. 7
( 15)


use of data forecasts, whereas the fourth policy does not require any data estimate.

4.8.1 The Newsboy Problem

In the Newsboy Problem, a resupply decision has to be made at the beginning of
a period (e.g. a spring sales season) for a single commodity whose demand is not
known in advance. The demand d is modelled as a random variable with a continuous
cumulative distribution function Fd (δ). Let c be the purchasing cost or the variable
manufacturing cost, depending on whether the goods are bought from an external
supplier or produced by the company. Moreover, let r and u be the selling price and
the salvage value per unit of commodity, respectively. Of course,

r > c > u.

There is no ¬xed reorder cost nor an initial inventory. In addition, shortage costs are
assumed to be negligible. If the company orders q units of commodity, the expected
revenue ρ(q) is
∞ ∞
ρ(q) = r min(δ, q) dFd (δ) + u max(0, q ’ δ) dFd (δ) ’ cq
0 0

q q
=r δ dFd (δ) + q dFd (δ) + u (q ’ δ) dFd (δ) ’ cq.
0 0

By adding and subtracting r q δ dFd (δ) to the right-hand side, ρ(q) becomes
∞ q
ρ(q) = rE[d] + r (q ’ δ) dFd (δ) + u (q ’ δ) dFd (δ) ’ cq, (4.35)
q 0

where E[d] is the expected demand. It is easy to show that ρ(q) is concave for q 0,
and ρ(q) ’ ’∞ for q ’ ∞. As a result, the maximum expected revenue is achieved
when the derivative of ρ(q) with respect to q is zero. Hence, by applying the Leibnitz
rule, the optimality condition becomes

r(1 ’ Fd (q)) + uFd (q) ’ c = 0,


where, by de¬nition, Fd (q) is the probability Pr(d q) that the demand does not
exceed q. As a result, the optimal order quantity S satis¬es the following condition:
r ’c
S) = .
Pr(d (4.36)
r ’u

Emilio Tadini & Sons is a hand-made shirt retailer, located in Rome (Italy), close
to Piazza di Spagna. This year Mr Tadini faces the problem of ordering a new bright
colour shirt made by a Florentine ¬rm. He assumes that the demand is uniformly
distributed between 200 and 350 units. The purchasing cost is c = ‚¬18 while the
selling price is r = ‚¬52 and the salvage value is u = ‚¬7. According to Equation
(4.36), Pr(d S) = (S ’ 200)/(350 ’ 200) for 200 S 350. Hence, Mr Tadini
should order S = 313 units. According to Equation (4.35), the expected revenue is
350 1
ρ(q) = 52 — 275 + 52 (q ’ δ) dδ ’ 18q = 34q,
350 ’ 200

for 0 200,
350 1
ρ(q) = 52 — 275 + 52 (q ’ δ) dδ
350 ’ 200
q 1
+7 (q ’ δ) dδ ’ 18q
350 ’ 200
= ’0.15q + 94q ’ 6000,

for 200 < q 350, and
350 1
ρ(q) = 52 — 275 + 7 (q ’ δ) dδ ’ 18q = ’11q + 12 375,
350 ’ 200

for q > 350. Hence, the maximum expected revenue is equal to ρ(313) = ‚¬8726.65.

4.8.2 The (s, S) policy for single period problems
If there is an initial inventory q0 and a ¬xed reorder cost k, the optimal replenishment
policy can be obtained as follows. If q0 S, no reorder is needed. Otherwise, the
best policy is to order S ’ q0 , provided that the expected revenue associated with this
choice is greater than the expected revenue associated with not producing anything.
Hence, two cases can occur:
(i) if the expected revenue ρ(S) ’ k ’ cq0 associated with reordering is greater
than the expected revenue ρ(q0 ) ’ cq0 associated with not reordering, then
S ’ q0 units have to be reordered;
(ii) otherwise, no order has to be placed.





tl tl

Figure 4.10 Reorder level inventory policy.

As a consequence, if q0 < S, the optimal policy consists of ordering S ’ q0 units
if ρ(q0 ) ρ(S) ’ k. In other words, if s is the number such that
ρ(s) = ρ(S) ’ k,
the optimal policy is to order S ’ q0 units if the initial inventory level q0 is less than
or equal to s, otherwise not to order. Policies like this are known as (s, S) policies.
The parameter s acts as a reorder point, while S is called the order-up-to-level.

If q0 = 50 and k = ‚¬400 in the Emilio Tadini & Sons problem, ρ(s) = ρ(S)’k =
‚¬8526.65 so that s = 277. As q0 < s, the optimal policy is to order S ’ q0 = 253

4.8.3 The reorder point policy
In the reorder point policy (or ¬xed order quantity policy), the inventory level is kept
under observation in an almost continuous way. As soon as its net value I (t) (the
amount in stock minus the unsatis¬ed demand plus the orders placed but not received
yet) reaches a reorder point l, a constant quantity q is ordered (see Figure 4.10).
The reorder size q is computed through the procedures illustrated in the previous
sections, by replacing d with d. In particular, under the EOQ hypotheses:

2k d
q= .
The reorder point l is obtained by requiring that the inventory level be nonnegative
during tl , with probability ±. This is equivalent to assuming that demand should not
exceed l during the interval tl . In the following, it is assumed that


• the demand rate d is distributed according to a normal distribution with expected
value d and standard deviation σd ;
• d and σd are constant in time;

• the lead time tl is deterministic or is distributed according to a normal distribu-
tion with expected value t¯l and standard deviation σtl ;

• the demand rate and the lead time are statistically independent.
The average demand rate d can be forecasted with one of the methods illustrated
in Chapter 2, while the standard deviation σd can be estimated as the square root of
MSE. Analogous procedures can be adopted for the estimation of tl and σtl .
Let z± be the value under which a standard normal random variable falls with
probability ± (e.g. z± = 2 for ± = 0.9772 and z± = 3 for ± = 0.9987). If tl is
deterministic, then

l = dtl + z± σd tl , (4.37)

where dtl and σd tl are the expected value and the standard deviation of the demand
in an interval of duration tl , respectively. If tl is random, then,

¯¯ ¯
l = d tl + z± σd t¯l + σt2 d 2 ,

¯¯ ¯

where d tl and σd tl + σt2 d 2 are the expected value and the standard deviation of
the demand in a time interval of random duration tl , respectively.
The reorder point l minus the average demand in the reorder period constitutes a
safety stock IS . For example, in case tl is constant, the safety stock is

IS = l ’ d tl = z± σd tl . (4.38)

Papier is a French retail chain. At the outlet located in downtown Lyon, the expected
demand for mouse pads is 45 units per month. The value of an item in stock is ‚¬4, and
the ¬xed reorder cost is equal to ‚¬30. The annual interest rate is 20%. The demand
forecasting MSE is 25. Lead time is 1 month and a service level equal to 97.7% is
required. On the basis of Equation (4.2), the holding cost is
h = 0.2 — 4 = 0.8 euros/year per item = 0.067 euros/month per item.
Therefore, from Equation (4.13),

2 — 30 — 45
q— = = 200.74 ≈ 201 items.
Moreover, σd can be estimated as follows:

σd = 25 = 5.



qi qi+1

qi qi+1


tl tl

ti ti+1 ti+2

Figure 4.11 Reorder cycle inventory policy.

From Equation (4.37), the reorder point l is
l = 45 + 2 — 5 = 55 units.

Consequently, the safety stock IS is

IS = 55 ’ 45 = 10 units.

4.8.4 The periodic review policy

In the reorder cycle policy (or periodic review policy) the stock level is kept under
observation periodically at time instants ti (ti+1 = ti + T , T 0). At time ti ,
qi = S ’ I (ti ) units are ordered (see Figure 4.11). The parameter S (referred to as
the order-up-to-level) represents the maximum inventory level in case lead time tl is
The periodicity T of the sampling (review period) can be chosen using procedures
analogous to those used for determining q — in the deterministic models. For instance,
under the EOQ hypotheses,
T= . (4.39)
The parameter S is determined in such a way that the probability that the inventory
level becomes negative does not exceed a given value (1 ’ ±). Since the risk interval
is equal to T plus tl , S is required to be greater than or equal to the demand in T + tl ,


with probability equal to ±. If the lead time tl is deterministic, then

S = d(T + tl ) + z± σd T + tl , (4.40)

where d(T + tl ) and σd T + tl are the expected value and the standard deviation of
the demand in T + tl , respectively. If the lead time is a random variable, then

¯ ¯
S = d(T + tl ) + z± σd (T + t¯l ) + σt2 d 2 ,
¯ 2

¯ ¯
where d(T + tl ) and σd (T + t¯l ) + σt2 d 2 are the expected value and the standard
¯ 2
deviation of the demand in T + tl , respectively.
The difference between S and the average demand in T + tl makes up a safety stock
IS . For example, if the lead time is constant,

IS = z± σd T + tl . (4.41)

Comparing Equation (4.41) with Equation (4.38), it can be seen that the reorder
cycle inventory policy involves a higher level of safety stock. However, such a policy
does not require a continuous monitoring of the inventory level.

In the Papier problem, the parameters of the reorder cycle inventory policy, com-
puted through Equations (4.39) and (4.40) are

2 — 30
T= = 4.47 months,
0.067 — 45

S = 45 — (4.47 + 1) + 2 — 5 — 4.47 + 1 = 269.54 units.
The associated safety stock, given by Equation (4.41), is

IS = 2 — 5 — 4.47 + 1 = 23.39 units.

4.8.5 The (s, S) policy
The (s, S) inventory policy is a natural extension of the (s, S) policy illustrated for the
one-shot case. At time ti , S ’ I (ti ) items are ordered if I (ti ) < s (see Figure 4.12). If
s is large enough (s ’ S), the (s, S) policy is similar to the reorder cycle inventory
method. On the other hand, if s is small (s ’ 0), the (s, S) policy is similar to a
reorder level policy with a reorder point equal to s and a reorder quantity q ∼ S. On
the basis of these observations, the (s, S) policy can be seen as a good compromise
between the reorder level and the reorder cycle policies. Unfortunately, parameters
T , S and s are dif¬cult to determine analytically. Therefore, simulation is often used
in practice.



qi qi qi+2 qi+2


qi+1 = 0
tl tl

ti ti+1

Figure 4.12 (s, S) policy.

Pansko, a Bulgarian chemical ¬rm located in Plovdiv, supplies chemical agents to
state clinical laboratories. Its product Merofosphine has a demand of 400 packages
per week, a variable cost of 100 levs per unit, and a pro¬t of 20 levs per unit. Every
time the manufacturing process is set up, a ¬xed cost of 900 levs is incurred. The
annual interest rate p is 20%. If the commodity is not available in stock, a sale is lost.
In this case, a cost equal to the pro¬t of the lost sale is incurred. The MSE forecast
equals 2500. The lead time can be assumed to be constant and equal to a week. The
inventory is managed by means of an (s, S) policy with a period T of two weeks. The
values s and S are selected by simulating the system for all combinations of s (equal
to 800, 900, 1000, 1100 and 1200, respectively) and S (equal to 1500, 2000 and 2500,
respectively). According to the results reported in Table 4.1, s = 1100 and S = 2000
are the best choice. This would result in an average cost per week equal to 612.7 levs.

4.8.6 The two-bin policy
The two-bin policy can be seen as a variant of the reorder point inventory method
where no demand forecast is needed, and the inventory level does not have to be
monitored continuously. The items in stock are assumed to be stored in two identical
bins. As soon as one of the two becomes empty, an order is issued for an amount equal
to the bin capacity.

Browns supermarkets make use of the two-bin policy for tomato juice bottles. The
capacity of each bin is 400 boxes, containing 12 bottles each. In a supermarket close


Table 4.1 Average cost per week (in levs) in the Pansko problem. The average ¬xed cost,
the average variable cost and the average shortage costs are reported in brackets.


s 1500 2000 2500

800 1120.8 625.0 994.9
(337.8 + 168.5 + 614.5) (224.8 + 236.8 + 163.3) (152.3 + 330.2 + 512.3)
900 644.7 622.9 908.9
(447.6 + 184.6 + 12.4) (225.0 + 236.8 + 161.0) (162.9 + 339.3 + 406.6)
1000 625.0 623.0 724.3
(450.0 + 184.9 + 0.0) (225.0 + 236.8 + 161.0) (197.9 + 375.7 + 150.5)
1100 635.0 612.7 634.6
(450.0 + 184.0 + 0.0) (229.7 + 239.1 + 143.9) (222.2 + 403.3 + 9.0)
1200 635.0 622.7 631.8
(450.0 + 185.0 + 0.0) (291.2 + 276.3 + 55.1) (224.9 + 406.8 + 0.0)

Table 4.2 Daily sales of tomato juice (in bottles) during the ¬rst week of
December last in a Browns supermarket.

Day Sales Inventory level

1 Dec 850 8510
2 Dec 576 7934
3 Dec 932 7002
4 Dec 967 6035
5 Dec 945 5090
6 Dec 989 4101
7 Dec 848 3253

to Los Alamos (New Mexico, USA) the inventory level on 1 December last was 780
boxes of 12 bottles each. Last 6 December, the inventory level was less than 400
boxes and an order of 400 boxes was issued (see Table 4.2). The order was ful¬lled
the subsequent day.

4.9 Selecting an Inventory Policy
It is quite common for a warehouse to contain several hundreds (or even thousands)
of items. In such a context, goods having a strong impact on the total cost have to be


managed carefully while for less important goods it is wise to resort to simple and
low-cost techniques.
The problem is generally tackled by clustering the goods into three categories (indi-
cated with the symbols A, B and C) on the basis of the average value of the goods
in stock. This method is often called the ABC technique. Category A is made up of
products corresponding to a high percentage (e.g. 80%) of the total warehouse value.
Category B is constituted by a set of items associated with an additional 15% of
the warehouse value, while category C is formed by the remaining items. Goods are
subdivided into these categories as follows: ¬rst, commodities are sorted by nonin-
creasing values with respect to the average value of the goods in stock; the items are
then selected from the sorted list, to reach the pre-established cumulated value levels.
On the basis of the 80“20 principle (or Pareto principle), category A usually con-
tains a small fraction (generally, 20“30%) of the goods whereas category C includes
many products. This observation suggests that the goods of categories A and B should
be managed with policies based on forecasts and a frequent monitoring (e.g. category
A by means of the reorder level inventory method and category B through the reorder
cycle inventory policy). Products in category C can be managed using the two-bin
policy that does not require any forecast.

The Walloon Transportation Consortium (WTC) operates a Belgian public trans-
portation service in the Walloon region. Buses are maintained in a facility located in
Ans, close to a vehicle depot. The average inventory levels, the unit values and the
total average value of the spare parts kept in stock are reported in Table 4.3. It was
decided to allocate to category A the products corresponding approximately to the
¬rst 80% of the total value of the stock, to category B the items associated with the
following 15%, and to category C the remaining commodities (see Table 4.4). It is
worth noting that category A contains about 30% of the goods, while each of the cat-
egories B and C accounts for about 35% of the inventory. The cumulated percentage
of the total value as a function of the cumulated percentage of the number of items
(Pareto curve) is reported in Figure 4.13.

4.10 Multiple Stocking Point Models
Good inventory policies for multiple interdependent stocking points can be very dif-
¬cult to devise. In this section a very simple model is described and analysed. In a
decentralized logistics system, a market is divided into n identical sales districts, each
of which is allocated to a warehouse, while in a centralized system every customer
is serviced by a unique facility. Under the EOQ hypotheses, the average inventory
levels of the two systems are linked by the following square-root law.

Property. If the EOQ hypotheses hold, and each warehouse in the decentralized
system services the same demand, then the total average inventory level I (n) in the


Table 4.3 Spare parts stocked by WTC.

Product Average Average unit Total average
code stock value (in euros) value (in euros)

AX24 137 50 6 850
BR24 70 2 000 140 000
BW02 195 250 48 750
CQ23 6 6 000 36 000
CR01 16 500 8 000
FE94 31 100 3 100
LQ01 70 2 500 175 000
MQ12 18 200 3 600
MW20 75 500 37 500
NL01 15 1 000 15 000
PE39 16 3 000 48 000
RP10 20 2 200 44 000
SP00 13 250 3 250
TA12 100 2 500 250 000
TQ23 10 5 000 50 000
WQ12 30 12 000 360 000
WZ34 30 15 450
ZA98 70 250 17 500

Cumulated % of the total value







Class A Class B Class C



Cumulated % of the
0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00 100.00
number of products

Figure 4.13 Pareto curve in the WTC problem.

decentralized system is
√ (1)
¯ ¯
I (n) = nI ,
where I (1) is the average inventory level in the centralized system.


Table 4.4 ABC classi¬cation of the spare parts in the WTC problem.

Total Total
Fraction Cumulated % average cumulated Cumulated
Product of total of the number value value % of the
code inventory of products (in euros) (in euros) total value Class

WQ12 3.3 3.30 360 000 360 000 28.87
TA12 10.8 14.10 250 000 610 000 48.92
LQ01 7.6 21.70 175 000 785 000 62.95 A
BR24 7.6 29.30 140 000 925 000 74.18
TQ23 1.1 30.40 50 000 975 000 78.19

BW02 21.1 51.50 48 750 1 023 750 82.10
PE39 1.8 53.30 48 000 1 071 750 85.95
RP10 2.1 55.40 44 000 1 115 750 89.47 B
MW20 8.2 63.60 37 500 1 153 250 92.48
CQ23 0.6 64.20 36 000 1 189 250 95.37

ZA98 7.6 71.80 17 500 1 206 750 96.77
NL01 1.6 73.40 15 000 1 221 750 97.98
CR01 1.8 75.20 8 000 1 229 750 98.62
AX24 14.8 90.00 6 850 1 236 600 99.17 C
MQ12 2.0 92.00 3 600 1 240 200 99.45
SP00 1.4 93.40 3 250 1 243 450 99.72
FE94 3.3 96.70 3 100 1 246 550 99.96
WZ34 3.3 100.00 450 1 2470,00 100.00

Proof. In the EOQ model the average inventory level is half the order size. Therefore,
I (1) is equal to 2 2kd/ h, where d is the demand of the whole market. In a decen-
tralized system, I (n) is the sum of the average inventory levels of the facilities, each
of which services 1/n of demand d. Hence, I (n) = 2 n 2k(d/n)/ h.

Kurgantora distributes tyres in Russia and Kazakhstan. The distribution network
currently includes 12 warehouses, each of which serves approximately the same
demand. In an attempt to reduce the total inventory level by 30%, the company
has decided to close some warehouses and allocate their demand to the remaining
facilities. Applying the square-root law, we see that the number of stocking points
should be reduced to 6, since

¯ ¯
I (12) = 12 I (1) ,

¯ ¯
I (n ) = n I (1) ,
¯ ¯
I (n ) /I (12) = n / 12 = 0.7,
n = 5.88.


4.11 Slow-Moving Item Models
As shown in the previous sections, a major issue for fast-moving product inventory
management is determining how often reorders should take place. On the other hand,
if demand is very low (e.g. a few units in 10“20 years), as in the case of spare parts
of a complex machinery (slow-moving products), the main issue is determining the
number of items to be purchased at the beginning of the machinery™s life cycle.
In this section we examine a model in which item purchase cost and shortage
penalties are taken into account while holding cost and salvage value (i.e. the value
of unused spare parts at the end of the machinery lifetime) are negligible. Let c and
u be the purchase cost of an item at the beginning of the planning horizon and during
the planning horizon, respectively (c < u). If n units of product are purchased at the
beginning of the planning period, and m units are demanded in the planning period,
the total cost is
C(n, m) = cn, if n m;
C(n, m) = cn + u(m ’ n), if n < m.
Let P (m) be the probability that m items are demanded. Then, the expected cost C(n)
in case n items are purchased is
∞ ∞
C(n) = C(n, m)P (m) = cn + u (m ’ n)P (m).
m=0 m=n+1

¯ ¯
C(n ’ 1) = C(n) ’ c + u[1 ’ F (n ’ 1)], (4.42)
¯ ¯
C(n + 1) = C(n) + c ’ u[1 ’ F (n)], (4.43)
where F (n) is the probability that n units (or less) are demanded. The minimum
expected cost is achieved if n— items are purchased at the beginning of the planning
¯ ¯
C(n— ’ 1) C(n— ), (4.44)
¯ ¯
C(n— + 1) C(n— ). (4.45)
Finally, combining Equation (4.44) and Equation (4.42), the following relation is
F (n— ’ 1) .
Similarly, combining Equation (4.45) and Equation (4.43) for n = n— gives
F (n— ) .
Consequently, the optimal number of items n— to be purchased is such that
F (n— ’ 1) F (n— ). (4.46)


Hydro Ltd uses ¬ve LIF03 generators in its hydroelectric power plants located in
Nigeria. Each piece of machinery has an average life of 20 years, during which the
expected number of engine failures is equal to 1.4. The cost of a spare part, purchased
when a generator is manufactured, is $60 000 while producing an additional unit costs
around $300 000. The failure process is modelled as a Poisson probability distribution
with expected value »,
» = 5 — 1.4 = 7 faults per life cycle.
Therefore, the probability P (n) that the number of demanded spare parts equals n
is given by
e’» »n
P (n) = , n = 0, 1, . . . ,
while the cumulative probability F (n) is
F (0) = P (0),
e’» »k /k! = F (n ’ 1) + P (n),
F (n) = n = 1, 2, . . . .

The values of P (n) and F (n) for n = 0, . . . , 10 are reported in Table 4.5.
u’c 300 000 ’ 60 000
= = 0.8,
c 300 000
on the basis of Equation (4.46), n— = 9 spare parts should be purchased and stocked.

4.12 Policy Robustness
The inventory policies illustrated in the previous sections often have to be slightly
modi¬ed in order to be used in practice. Fractional order sizes and shipment frequen-
cies have to be suitably rounded up or down (e.g. q — = 14.43 pallets should become
14 or 15 pallets). Fortunately, the total cost is not very sensitive to variations of the
order size around the optimal value. For the EOQ model the following property holds.

Property. In the EOQ model, errors in excess of 100% on the optimal order size
cause a maximum increase of the total cost equal to 25%.
Proof. Recall that the average cost in the EOQ model is given by
µ(q) ’ cd = kd/q + hq/2.

If q = q — = 2kd/ h (see Equation (4.13)), then

µ(q — ) ’ cd = 2kdh.


Table 4.5 Probability distribution of spare part demand in the Hydro Ltd problem.

n P (n) F (n)

0 0.0009 0.0009
1 0.0064 0.0073
2 0.0223 0.0296
3 0.0521 0.0818
4 0.0912 0.1730
5 0.1277 0.3007
6 0.1490 0.4497
7 0.1490 0.5987
8 0.1304 0.7291
9 0.1014 0.8305
10 0.0710 0.9015

If q = 2q — , then √
µ(2q — ) ’ cd = 5

µ(2q — ) ’ cd
= 1.25
µ(q—) ’ cd
1.25(µ(q — ) ’ cd), q ∈ [q — , 2q — ].
µ(q) ’ cd
Similarly, it can be shown that rounding the reorder size to the closest power of 2
(power-of-two policy) induces a maximum cost increase of about 6%.

In the Al-Bufeira Motors problem (see Section 4.4.2), the total cost associated with
¬ve shipments per year (i.e. q = 44 units),
800 — 220 216 — 44
µ(q — ) = + = 8752 dollars per year,
44 2
is higher than the optimal solution (µ(q — ) = 8719.63 dollars per year) only by 0.37%.

4.13 Questions and Problems
4.1 In most industrialized countries the average ITR is around 20 for dairy products
and around 5 for household electrical appliances. Discuss these ¬gures.

4.2 Modify the EOQ formula for the case where the stocking point has a ¬nite
capacity Q.


4.3 Modify the EOQ formula for the case where the holding cost is a concave
function of the number of items kept in inventory.
4.4 Devise an optimal inventory policy for the EOQ model with a ¬nite time horizon
TH .
4.5 Modify the EOQ formula for the case where the order size q is delivered by a
number of vehicles of capacity qv each having a ¬xed cost kv .
4.6 Draw the auxiliary graph used for solving the Sao Vincente Chemical problem
as a shortest-path problem.
4.7 Modify the Wagner“Within model for the case where the stocking point is
capacitated. Does the ZIO property still hold?
4.8 What is the optimal order quantity in the Newsboy Problem if the stocking
point is capacitated?
4.9 If type A products are overstocked, the total cost increases dramatically, while if
type C products are overstocked, the total cost does increase too much. Calculate
the cost increase whenever the inventory level of A products is increased by
20%. Repeat the calculation for C products.
4.10 Show that the power-of-two policy induces a maximum cost increase of about

4.14 Annotated Bibliography
An in-depth treatment of inventory management can be found in:
1. Zipkin PH 2000 Foundations of Inventory Management. McGraw-Hill, New
A simpli¬ed approach is available in:
2. Lewis CD 1998 Demand Forecasting and Inventory Control: A Computer Aided
Learning Approach. Wiley, New York.
An introduction to the use of simulation methods with some applications on the
inventory management is:
3. Law L and Kelton WD 2000 Simulation Modelling and Analysis, 3rd edn.
McGraw-Hill, New York.


Designing and Operating a

5.1 Introduction
Warehouses are facilities where inventories are sheltered. They can be broadly classi-
¬ed into production warehouses and DCs. This chapter deals with warehouse design
and operation, with an emphasis on DCs. In the following, a product is de¬ned as
a type of good, e.g. wine bottles of a speci¬c brand. The individual units are called
items (or stock keeping units (SKUs)). A customer order is made up of one or more
items of one or more products.

Flow of items through the warehouse. Warehouses are often used not only to
provide inventories a shelter, but also to sort or consolidate goods. In a typical DC,
the products arriving by truck, rail, or internal transport are unloaded, checked and
stocked. After a certain time, items are retrieved from their storage locations and
transported to an order assembly area. In the simplest case (which occurs frequently
in CDCs (see Chapter 1)), the main activity is the storage of the goods. Here, the
merchandise is often received, stored and shipped in full pallets (all one product)
and, as a result, material handling is relatively simple (see Figure 5.1). In the most
complex case (which occurs frequently in RDCs (see Chapter 1)) large lots of prod-
ucts are received and shipments, containing small quantities of several items, have to
be formed and dispatched to customers. Consequently, order picking is quite com-
plex, and product sorting and consolidation play a mayor role in order assembly (see
Figure 5.2).

Ownership of the warehouses. With respect to ownership, there are three main
typologies of warehouses. Company-owned warehouses require a capital investment
in the storage space and in the material handling equipment. They usually represent
the least-expensive solution in the long run in the case of a substantial and constant
demand. Moreover, they are preferable when a higher degree of control is required
to ensure a high level of service, or when specialized personnel and equipment are

Introduction to Logistics Systems Planning and Control G. Ghiani, G. Laporte and R. Musmanno
© 2004 John Wiley & Sons, Ltd ISBN: 0-470-84916-9 (HB) 0-470-84917-7 (PB)

Article A





Figure 5.1 The ¬‚ow of items through the warehouse; the goods are received and
shipped in full pallets or in full cartons (all one product).

Article A Article B Article C

Receiving Receiving Receiving

Holding Holding Holding

Picking Picking Picking

Batch forming



Figure 5.2 The ¬‚ow of items through the warehouse; the goods are received in full pallets
or in full cartons (all one product) and shipped in less than full pallets or cartons.

needed. Finally, they can be employed as a depot for the company™s vehicles or as a
base for a sales of¬ce. Public warehouses are operated by ¬rms providing services
to other companies on a short-term basis. As a rule, public warehouses have stan-
dardized equipment capable of handling and storing speci¬c types of merchandise
(e.g. bulk materials, temperature-controlled goods, etc.). Here, all warehousing costs
are variable, in direct proportion to the storage space and the services required. As
a result, it is easy and inexpensive to change warehouse locations as demand varies.
For these reasons, public warehouses can suitably accommodate seasonal inventories.



Holding Handling
15% 50%


Figure 5.3 Common warehouse costs.

Finally, leased warehouse space is an intermediate choice between short-term space
rental and the long-term commitment of a company-owned warehouse.

Warehouse costs. The total annual cost associated with the operation of a ware-
house is the result of four main activities: receiving the products, holding inventories
in storage locations, retrieving items from the storage locations, assembling customer
orders and shipping. These costs depend mainly on the storage medium, the stor-
age/retrieval transport technology and its policies. As a rule, receiving the incoming
goods and, even more so, forming the outgoing lots, are operations that are dif¬-
cult to automate and often turn out to be labour-intensive tasks. Holding inventories
depends mostly on the storage medium, as explained in the following. Finally, picking
costs depend on the storage/retrieval transport system which can range from a fully
manual system (where goods are moved by human pickers travelling on foot or by
motorized trolleys) to fully automated systems (where goods are moved by devices
under the control of a centralized computer). Common warehouse costs are reported
in Figure 5.3.

5.1.1 Internal warehouse structure and operations
The structure of a warehouse and its operations are related to a number of issues:
• the physical characteristics of the products (on which depends whether the
products have to be stored at room temperature, in a refrigerated or ventilated
place, in a tank, etc.);
• the number of products (which can vary between few units to tens of thousands);
• the volumes handled in and out of the warehouse (which can range between a
few items per month to hundreds of pallets per day).


Receiving zone

Storage zone

Shipping zone

Figure 5.4 Warehouse with a single receiving zone and a single shipping zone.

Typically, in each DC there are (see Figure 5.4)
• one or more receiving zones (each having one or more rail or truck docks),
where incoming goods are unloaded and checked;
• a storage zone, where SKUs are stored;
• one or more shipping zones (each having one or more rail or truck docks), where
customer orders are assembled and outgoing vehicles are loaded.
The storage zone is sometimes divided into a large reserve zone where products
are stored in the most economical way (e.g. as a stack of pallets), and into a small
forward zone, where goods are stored in smaller amounts for easy retrieval by order
pickers (see Figure 5.5). The transfer of SKUs from the reserve zone to the forward
zone is referred to as a replenishment. If the reserve/forward storage is well-designed,
the reduction in picking time is greater than replenishment time.

5.1.2 Storage media
The choice of a storage medium is strongly affected by the physical characteristics of
the goods in stock and by the average number of items of each product in a customer
order. Brie¬‚y, when storing solid goods three main alternatives are available: stacks,
racks and drawers. In the ¬rst case, goods are stored as cartons or as pallets, and
aisles are typically 3.5“4 m wide (see Figure 5.6). Stacks do not require any capital
investment and are suitable for storing low-demand goods, especially in reserve zones.
In the second case, goods are stored as boxes or pallets on metallic shelves separated
by aisles. Here quick picking of single load units is possible. When SKUs are moved
by forklifts, the racks (see Figure 5.7) are usually 5“6 m tall and aisles are around


Receiving Reserve Forward Shipping
zone zone zone zone

Figure 5.5 Warehouse with reserve and forward storage zones.

Figure 5.6 Block stacking system.

3.5 m wide. Instead, as explained in the following, in automated storage and retrieval
systems (AS/RSs), racks are typically 10“12 m tall and aisles are usually 1.5 m wide
(see Figure 5.8). Finally, in the third case, items are generally of small size (e.g.
metallic small parts), and are kept in ¬xed or rotating drawers.

5.1.3 Storage/retrieval transport mechanisms and policies
A common way of classifying warehouses is the method by which items are retrieved
from storage.


Figure 5.7 Rack storage.

Figure 5.8 An AS/RS.

Picker-to-product versus product-to-picker systems. The picking operations can
be made
• by a team of human order pickers, travelling to storage locations (picker-to-
product system);
• by an automated device, delivering items to stationery order pickers (order-to-
picker system).
Clearly, mixed solutions are possible. For instance, in picker-to-belt systems, the
items are retrieved by a team of human order pickers and then transported to the order
assemblers by a belt conveyor. Picker-to-product systems can be further classi¬ed
according to the mode of travel inside the warehouse. In person-aboard AS/RS, pickers



Figure 5.9 Item retrieval by trolley.

are delivered to storage locations by automated devices which are usually restricted
to a single aisle. In walk/ride and pick systems (W/RPSs), pickers travel on foot or by
motorized trolleys and may visit multiple aisles (see Figure 5.9).
The most popular order-to-picker systems are the AS/RSs. An AS/RS consists of
a series of storage aisles, each of which is served by a single storage and retrieval
(S/R) machine or crane. Each aisle is supported by a pick-up and delivery station
customarily located at the end of the aisle and accessed by both the S/R machine and
the external handling system. Therefore, assuming that the speeds vx and vy along
the axes x and y (see Figure 5.10) are constant, travel times t satisfy the Chebychev
x y
t = max , ,
vx vy
where x and y are the distances travelled along the x- and y-axes.
AS/RSs were introduced in the 1950s to eliminate the walking that accounted for
nearly 70% of manual retrieval time. They are often used along with high racks and
narrow aisles (see Figure 5.11). Hence, their advantages include savings in labour
costs, improved throughput and high ¬‚oor utilization.

Unit load retrieval systems. In some warehouses it is possible to move a single
load at a time (unit load retrieval system), because of the size of the loads, or of
the technological restrictions of the machinery (as in AS/RSs). In AS/RSs, an S/R
machine usually operates in two modes:
• single cycle: storage and retrieval operations are performed one at a time;

• dual cycle: pairs of storage and retrieval operations are made in sequence in an
attempt to reduce the overall travel time.


Side aisle
Central aisle

Figure 5.10 An S/R machine.
Side aisle

Figure 5.11 Item storage and retrieval by an AS/RS and a belt conveyor.

There exist systems in which it is possible to store or pick up several loads at the
same time (multi-command cycle).

Strict order picking versus batch picking. In multiple load retrieval systems,
customer orders can be assigned to pickers in two ways:

• each order is retrieved individually (strict order picking);

• orders are combined into batches.

In the latter case, each batch may be retrieved by a single picker (batch picking).
Otherwise, the warehouse is divided into a given number of zones and each picker is
in charge of retrieving items from a speci¬c zone (zone picking).


5.1.4 Decisions support methodologies
Warehouses are highly dynamic environments where resources have to be allocated in
real-time to satisfy customer orders. Because orders are not fully known in advance,
design and operational decisions are affected by uncertainty. To overcome the inher-
ent dif¬culty of dealing with complex queueing models or stochastic programs, the
following approach is often used.

Step 1. A limited number of alternative solutions are selected on the basis of expe-
rience or by means of simple relations linking the decision variables and simple
statistics of customer orders (the average number of orders per day, the average
number of items per order, etc.).

Step 2. Each alternative solution generated in Step 1 is evaluated through a detailed
simulation model and the best solution (e.g. with respect to throughput) is selected.

5.2 Warehouse Design
Designing a warehouse amounts to choosing its building shell, as well as its layout
and equipment. In particular, the main design decisions are

• determining the length, width and height of the building shell;

• locating and sizing the receiving, shipping and storage zones (e.g. evaluating
the number of I/O ports, determining the number, the length and the width of
the aisles of the storage zone and the orientation of stacks/racks/drawers);

• selecting the storage medium;

• selecting the storage/retrieval transport mechanism.

The objective pursued is the minimization of the expected annual operating cost
for a given throughput, usually subject to an upper bound on capital investment.
In principle, the decision maker may choose from a large number of alternatives.
However, in practice, several solutions can be discarded on the basis of a qualitative
analysis of the physical characteristics of the products, the number of items in stock
and the rate of storage and retrieval requests. In addition, some design decisions are
intertwined. For instance, when choosing an AS/RS as a storage/retrieval transport
mechanism, rack height can be as high as 12 m, but when traditional forklifts are used
racks must be much lower. As a result, each design problem must be analysed as a
unique situation.
Following the general framework introduced at the end of Section 5.1, we will
make in the remainder of this section a number of qualitative remarks and we will
illustrate two simple analytical procedures.


5.2.1 Selecting the storage medium and the storage/retrieval
transport mechanism
The choice of storage and retrieval systems is in¬‚uenced by the physical characteristics
of the goods, their packaging at the arrival and the composition of the outgoing
lots. For example, in a single storage zone warehouse, palletized goods are usually
stocked on racks if their demand is high enough, otherwise, stacks are used.Automated
systems are feasible if the goods can be automatically identi¬ed through bar codes
or other techniques. They have low space and labour costs, but require a large capital
investment. Hence, they are economically convenient provided that the volume of
goods is large enough.

5.2.2 Sizing the receiving and shipment subsystems
The receiving zone is usually wider than the shipping area. This is because the incom-
ing vehicles are not under the control of the warehouse manager, while the formation of
the outgoing shipments can be planned in order to avoid congesting the output stations.

Determining the number of truck docks
Goods are usually received and shipped by rail or by truck. In the latter case, the
number of docks nD can be estimated through the following formula,
nD = ,
where d is the daily demand from all orders, t is the average time required to
load/unload a truck, q is the truck capacity, and T is the daily time available to
load/unload trucks.

Sintang is a third-party Malaysian ¬rm specialized in manufacturing electronic
devices. A new warehouse has been recently opened in Kuching. It is used for storing
digital satellite receivers, whose average daily demand is d = 27 000 units. Outgoing
shipments are performed by trucks, with a capacity equal to 850 boxes. Since the
average time to load a truck is t = 280 min and 15 working hours are available every
day, the warehouse has been designed with the following number nD of docks:
27 000 — 280
nD = = 10.
850 — 900

5.2.3 Sizing the storage subsystems
The area of the storage zone must be large enough to accommodate goods in peak
periods. On the other hand, if the storage zone exceeds the real needs of the ¬rm,


storage and retrieval times become uselessly high. This could decrease throughput or
increase material handling costs.

Determining the capacity of a storage area
The size of a storage area depends on the storage policy. In a dedicated storage
policy, each product is assigned a pre-established set of positions. This approach is
easy to implement but causes an underutilization of the storing space. In fact, the
space required is equal to the sum of the maximum inventory of each product in time.
Let n be the number of products and let Ij (t), j = 1, . . . , n, be the inventory level of
item j at time t. The number of required storage locations md in a dedicated storage
policy is
md = max Ij (t). (5.1)
j =1
In a random storage policy, item allocation is decided dynamically on the basis of
the current warehouse occupation and on future arrival and request forecast. Therefore,
the positions assigned to a product are variable in time. In this case the number of
storage locations mr is
mr = max Ij (t) md . (5.2)
j =1
The random storage policy allows a higher utilization of the storage space, but
requires that each item be automatically identi¬ed through a bar code (or a similar
technique) and a database of the current position of all items kept at stock is updated
at every storage and every retrieval.
In a class-based storage policy, the goods are divided into a number of categories
according to their demand, and each category is associated with a set of zones where
the goods are stored according to a random storage policy. The class-based storage
policy reduces to the dedicated storage policy if the number of categories is equal to
the number of items, and to the random storage policy if there is a single category.

Potan Up bottles two types of mineral water. In the warehouse located in Hangzhou
(China), inventories are managed according to a reorder level policy (see Chapter 4).
The sizes of the lots and of the safety stocks are reported in Table 5.1. Inventory levels
as a function of time are illustrated in Figures 5.12 and 5.13. The company is currently
using a dedicated storage policy. Therefore, the number of storage locations is given
by Equation (5.1):
md = 600 + 360 = 960.
The ¬rm is now considering the opportunity of using a random storage policy. The
number of storage locations required by this policy would be (see Equation (5.2))
mr = 600 + 210 = 810.


Table 5.1 Lots and safety stocks (both in pallets) in the Potan Up problem.

Product Lot Safety stock

Natural water 500 100
Sparkling water 300 60






Figure 5.12 Inventory level of natural mineral water in the Potan Up problem.





. 7
( 15)