(T,OUL) process and added a reorder point to it. So, the idea would be that
when we got to a review point, we would only order the difference between
the inventory position and the OUL if the inventory position were below the
ROP. Would the new (T,OUL,ROP) process have more, less, or the same
average inventory level? Without the ROP, we would order at every review
time, but with the ROP we might not order at some of the review times.
Consequently, we would expect less inventory on average if a ROP were
added to an existing (T,OUL) process. There are many other hybrid
processes as well. Many of them are difficult to model, so it is good to
understand these two example replenishment processes (T,OUL) and
(Q,ROP) and to think about how variations affect inventory levels and
protection periods in particular. When a new process is created, it is always
good to think carefully about how new steps affect risk of failure and the
protection period in general.
As we have already mentioned, being able to represent your ideas in an
inventory graph can facilitate careful and rigorous thinking about how the
process will perform and where it will have potential failures. When you
draw these graphs you can use them to brainstorm various scenarios that
might occur. This can help you think through how various steps affect the
protection period. It can be more helpful if you have a colleague who is also
competent in inventory management and who can also use inventory
graphs to discuss new inventory systems or changes to existing systems. In
addition to drawing inventory graphs, it is also useful to have someone who
can build a discrete event simulation model of the process.
Discrete event simulation can be done in Excel or a discrete event
simulation software package. It allows you to model business processes
such as replenishment processes and allows for uncertainty to be brought
into things like demand and lead time. After a lot of discussion with the
inventory graphs, carefully designed flow charts can be created. These are
used to then create the discrete event simulation. In the process of
developing the discrete event simulation of the inventory process, questions
will arise about details of how the process works. The actual development of
the discrete event simulation requires a great deal of specificity. Usually the
first few times these models are run, problems occur because something
has not been specified. This leads to additional refinement of thinking
about the process that is being created or modified. In the previous
discussion about (T,OUL) and (Q,ROP) processes, we gave specific
formulas for estimating OUL or ROP. Even for these processes, for many
distributions, we cannot write a specific formula for OUL or ROP. In those
cases, discrete event simulation is needed. It gets even more complicated
for many hybrid processes, and then discrete event simulation is about the
only alternative. There are many different levels of skill in discrete event
simulation. However, it is useful for someone involved in inventory
management to at least be able to make rough cut discrete event simulation
models in Excel.

EXPECTED UNITS OUT PER REPLENISHMENT

CYCLE
We now return to the (Q,ROP) process and consider the optimal order
quantity, but first we must consider the total cost of a given order quantity.
For a given order quantity, the expected cost of carrying cycle stock is
(Q/2)hc.9 In addition, the expected cost of carrying safety stock is (ROP –
EDDLT)hc. The number of orders placed per year is (D/Q), so if the cost of
each order is S, then the annual ordering cost is (D/Q)S. During each lead
time there is a chance of stocking out and losing sales. The expected
number of units out of stock per replenishment cycle is

This is called the loss integral.10 We show you how to calculate this with a
normal distribution in Excel.
In this formula, Z is the number of standard deviations above the mean
demand during lead time that is represented by the ROP. If you use
=NORMDIST(ROP,EDDLT, σDDLT, 1) to find the PPIS, you can then use
=NORMSINV(PPIS) to find Z. Then you can apply the loss integral
formula.
Using the example from Table 3-1, suppose we set ROP to 52 units.
=NORMDIST(ROP,EDDLT, σDDLT, 1)
=NORMDIST(52,49,12,1) returns a value of 0.5987. Then using
=NORMSINV(PPIS)
=NORMSINV(0.5987) we get 0.25, which is the value of Z.
Now, to get the expected number of units out of stock per replenishment
cycle, we have

This returns a value of 3.4 units per replenishment cycle. Now, suppose the
expected cost of a unit out of stock is $10 per unit out of stock. Then every
time we place an order our expected cost of lost sales is about $34. This is
an additional ordering-related cost. Suppose you only use truckload 11 for
transportation and that each truckload costs $150. Other ordering related
costs are $20, including accounts payable variable costs, receiving, and so
on. Then the total cost associated with placing an order is $150 + $34 +
$20 = $204 per order.

TOTAL ANNUAL COST AS A FUNCTION OF

ORDER QUANTITY
Since there are (D/Q) replenishments per year,12 the expected number of
units out of stock per year is (D/Q)U(ROP). Suppose the cost per unit is m,
then the expected cost of out of units out of stock per year is
m(D/Q)U(ROP). The in-transit holding cost is (L/365)*D*hc. If lead time is
in days, then 365 days is used for the denominator; if lead time is in weeks,
then 52 weeks is used for the denominator, and so on. So the expected cost
is

We now discuss where transportation costs fit into this analysis. Truckload
(TL) costs are based on point-to-point service. That is, the amount charged
for the transportation is based on the rate a carrier changes from point A to
point B. If TL is used, the same cost is incurred regardless of how much is
shipped as long as it is less than the TL capacity. In this case, the cost of a
TL is added to the ordering cost because each time an order is placed, the
transportation cost of a TL must be paid. On the other hand, if less than
truckload (LTL) is used, then the cost is based on the weight. LTL rates 13 are
first based upon product class from National Motor Freight Classification
(NMFC) published by the National Motor Freight Traffic Association
(NMFTA). Then the carrier’s tariff is used based on the origin and
destination. Next the rates are discounted by the carrier. Finally the cost is
based on the weight shipped. Consequently, transportation is based on
weight shipped, which can be translated into a cost per unit. As a result, the
transportation cost is added to the value of the item, c. So, if TL is used,
transportation costs are a part of the ordering cost, whereas if LTL is used,
transportation costs are a part of the unit cost. In general, if the
transportation cost is based on the order not on the quantity, the
transportation cost goes in the ordering cost, whereas if it is based on the
amount shipped, it goes in the unit cost.
Taking the derivative of the total cost function above with respect to Q, and
setting it equal to zero, we have

Solving for Q, we have the EOQ.14

Taking the second derivative we find

Meaning that the function is convex, so we have found a global, unique
minimum.
All of this can be done with discrete empirical distributions as well.
The optimal solution may result in using TL but not filling the truck all the
way. That is, it might be less expensive to use TL than LTL, and, at the same
time, it is not optimal to utilize the transportation to 100 percent. Usually,
however, more than one item is shipped on the truck.
When evaluating ordering decisions and transportation decisions, there is a
trade-off between inventory holding costs and transportation costs that
must be taken into account.
Let

and

So B is the fixed cost of ordering, and I is the amount of inventory.

Figure 3-13 illustrates the cost trade-offs associated with various levels of Q
in a (Q,ROP) continuous review replenishment process.

Figure 3-13 Cost tradeoffs

Generally, the cost curve C(Q) is relatively flat at the bottom. Hence, being
off target doesn’t have a large impact on total cost. Being off the optimal
order quantity in terms of ordering too much has a slow rate of growth in
cost, whereas, ordering too little can increase total cost dramatically at
some point. As Q increases, the component of Ihc that is increasing is only
the cycle stock, not the safety stock or the in-transit stock.

Furthermore, Ihc increases linearly. However, as Q decreases, 

increases at an increasing rate.
The following example is important because it illustrates much of what has
been talked about in this chapter, but it also shows an important caveat
when using the EOQ or other optimization models. It also brings
transportation costs into the discussion. So, the following example is not
just an illustration of how to use what we have learned, it actually contains
new material that is easiest to explain in the context of an example.
A regional retailer, Value Dime and Five has one distribution center that
serves 500 stores. It only sells one SKU of toilet paper, its private label
(Hunter TP) 2 ply extra coarse toilet paper with 543 sheets per roll (white
with bold rough embossing). It replenishes stores in case pack quantities,
and each case contains 80 rolls. Value Dime and Five only buys it by the
truckload, which holds 560 cases, and pays $40 per case. Demand during
lead time is approximately normally distributed with a mean of 80 cases
and a standard deviation ( σDDLT ) of 30 cases. Since the lead time is usually a
day and average daily demand faced by the distribution center is 80, Value
Dime and Five orders when inventory position is 100. This might seem
strange, but the company has a good reason for it. Stores typically have
plenty of inventory, so even if Value Dime and Five is late on filling an
order from a store, the stores usually don’t run out of stock. Any orders that
can’t be filled when received from a store are filled later, as soon as the
inventory is available in the distribution center. The transportation cost per
truckload is $400. The owner of Value Dime and Five recognizes that
transportation costs are high, so from an efficiency perspective he wants to
maximize truck utilization and he only buys full truckloads. Based on his
weighted average cost of capital and damage analysis, he estimates his
inventory carrying cost factor to be about 0.25 of the value of inventory per
year. He estimates all other costs of ordering associated with purchasing,
accounts payable, receiving, and so on to be about $50 per order. For each
case that is back-ordered by a store because the distribution center doesn’t
have it in stock, the cost is about $5 per case and is the result of
administrative workarounds. The terms of sale are FOB Origin, Freight
Prepaid because Value Dime and Five has a much larger transportation
spend, allowing it to get better deals on transportation rates. Since it is FOB
Origin, Value Dime and Five owns the inventory in transit. Value Dime and
Five estimates its in-transit inventory carrying cost factor to be about 0.23
of the value of inventory per year. Value Dime and Five has a new analyst
who is pushing the idea that the company should not focus on maximizing
transportation utilization, but rather should focus on minimizing total cost.
The analyst is proposing the company use the economic order quantity
model in determining how much to order at a time rather than just
ordering a truckload. Value Dime and Five is open every day of the year.
Average daily demand for the distribution center is 80 cases per day, so
annual demand is 80 × 365 = 29,200. Since Q = 560 cases and D = 29,200
cases per year, the expected number of orders per year is D/Q = 52. We
begin by looking at costs associated with ordering. Given the expected
number of orders per year, the annual ordering cost will be 52 × $50 =
$2,600 per year, and the annual transportation cost will be 52 × $400 per
truckload = $20,800 per year. The next component we need to consider
that occurs each time an order is placed is the back-ordering cost. To
calculate the expected number of cases back-ordered per replenishment
cycle, we first need to make sure we can use the normal distribution. Many
tests exist to check this, but we are simply looking to make sure that no
more than 1 percent of the probability density is below zero. To do this we
use the following in Excel.
=NORMDIST(0,EDDLT, σDDLT, 1)
=NORMDIST(0,80,30,1)
=0.004
This shows that less than 1 percent of the probability density is below zero;
in fact, less than a half of 1 percent is below zero in this case.
We now calculate the expected number of cases back-ordered per
replenishment cycle. First we must calculate the PPIS metric.
=NORMDIST(ROP,EDDLT, σDDLT, 1)
=NORMDIST(100,80,30,1) returns a value of 0.75. This means they should
run out of stock during the lead time about 25 percent of the time. Of
course the lead time is only one day and stores tend to have plenty of
inventory, based on the example. Then, using PPIS we can find Z, which we
need in the calculation of the loss integral.
=NORMSINV(PPIS) =NORMSINV(0.75) we get 2/3 (0.66), which is the
value of Z.
Now, to get the expected number of units out of stock per replenishment
cycle, we have

This returns 4.5 cases per replenishment cycle. For each case back-ordered,
there is a $5 cost, so the cost of back-ordering per replenishment cycle is
4.5 × $5 = $23 per cycle. Since there are 52 orders per year, the annual cost
is 52 × $23 = $1,179.
So the total ordering related costs, ordering costs plus transportation costs,
plus back-ordering costs are $2,600 + $20,800 + $1,179 = $24,579 per
year. Clearly, the preponderance of the ordering related costs are
transportation costs in this example, but there are many situations where,
especially when there are lost sales instead of back-orders, lost sales costs
might be the highest of the costs. It would depend on the product
characteristics and the replenishment process and parameters, but at least
in this example, transportation costs are clearly dominating.
Now we look at the inventory related costs, starting with cycle stock.
Since each truckload is 560 cases, Q = 560 cases and the expected cycle
stock is Q/2 = 280 cases. The cost of each case is $40, so the expected
investment in cycle stock is 280 cases × $40 per case = $11,200, and the
expected annual cost of holding cycle stock is $11,200 × 0.25 = $2,800.
Safety stock is (ROP – EDDLT) = 100 cases – 80 cases = 20 cases. So, the
investment in safety stock is 20 cases × $40 per case = $800, and the
expected safety stock holding cost is 0.25 × $800 = $200.
The in-transit stock is (1 day/365 days) × 29,200 cases per year = 80 units.
So the investment in in-transit stock is 80 units × $40/case = $320 per
year, and the expected in-transit stock is $320 per year × 0.23 = $736 per
year. So the total holding costs are $2,800 + $200 + $736 = $3,736.
The total relevant inventory related cost is = ordering related costs +
inventory holding costs = $24,579 per year + $3,736 per year = $28,315 per
year.
Looking back at Figure 3-13, we see that since the cycle stock cost is $2,800
and the ordering-related cost is $24,579, we are clearly not ordering
enough at a time. As you can see, you are ordering below the economic
order quantity if ordering costs are greater than cycle stock inventory
holding costs. But here is an important caution. If we went to ordering two
truckloads at a time, our ordering costs would increase by the amount of a
TL. As it is drawn in Figure 3-14, this wouldn’t be enough to make a
difference. In fact, it turns out that a fully utilized TL is the optimal quantity
in this example. Now, if the other ordering related costs dominated the
transportation costs, this might not be the case.

Figure 3-14 Truckloads and total cost

If we use the EOQ model and are not careful, it would recommend an order
quantity that wasn’t feasible with the cost inputs. Let’s calculate the EOQ:
So the EOQ is recommending ordering 1,661 cases per order. But that is
1,661 cases per order / 560 cases per TL = 3 truckloads. However, the total
cost at three truckloads is higher than it is at one truckload. Figure 3-
14 shows because transportation costs dominate so much, one full
truckload really is the best solution. In Figure 3-14 notice that until the
order quantity equals one truckload, the annual ordering cost curve is
decreasing as would be expected, but then it jumps up by $400, the cost of
a truckload. The EOQ is recommending three truckloads, but that is not
feasible at $400, rather, three truckloads is $1,200.
Continuing with this example, we are saying that one truckload is optimal.
However, this is true if we don’t allow for the possibility of the use of other
modes. If rail is a possibility, then several truckloads fit in a railcar. For that
to be an option, the origin and destination need to be near a rail siding.
Using rail would certainly increase lead time, but in this case, it doesn’t
seem to be as big of a problem. Currently they are not holding very much
safety stock at the distribution center, so they could easily double or triple
their safety stock without making much of a difference on total cost. In fact,
they could and probably should consider using intermodal transportation,
which would increase the lead time but decrease the transportation cost,
although it wouldn’t increase the transportation unit capacity. The obvious
goal here is to reduce transportation costs.
Now, let’s keep everything else the same in this example, except increase
the cost per unit to $1,000. Clearly we are not talking about toilet paper
now, but this would change the EOQ to 332 cases, which is less than a full
truckload (60 percent utilization = 332 cases per order / 560 cases per
truckload). In this case it is optimal to underutilize the transportation
capacity.

Fill Rate
Let’s go back to Figure 3-13. Now, if you decrease Q, you are replenishing
more frequently and being exposed to more stockouts during the lead time,
so the expected number of units out of stock increases. That is, lost sales
increase. Consequently, a firm might want to increase safety stock to hit the
same fill rate. Keep in mind, if Value Dime and Five decreases Q, cycle
stock will decrease. So, if Value Dime and Five has to increase safety stock,
the question is will it have to increase it more than the decrease in cycle
stock? Also, when safety stock is increased, ROP is increased, so U(ROP)
decreases. So, let’s keep track of what is going on here. Q is decreased,
resulting in a lower fill rate (more lost sales) and lower cycle stock. To
address the lower fill rate, safety stock is increased by increasing ROP,
which in turn reduces U(ROP). The bottom line is that because the safety
stock increase is reducing U(ROP), the increase in safety stock needed to
offset the increase in the cost of lost sales is not as much as it otherwise
would be.
Figure 3-15 shows Q* is reduced to Q’ (labeled as “1” on the horizontal axis),
reducing the order quantity from its optimal level.

Figure 3-15 Optimal order quantity is reduced

This increases the cost as labeled on the vertical axis “1”. Now, as a result of
the reduction in fill rate, safety stock is increased (labeled “2”). Notice that
the inventory cost line shifts up. This is because, for any level of cycle stock,
we have more safety stock. But the (D/Q)B equation shifts down because B
is reduced. This results in a new optimal order quantity. As drawn the total
cost looks lower than it was for the original EOQ, Q*. However, it might be
higher or it might be lower, depending on the interaction between safety
stock and fill rate. The point is, the EOQ takes all of the other costs as given
and then optimizes. There are two problems with this: (1) The other costs
might not be at their optimal levels, and (2) it ignores the fact that over
time, competitors might increase service levels, increasing the optimal level
of service for the focal firm.

Trade-off Analysis
Neoclassical economic theory applied to inventory questions would propose
that an equilibrium exists between various costs. That is, when one of the
costs increases, firms respond by increasing other costs that are traded off
against them. For example, if inventory costs increase, they would increase
transportation spending to minimize the amount of inventory held; if
transportation costs increase, they spend more on inventory to minimize
those transportation costs. However, there is ample evidence to suggest
that competition actually induces disequilibrium. That is, the process of
competition itself is disequilibrating, because firms seek a comparative
advantage in their inventory management to achieve a competitive
advantage in market position. The competition they engage in teaches firms
which inventory approaches work best, because competitors that choose
effective or efficient inventory management methods achieve superior
performance.
When they suffer from inferior financial performance, firms recognize their
inferiority in terms of efficiency, value, or both, which then highlights their
inferior resources. If they recognize that their disadvantage emanates from
inventory management, firms likely (1) try to purchase or build the same
inventory management transaction system their competitors use, (2) try to
purchase or build the same inventory management decision support system
their competitors use, (3) purchase or build more innovative inventory
management transaction systems or decision support systems, (4) copy or
create more innovative business processes, (5) copy or create superior
management capabilities, (6) hire managers away from competitors, (7)
create an organizational culture similar to that of their competitors, and/or
(8) copy or innovate on their inventory positioning. In this process of
working to match or beat the competition, firms constantly face the threat
of inferior performance, so they all try to match or beat the competition.
This results in what seems like a constant release of new versions of
inventory management systems, ongoing recruitment of logistics executives
from other leading firms, persistent publication of new inventory
management ideas in trade and academic publications, and perpetual
hiring of management consulting firms that focus on inventory
management. On the other hand, in the short run, optimal inventory levels
probably do exist, but one must be continually revising inventory policies in
light of competition and consumer behavior.
Figure 3-16 shows the trade-off between inventory holding costs and
ordering costs when LTL is used.

Figure 3-16 Trade-off between inventory holding costs and ordering costs when
LTL is used

On the horizontal axis, we see the intervals based on LTL weight breaks.
For LTL, between an origin and destination, the transportation rate
increases at a decreasing rate. Consequently, the unit cost of the item
follows the same patterns and so does the inventory holding cost per unit of
inventory. L5C stands for “less than 500 lbs.,” 5 C stands for “above 500
lbs.,” 1 M stands for “greater than 1,000 lbs.,” and so on. These are typical
weight breaks in LTL. This causes the inventory holding cost curve to be
concave to the origin. In Figures 3-15 and 3-16, which did not involve LTL,
inventory costs were linear in Q. This effect that LTL has causes the optimal
order quantity to be higher than the point where inventory holding costs
are equal to ordering costs, as in Figures 3-15 and 3-16.
In Figure 3-17, the vertical dashed line is where we switch from LTL to TL.
Now our transportation cost is not a part of the unit product cost c because
it doesn’t matter how much you put in the truck, you still pay for the
truckload. So when this happens, the TL cost becomes an ordering cost
because every time you order, you must pay for a TL. So, the Ihc shifts
down to the bold Ihc, and the (D/Q)B shifts up to the bold (D/Q)B. If this is
the optimal place to shift to TL, the total cost should not go up as fast after
this quantity.

Figure 3-17 From LTL to TL

In Figure 3-18 we see the result of reducing the ordering cost per order
from B to B’; namely, it reduces the optimal order quantity.
Figure 3-18 Reduction in the optimal order quantity

Notice that the entire ordering cost curve shifts down. If firms are
attempting to reduce setup costs, make purchasing more efficient, reduce
invoice match rate errors,15 improve the efficiency of receiving, accounts
payable, and other related processes, then we should expect B to be
reduced. Better transportation processes for reduction in TL costs, such as
intermodal, can also reduce B. Improved execution that results in lower
expected number of units out of stock per replenishment cycle for a given
level of ROP will reduce B. B can also be reduced if the cost of a stockout
per unit of out of stock is decreased. This can occur if substitutes are
available that did not used to be available, the margin on items decreases,
competition among firms selling the product decreases, and other events
affect the expected cost of lost sales per unit of out of stock.
Figure 3-19 shows a company whose cost of holding inventory is being
reduced. That is, the cost per unit is reduced or the inventory holding cost
is going down.
Figure 3-19 Cost of holding inventory reduced

The cost per unit could be reduced through more efficient acquisition,
better negotiation, improved production processes, more competition in
the market providing the product, and others. The inventory holding cost
factor could be reduced as a result of more access to capital, thus lowering
the cost of capital, more efficient use of storage space, less product damage
while in storage, less pilferage and spoiling, and so on. In Figure 3-
18 and Figure 3-19, we have process improvement leading to changes in the
optimal order quantity. In Figure 3-18 we see process improvement leading
to a reduction in the optimal order quantity, whereas in Figure 3-19 we see
process improvement leading to an increase in the optimal order quantity.
Hence, we cannot provide an unequivocal result that firms that are
improving their processes should also be reducing inventory. That is clearly
not necessarily true based on our discussion so far. If the effects of process
improvement on ordering costs are strong enough, they may overcome the
effects of process improvement on inventory holding costs and result in the
optimal order quantity being reduced. But that is not generally true.
QUANTITY DISCOUNTS
Earlier when we solved the total cost function for the optimal order
quantity, the economic order quantity, the term representing the annual
purchase cost, Dc (which is annual demand D times cost per unit c), fell out
when we took the first derivative with respect to the order quantity, Q,
because c was not a function of Q. However, if there are quantity discounts,
then c is a function of Q. Many times the unit cost is referred to as a
variable ordering cost as opposed to the fixed ordering cost per order, S.
Quantity discounts can be applied in several ways, but we discuss two of
them: (1) all units quantity discount and (2) incremental units quantity
discount. With an all units quantity discount, the price of all units decreases
if the order quantity is greater than some designated amount, making the
total cost function discontinuous. With the incremental units quantity
discount, only the units above a certain quantity have the lower price
applied to them, so the cost function has a kink in it, but it is not
discontinuous.
For simplicity our analysis of the quantity discount considers the variable
ordering cost, the fixed ordering cost, and the cycle stock holding cost. In
addition, we only consider one price break for the quantity discount, but
the analysis for multiple price breaks in the quantity discount is essentially
just a straightforward extension of this analysis. Hence, our total cost
function is the following:

There are three key cases we want to consider, as illustrated in Figure 3-

20, Figure 3-21, and Figure 3-22.
Figure 3-20 Quantity discount below EOQ

Figure 3-21 Quantity discount above EOQ

Figure 3-22 EOQ lower than quantity discount

The vertical line in Figure 3-20 is the point of discontinuity of the total cost
function and is where the price break occurs. In Figure 3-20 it is clear that
taking the quantity discount makes sense. You can see from the graph that
even without the quantity discount, it would have been optimal to order a
quantity above the price break. This price break is not effective in terms of
incenting this firm to behave differently. The firm simply can order at a
lower price.
In Figure 3-21 the price break is effective because the price break incents
the firm to order more than it would otherwise. As you can see from the
graph, the total cost at the price break is lower than the total cost at the
EOQ level without the price break.
Finally, the price break in Figure 3-22 is not effective in the sense that it
does not incent the firm to order more. It is less expensive for the firm to
order the EOQ without the discount than to order an amount high enough
to receive the price break.

ENDNOTES
1. Over the years one of the authors has studied and taught from a number
of textbooks on inventory theory. Some of his favorites, which he has
learned the most from include the following: Hadley, George, and Thomson
M. Whitin. Analysis of Inventory Systems. New York: Prentice Hall, 1963.
Zipkin, P. Foundations of Inventory Management. Irwin, New York:
McGraw-Hill, 2000. Silver, Edward Allen, David F. Pyke, and Rein
Peterson. Inventory Management and Production Planning and
Scheduling. Vol. 3. New York: Wiley, 1998. Nahmias, Steven. Production
and Operations Analysis. New York: McGraw-Hill, 2005. Porteus, Evan
L. Foundations of Stochastic Inventory Theory. Palo Alto, CA: Stanford
University Press, 2002. This book is different from each of those books
because this book has a much more applied orientation; however, we are
indebted to these great works.
2. We use the terms stockout and out-of-stock interchangeably.
3. The idea here is that the difference in demand during lead time between
all of these observations is a result of randomness. That is, we are assuming
there is no seasonal variation, trend, or other causal factors such as changes
in prices or competitive responses. However, later in this chapter we
discuss competitive effects on inventory management.
4. Up to this point, and at future points, we are talking about demand
during lead time. However, we also consider situations where we estimate
demand and lead time separately.
5. It is called continuous review, but it should probably be
called continuous review and continuous response.
6. Later in this chapter, we see that inventory position can be above on
hand even before the order is placed. This happens when the lead time is
longer than the time between orders.
7. This formula comes from taking the variance of a random sum of

random variables. The random sum of random variables is given by   

where L is the random variable representing lead time, andXi is the random
variable representing demand per unit of lead time. Then you

calculate  .
8. Recall in the (Q,ROP) process you could only stockout during L.
9. The inventory carrying cost is h, so hc is the cost of carrying one unit of
inventory for one year. If the average inventory for the year is multiplied by
c, then you have the average investment in inventory.
10. Hadley, George, and Thomson M. Whitin. Analysis of Inventory
Systems. New York: Prentice Hall, 1963.
11. We will assume that you use truckload carriers even if the truck is not
utilized to 100 percent of capacity.
12. D is the expected annual demand.
13. Defee, C. Clifford, Joe B. Hanna, and Robert Overstreet. “LTL Pricing:
Looking Back to the Future.”Journal of Transportation Management 22.2
(2011): 45-58.
14. Wilson, R. H. “A Scientific Routine for Stock Control.” Harvard
Business Review 13.1 (1934): 116-129.
15. Invoice match rate errors occur when at least two of the following do
not match: invoice, receiving document, and purchase order. When these
do not match, they must be reconciled, and this can be labor intensive. A
certain percentage of orders have invoice match rate errors. If business
processes can be improved to reduce the percentage of errors that occur,
the cost of accounts payable per order is reduced.