Anda di halaman 1dari 10

MANAGING PERISHABLE is the newsvendor model discussed in the

INVENTORY section titled ‘‘Extending the Newsvendor


Model’’.) Thus, when the shelf-life of the items
OPHER BARON stored is finite, traditional control policies
Rotman School of Management, may not be sufficient. For this reason, special
University of Toronto, Toronto, purpose models that focus on the manage-
Ontario, Canada ment of inventory of perishable items were
developed. Our purpose is to highlight the
reasons for these differences and enable fur-
ther investigation of the models developed for
perishable items by providing relevant refer-
INTRODUCTION ences. A more technical discussion of the tech-
niques employed in the analysis of perishable
This article discusses inventory management items is given in the article titled Mathe-
of perishable items, including the differences matical Models for Perishable Inventory
between inventory models that consider Control in this encyclopedia.
items’ perishability and the standard models This article continues as follows: the
that ignore this issue. We hope that this section titled ‘‘Modeling Inventory Prob-
article will help fostering the interest of lems for Perishable Items’’ highlights the
students and researchers in inventory con- differences between inventory models for
trol of perishable items, and ultimately help perishables and models that ignore per-
improving the inventory control of perishable ishability. The section titled ‘‘Taxonomy of
items in practice. Inventory Models for Perishable Items with
To emphasize the importance of appro- Infinite Horizon’’ provides a taxonomy of
priately managing inventory of perishables, models for controlling inventory of perish-
note that Ref. 1 estimates the total cost of ables based on Ref. 2, and lists some of the
unsalable merchandise of suppliers to drug- basic results of such models. An important
stores and supermarkets in 2005 by $2.05 observation from this section is that effective
billion. This cost goes directly to the bottom heuristics are required for the management
line of many of these companies. Moreover, of perishable items. Thus, the section titled
higher clock speed in many industries results ‘‘Developing a Heuristic Control Policy’’
in shorter and shorter life-cycles of many demonstrates the development of such a
products. The inventory control of such prod- heuristic based on Ref. 3, and finally, the
ucts might become similar to the one of per- section titled ‘‘Future Research Directions’’
ishable items as well after some period, when lists probable directions for future research
a newer version is presented, the value of old on the management of perishable items.
items declines significantly. One important
difference between items with short life-cycle MODELING INVENTORY PROBLEMS FOR
and perishables is that in some cases, the firm PERISHABLE ITEMS
controls the introduction of the new technol-
ogy. Still, much of the insight derived for The analysis of inventory systems is primar-
managing perishable inventory may be valid ily focused on the tactical question of which
for products with short life cycles. inventory control policies to use and the oper-
Most of the traditional inventory models ational questions of when and how much
analyzed in the literature (e.g., EOQ, (Q, r), inventory to order. By and large, these are
and (S, s) policies) assume an infinite shelf- the main questions for managing the inven-
life for items (see Multiechelon Multiprod- tory of perishable items as well. This section
uct Inventory Management). (An exception reminds the reader of the assumptions used

Wiley Encyclopedia of Operations Research and Management Science, edited by James J. Cochran
Copyright © 2010 John Wiley & Sons, Inc.
1
2 MANAGING PERISHABLE INVENTORY

in modeling standard inventory systems for the inventory of good items is deteriorating
nonperished items and then lists the special with time at some rate, as in Ref. 4.
features of models for perishables. It con- The two most common models for perisha-
cludes by generalizing the newsvendor model bility are outdatedness due to reaching expiry
to consider items with finite shelf-life. In the date (e.g., food items or medicine) and sudden
process, we present relevant notation. perishability due to disaster (e.g., spoilage
because of extreme weather conditions). The
Inventory Modeling for Nonperishables perishability due to outdatedness is typically
modeled as a deterministic time to perisha-
In inventory models for nonperishables, time
bility and the perishability due to disaster is
is measured either in a discrete (i.e., periodic)
typically modeled as an exponential (or its
or continuous fashion. The choice of time
discrete counterpart, geometric) time to per-
measurement is often related to the review
ishability. This is because the memory-less
period (periodic or continuous) and to the
property of these distributions often results
model horizon, T, that may be a single period,
in more tractable models.
finite, or infinite horizon. The replenishment
The treatment of lead time in the manage-
lead time, L, may be assumed to be zero,
ment of inventory for perishable items is also
deterministic, or stochastic (with or without
not trivial. One difficulty is that the items
order crossings). The customers arrival rate
might perish during the delivery time. This
per time period, λ, may be deterministic or
might be addressed by assuming that the sup-
stochastic, each individual’s demand, D, may
plier supplies fresh items upon their delivery
be deterministic or stochastic, and D may be
and changes their lead time accordingly. For
discrete or continuous. A typical assumption
example, when the lead time is determin-
is that the arrival process is Poisson with
istic and items’ shelf-life is exponential, we
rate λ; D = 1 (see Poisson Process and
can assume that delivered items are fresh
its Generalizations). Finally, when there
by modeling the lead time as a geometric
are no items on the shelf, that is there are
random variable.
shortages, demand is typically assumed to
In the example above, changing the lead
be backlogged or lost.
time to support the assumption that sup-
The costs considered in the management
plied items are fresh complicates the model
of standard inventory systems are per unit
because it introduces another uncertainty in
procurement cost, c; per unit selling price,
the model; namely, the corrected lead time
p; order setup cost, K ≥ 0; and holding cost
becomes uncertain. To avoid complicating the
per unit per time period, h > 0. Another cost
model, much of the literature on perishable
is that of shortages. The shortage costs can
items assumes that supplied items (after the
include a one-time cost per shortage, Kb ≥ 0
lead time) are fresh.
if shortages are backlogged, or Kl ≥ 0 when
Another complication is that when lead
shortages result in lost sales, and a cost per
times are long relative to the items’ shelf-
unit per unit time, cb ≥ 0, in cases of backlog,
lives, there might be several orders out at
or a cost per unit, cl ≥ 0, in the case of lost
the same time. Then, orders might cross
sales.
each other. This potentially affects the shelf-
life distribution of delivered items. (Order
Special Features of Models of Perishables
crossing typically increases the complexity of
In addition to the above list of characteristics, traditional models as well.)
modeling of inventory for perishable items A general difficulty caused by the pres-
also requires a characterization of the time ence of lead time in a stochastic environment
to perishability, Lp . This time to perishabil- is that shelf-life of items in stock depends on
ity may be either deterministic or stochastic. which batch they arrived at. Thus, the con-
Moreover, when orders arrive in batches, all troller should keep track not only of the quan-
items in a batch may share the same time to tity of the items on shelf but also the length
perishability, or each item may have its own of time different items are on the shelf; that
shelf-life. An alternative assumption is that is, their age and its distribution.
MANAGING PERISHABLE INVENTORY 3

Another challenge in the analysis of analysis of the optimal control policy for
perishable items is that items that have not perishables, it is enough that a newly arriv-
yet perished may have different remaining ing batch may have a shorter shelf-life than
shelf-life. (Either because items have a that of existing items. Moreover, in cases
unique shelf-life, or because batches with when customers may control the items’ dis-
common shelf-lives arrived in different patching they may prefer fresher items (con-
periods.) Thus, the information required to sider items such as milk), causing the actual
completely characterize the on-hand and dispatching to differ from first-in-first-out.
on-order inventory includes not only the
quantity of inventory but also the remaining
shelf-lives of each unit in the inventory. This Extending the Newsvendor Model
increase in information requirement makes
the analysis of multiechelon supply chains The newsvendor model discussed in the
for perishable products especially challeng- article titled Newsvendor Models in this
ing (because even for standard items, such encyclopedia is an exception in the standard
an analysis often relies on dynamic program- inventory literature, because it considers
ming and suffers from the curse of dimension- perishability. This model considers the order
ality). Therefore, some important theoretical quantity required to maximize the profit
contributions developed in the analysis of over a single period when the demand over
perishable items address the information the period is uncertain. In addition to the
required to characterize the inventory level costs listed above, the newsvendor model
process. (See the discussion in the section also considers the salvage cost of perishable
titled ‘‘Models Without Order Setup Cost items cs , which can be negative or positive.
or Lead Times’’ and in the article titled The newsvendor model can also be used to
Mathematical Models for Perishable model perishable items that can be ordered
Inventory Control in this encyclopedia). only once. Now, we model the length of the
Another relevant decision in managing single period (i.e., the lifetime) as uncertain,
perishables when items of several different but there are no further procurement oppor-
ages coexist, is items dispatching to fulfill tunities. In this case, the uncertainty in
orders. In such cases, it might not be optimal the demand depends on both, the uncertain
to dispatch products in a first-in-first-out demand per time period and the uncertain
fashion. For an extreme case, consider the length of the selling season. A similar idea
following example. was first investigated in Ref. 5, which in
contrast to the newsvendor model, allows
Example 1. A last-in-first-out dispatching several ordering opportunities before the
policy would decrease the effect of perisha- perishability time.
bility when an item’s shelf-life follows a new To express the uncertainty in demand in
worse-than-used distribution. (Let FLp (s) these settings, we assume that the shelf-
denote the cumulative distribution function life Lp is a discrete random variable
 with
 a

of the shelf-life. Then, if z-transform Z Lp (z) = ∞ i=0 z Pr Lp = i and
i

   that the demand in time period i is inde-


1 − F Lp (s + t ) ≥ 1 − F Lp (s) 1 − F Lp (t ) pendent and identically distributed, Di , with
∀s ≥ 0, ∀t ≥ 0, a probability density function (pdf), f Di (x),
and a moment-generating function, L∗D (α) =
 ∞ αx i
we say that Lp has a new worse-than-used −∞ e f Di (x) dx. Then, the horizon consid-
distribution.) ered by the planner is T = Lp , the overall
demand during the period, DT , is a random
While perishable items with a new worse- sum of random variables with a moment-
than-used shelf-life are hard to find in prac- generating function L∗D :
T
tice, this example highlights the difficulty of
characterizing the optimal dispatching pol-    
icy. To substantially complicate the exact L∗D (α) = E L∗D (α) Lp = Z Lp L∗D (α) .
T i i
4 MANAGING PERISHABLE INVENTORY

 
 Lp ∼ Poisson p and Di
Example  2. If Models for Perishable Inventory Control
∼ Normal μ, σ 2 and are 
i.i.d,  Then in this encyclopedia.
μα+(σ α)2 / 2
Z Lp (z) = ep(z−1) , L∗D (α) = e and Discrete Review Systems
i
the moment-generating function of overall
demand over the period is Models with No Fixed Order Setup Cost and
No Lead Time. The simplest model is one
   
p exp μα+(σ α)2 / 2 −1 in which items’ shelf-life equals the review
L∗D (α) = e . period. Then, different periods are indepen-
T
dent and the standard newsvendor solution is
optimal in each period. The second simplest
Because the moment-generating function model, in which items’ shelf-life equals two
fully characterizes the demand distribution review periods, is already much more inter-
over the products’ random shelf-life, the stan- esting. This model was pioneered in Ref. 6,
dard solution of the newsvendor model can where it was shown that the optimal control
be implemented to maximize the profit even policy is state dependent.
when the valuable shelf-life of the items is That the optimal control policy is state
uncertain. dependent, is somewhat surprising, because
the optimality of the base stock control policy
is easily established in discrete review mod-
TAXONOMY OF INVENTORY MODELS FOR els for nonperishables without order setup
PERISHABLE ITEMS WITH INFINITE cost. Moreover, this result is also important
HORIZON because it implies that for models in this cate-
gory, the simple base stock-level control is not
As shown above, the newsvendor model optimal, in sharp contrast to the optimality
is useful in the inventory management of of this control for nonperishables.
perishable items that can only be ordered Thus, for perishable items, the opti-
once. However, many perishable items such mal inventory control requires solving a
as food and medicines are consumed over dynamic programming problem. The dif-
a much longer horizon; then a single order ficulty in finding and implementing the
is not practical. Finite horizon models were optimal state-dependent controls motivated
addressed in the literature almost solely in researchers to search for effective heuristics.
the discrete review settings. In contrast, For example, Broadheim et al. [7] suggested
infinite horizon models were considered in a heuristic that is based only on the age
both discrete and continuous review settings. of the newest item in the system and the
Below, we describe the primary models, simulation-based study in Ref. 8 considered
assumptions, and results in the inventory several heuristics including one that keeps
literature on perishable items. We follow the the total number of items in the system
classification of Ref. 2 in the section titled fixed, ignoring their ages.
‘‘Taxonomy of Inventory Models for Perish-
able Items with Infinite Horizon’’ because it Models with Fixed Ordering Cost but no
is both thorough and recent. The interested Lead Time. The research on this category
reader is encouraged to read Ref. 2. was pioneered in Ref. 9, which highlights
The infinite horizon models for perishables the complexity of the optimal control struc-
include a cost, cp , per perished item. This cost ture for these models and suggests heuristic
replaces the salvage value of perished items for control of the system based on the (S, s)
in the newsvendor model. It is also possible model.
to include a fixed cost, Kp , in cases where A nice observation for this model of the
batches of items perish; however, most of the Poisson demand with backlogs case is that
models do not include the latter cost. the optimal reorder level is always at or below
A more detailed discussion of the state- −1. This result, which was established in
of-the-art results for each of these models Ref. 10, echoes the one in Ref. 11 for the
is given in the article titled Mathematical continuous review system. The proof follows
MANAGING PERISHABLE INVENTORY 5

because—with no lead time—higher reorder Continuous Review Systems


points would increase holding cost without
Models without Order Setup Cost or Lead
reducing the backlog. (This result also holds
Times. This category was motivated to model
in the case where customer’s arrival process
blood banks. It was originated by Graves [14],
is Poisson and each customer requests some
integer quantity of the product.) who assumed that the items are continuously
produced, perish after a deterministic time,
Models with Lead Time. The inclusion of and that demand follows a compound Pois-
positive lead time leads to much more com- son process with either a single unit or an
plicated models, as the number of states exponential demand at each arrival. A large
required for the dynamic program increases. body of work within this category was done by
With lead time, one needs to also keep track Perry et al. in Ref. 15 and references therein;
of the age of ordered items. This model is con- thus Karasesman et al. [2] named it the Perry
sidered in Ref. 12, where methods to solve it model.
are suggested. Recently, Berk [13] analyzed An interesting observation that is used in
the (Q,r) periodic review inventory model for the analysis of the Perry model is that its per-
perishables with lead time by using the effec- formance can often be characterized based on
tive shelf-life distribution and assuming lost knowledge of the virtual death process, that
sales. is, the time until the next perishability [16].
The main assumption made in many of the
Discussion. The assumption of zero lead
models within this category is that good items
time in the case of a discrete review system
arrive one by one independently of the deci-
is often justified if the lead time is shorter
sion making. Thus, in contrast to standard
than the review period. Therefore, the analy-
inventory models, the controller does not
sis of such models (with periodic review and
decide on when and how much to order. Still,
no lead time) is valuable. The zero lead time
the controller might affect the arrival rate of
assumption is not as meaningful for contin-
good items by advertising, for example.
uous review models or when the lead time is
Because of this lack of control on the input,
long and makes reviewing the items at such
intervals too expensive. most work on the Perry model focused on
From an analysis point of view, treating performance analysis rather than on finding
perishable items requires consideration optimal controls.
of the age distribution of the on-hand
inventory in both ordering and dispatching
Models without Order Setup Cost but
of perishable items. Dispatching is also
important because, typically (and in contrast with Positive Lead Time. The first work on
to Example 1), the value of older items continuous review models without setup
in stock is lower than that of the newer cost but with lead time is given in Ref. 17
ones. The challenges in both ordering and where, as in many subsequent works, the
dispatching of perishables causes the tra- performance of an (S − 1, S) control policy
ditional optimal policies for nonperishable is investigated. However the optimality of
items to be suboptimal for perishables. this control policy is not established, even
It turns out that the optimal control when time to perishability and interarrival
policies for perishable items are fairly hard times are exponential. In fact, in view of
to characterize. This difficulty encouraged the fact that base stock is not the optimal
researchers to develop effective heuristics policy in the corresponding discrete review
for the periodic control of perishable items in model, even in the absence of lead time, this
the absence of lead time. However, no such control policy is probably not optimal for the
heuristics were developed for models with continuous review models either. While some
long lead time (Williams and Patuwo [12] study of problems with fixed lead time and
give an exact analysis of several relevant time to perishability has been pursued [18],
cases and suggest guidelines for developing the theoretical results about these applicable
such heuristics). models are far from being complete.
6 MANAGING PERISHABLE INVENTORY

Models with Order Setup Cost. Models with DEVELOPING A HEURISTIC CONTROL
order setup cost can be portioned into ones POLICY
with a fixed or variable order size, corre-
sponding to the (Q,R) or the (s,S) standard As discussed above, there is a need for
inventory models. Both types of models can developing effective control policies for both,
be analyzed as cost minimization problems, continuous and discrete review models of
possibly subject to some service level con- perishable items. Below, we discuss such
straint. a heuristic based on Ref. 3. Consider a
For fixed batch size problems, two policies continuous review model with no lead time,
were considered: the standard (Q, R) and the where the customers arrival process is
(Q, R, T) policy, where orders of a batch are Poisson with rate λ and each customer
triggered by either the time passed (since the requests a continuous random quantity D of
inventory level was Q) being T, or by the items with a mean demand size of E (D). The
inventory level falling below R [19]. costs considered are the order setup cost,
The vast majority of work allowing for K, holding cost per item per time period,
variable order size assumes no lead time, h, and cost of perishable items, cp . Let us
as in Ref. 11. In Ref. 11, it is established focus on an (S, s) control policy and allow no
that with Poisson demand the (S,s) is the backlog, which is legitimate due to the zero
optimal policy when shortages are either lead time assumption. Thus, s = 0 is optimal
backlogged or result in lost sales, and also and due to the Poisson arrival process, the
that s < −1 or s = 0 for these models, respec- times where the inventory level is raised
tively. (The insight behind the proof on the to S are renewal epochs (see Definition
optimal reorder point is similar to the one and Examples of Renewal Processes,
for the corresponding discrete review model properties of renewal processes)
discussed in the section titled ‘‘Models with In the standard (S, 0) model without per-
Fixed Ordering Cost but no Lead Time.’’) ishable items, the end of the inventory cycle,
Recently, Gurlur and Ozkaya [20] consid- denoted by TS is caused only due to the
ered a zero lead time with backlog (s, S) policy arrival of a demand. Of course, TS is a ran-
for perishables. They used sums and inte- dom variable that depends on S and on the
grations of relevant distribution functions to demand arrival process. But when items are
express the expected cost rate function for perishable, cycles can also end due to per-
their model and developed a heuristic for the ishability, at time Lp , which is the random
positive lead time case. variable describing the time to perishability
of a new batch. To emphasize that the model
Discussion. To summarize, stochastic focuses on perishable items, we denote the
analysis is the main tool used in the inves- time to end the cycle by τ = min TS , Lp and
tigation of continuous review models for the corresponding control policy by (S, τ ), as
perishable items. This analysis is not trivial in Ref. 3.
and the characterization of the optimal Figure 1 shows a sample path of inventory
control policies is more complicated than in over more than two inventory cycles. The
the periodic review models. Therefore, there first cycle ends due to the demand, as in a
is a need to find effective control policies standard (S, 0) model; then τ = T S1 ≤ L p1 .
for these models as well, especially in the The second cycle ends due to perishability;
presence of lead time. then τ = Lp2 < TS2 .
There is still a place for developing effi- Let V (t) denote the inventory level at
cient heuristics for the control of continuous time t ∈ [0,τ ) from the beginning of the cycle.
review policies for perishables with lead time. Thus, V = {V(t) : t ≥ 0} is a regenerative pro-
Both the heuristic reported in Ref. 20 and cess such that V (0) = S. We let V(τ ) denote
the one for the no lead time case from Ref. 3, the inventory level at the end of the cycle and
which is also reproduced in the next section, use E as the expected value operator (both
seem good starting points for finding such E(V) and E(V(τ )) are well defined because V
heuristics. is a regenerative process). Then, the long-run
MANAGING PERISHABLE INVENTORY 7

Sample path (S,τ ) models.


Order times

Interarrival times
A ~exp(λ)
1

S
Demand X1

A2
V(t)
X2
A3

X3

t
Time to perishability Lp1
Lp2 Lp3

Figure 1. Sample path (S, τ ) models.

average cost C(S) is Since for every t < τ we have V(t) = S −


  t λ E(D) and, as in the EOQ model, the aver-
K + cp E V(τ ) I{τ < Ts } age inventory resulting from an inventory
C(S) =
E(τ ) triangle of height Y is Y/2, we have
+ h E (V), (1)

S/(λE(D))
tλE (D)
E (V) = S − tλE (D) +
where I{·} is the indicator function with a 0 2
value 1, if the event {·} occurs and 0 otherwise.
S S
To find a heuristic control policy, we fLp (t) dt + Pr Lp > (3)
2 λE (D)
replace the stochastic demand assumption
with a deterministic demand with a rate
and
λE (D). Then, starting at the reorder level S
for any time t before the end of the cycle, we

  S/ (λE(D))
have E V(τ ) I{τ < TS } = (S − t λ E (D))
0

V (t) = S − t λ E (D) t ∈ [0, τ ). f Lp (t) dt. (4)

Therefore, if items do not perish within a


cycle its length is TS = S/(λE (D)). Letting Remark. This heuristic assumes a contin-
f Lp (t) denote the pdf of Lp , the expected cycle uous and deterministic demand process, thus
length is given by  new demand arrives immediately after an old

E (τ ) = E min Lp , TS batch perishes. In practice, when there is no

S/(λE(D)) lead time, ordering the next batch after items
= t f Lp (t) dt perish can be postponed until the next cus-
0 tomer’s arrival. That is, the order at the end

S S of the second cycle in Fig. 1 can be postponed
+ Pr Lp > . (2) until the first arrival in the third cycle. Such
λ E (D) λ E (D)
8 MANAGING PERISHABLE INVENTORY

postponement would increase E (τ ) and can (around 2% on average). Thus, Baron et al.
be incorporated into the heuristic by adding [3] suggested that this heuristic be used for
the expected time to an arrival to the expected cases where S∗ cannot be found.
cycle length whenever cycles end due to per-
ishability. In contrast, when the demand size
FUTURE RESEARCH DIRECTIONS
is continuous, it is likely that when the cycle
ends due to demand there is some demand
Below, we discuss several research direc-
that should be satisfied from the next batch. tions and explain their importance. Many
Then when no backlog is allowed, a similar of these extensions have attracted significant
postponement is not feasible for cycles that research for standard items, but in our opin-
end due to demand. In any case, postponing ion, they are not addressed well enough for
orders when cycles end due to perishability perishable items.
might not be beneficial when lead time is pos- The most important research direction
itive. Thus, in the example given below, we for managing inventory of perishables is to
ignore such postponement and use E (τ ) as develop effective heuristics for inventory con-
given in Equation (2). trol of such items. Some heuristic control
policies were developed earlier, such as by
Nahmias [21], who discusses heuristics for
Example 3. The heuristic when Lp ∼ the deterministic shelf-life case. However,
exp (ξ ): Assuming Lp ∼ exp (ξ ) using Equat- developing effective heuristics for additional
ions (2)–(4), we get the approximated control settings is still required. Such settings should
problem: include both continuous and periodic review
systems and focus on cases with positive lead
⎛ 
S
⎞
−ξ times. A closely related research direction
λE(D) 1−e λE(D)
⎜ ⎟ is to provide theoretical guarantees on the
K + π ⎝S − ξ ⎠
performance of different heuristics, similar
min Ch (S) = to the established efficiency of power of two
S −ξ S
1−e λE(D) policies for standard inventory items [22].
ξ The standard assumption in the literature
⎛ ⎞
S
−ξ λE(D) on continuously reviewed perishable items is
λE (D) 1 − e
⎜ ⎟ that the planning horizon is infinite. Investi-
+ h⎜
⎝S −
⎟,
⎠ gating the inventory control of perishables for

finite horizons in such settings also deserves
more research attention.
(5) Four other issues that have garnered a
lot of attention for standard inventory items
while in general there is no closed form are their management (i) in multiechelon
solution to problem (5), it is easily solved settings, (ii) when aiming to coordinate
numerically. the supply chain, (iii) for multiple items,
and (iv) in the presence of competition (see
This heuristic, of solving Equation (5) was Business Process Outsourcing multiech-
investigated in Ref. 3, for cases in which the elon, multiple items and Supply Chain
time to perishability is exponential or deter- Coordination). An important feature
ministic, arrivals follow a Poisson process, in investigating these issues is internal
and customers’ demand is either a unit or transshipment. Many models address-
exponentially distributed. For these cases, ing nonperishable items ignore internal
Baron et al. [3] expressed the optimal order shipments, because the cost of internal
up to level, that is, the one that minimizes transshipments might offset their benefit.
Equation (1) denoted by S∗ , and compared it However, because transshipments of per-
to the order up to the level based on solv- ishable items may reduce the proportion of
ing Equation (5), SH . For these cases, the perished items, the value of internal trans-
cost error of the heuristic is relatively small shipments for perishable items is higher
MANAGING PERISHABLE INVENTORY 9

than for nonperishables. Thus, allowing subject to service level constraints even for a
internal transshipments is preferable in single customer type has won little attention.
models that consider multiple sites. Such constraints could be quite specific in
Initial steps at investigating the above defining appropriate service levels in relation
issues for supply chains of perishable items to the residual life of items sold. Customers
were taken (Section 4 in Ref. 2). For example, might not be satisfied with only a high
for the supply of blood in multiechelon on-shelf availability, but may also demand
settings (Ref. 23 and references within), that available items are fresh enough.
consider both, a rotation (or recycling) policy Moreover, in some cases different customer
and a retention one. In a rotation system at types prefer different levels of freshness
the end of each period (e.g., day) the supplier that require different storage processes. For
gathers all unsold (and not yet perished) example, the shelf-life of tomatoes is longer
items from the retailers and then supply if they are kept on the vine, refrigerated
them at the beginning of the next period; in (that may also reduce their flavor), or turned
a retention system, supplied items are left into a tomato paste, tomato juice, or ketchup.
with the retailers until they are demanded or Thus, it is likely that formulating problems
perished. However, the literature addressing with different customer types would be very
issues (i)–(iv) for perishables typically application dependent. Still, in view of the
considered fairly restrictive settings. importance of these applications, for example
Therefore, there is room for addressing these in the health-care industry, this is a topic
issues and developing effective management deserving further research.
policies for managing the flow and allocation The above discussion of different service
of perishable items in supply chains. levels and storage requirements for perish-
Note that to properly address these able items raises another question; namely,
issues, investigation of the strategic plan- the effective analysis of the logistics required
ning of supply-chain networks is required for the delivery of perishable items. Logistics
for perishable items. Appropriate models planning for perishable items involves capac-
for supply-chain networks require as input, ity, storage, production, and transportation
the inventory control policy used and decisions as well as their effect on items’
their performances (see Supply Chain shelf-life. For example, some fruits are deliv-
Coordination). For example, locating ered to North America in ships, with part
warehouses and distribution centers should of their ripening occurring during this long
consider the costs implied by different and often uncertain transportation window.
supply-chain configurations. However, in the Once ripened, such fruits are often stored
absence of efficient methods to manage the at cold temperatures to increase their shelf-
inventory within the supply chain and to life. Thus, managing the logistics process
estimate the implied inventory-related costs, required to bring valuable items to final cus-
designing the network of the supply chain tomers is not simple, and deserves further
might be of limited value. attention. Again, such research is likely to be
Another model that is well established application dependent.
for standard items is joint manufacturing, An important issue that is unique for per-
storing, and allocation of products for differ- ishable items is the dispatching policy which
ent customer types. This is also relevant for may be complex even for a single customer
perishable items. An example is the choice type. In addition to the complexity high-
made at a central depot that serves several lighted by Example 1, the dispatching policy
different retailers with different importance. can also affect the demand. For example, con-
Then, an allocation of items may need to sumers of food items such as milk, often check
consider performance measures as seen by the ‘‘best before’’ date of items on the shelf.
the different customer types. Such measures Thus, demand for a specific item depends not
of service levels could be on-stock availability only on its availability on the shelf and its
of ‘‘fresh enough’’ items, fill rates, and so on. age, but also on the age and shelf availabil-
In fact, inventory control of perishable items ity of items of different ages. That is, there
10 MANAGING PERISHABLE INVENTORY

are substitutability effects among items that periods subject to stochastic demand. Nav Res
differ by age. In addition, in this setting, the Logist Q 1973;20:207–229.
controller can only affect rather than dictate 7. Broadheim E, Derman C, Prastacos GP. On
the dispatching policy. This is a mirror prob- the evaluation of a class of inventory policies
lem to the models in the section titled ‘‘Models for perishable products such as blood. Manag
Without Order Setup Cost or Lead Times,’’ Sci 1975;22:1320–1325.
where the controller can only affect rather 8. Nahmias S. A comparison of alternative
than dictate the arrival of new items. Again, approximations for ordering perishable inven-
tory. INFOR 1975;13:175–184.
initial works on this subject exist (Section 5.2
of Ref. 2), but there is still plenty of opportu- 9. Nahmias S. The fixed charged perisha-
ble inventory problem. Oper Res 1978;26:
nity for research in more realistic settings.
441–481.
As the last but not the least research
10. Lian Z, Liu L. A discrete-time model for
direction, we list the following simplifying
perishable inventory systems. Ann Oper Res
assumptions that are often made in works
1999;87:103–116.
on perishable items. Most of the references
11. Weiss H. Optimal ordering policies for con-
above assume that (i) demand is known (in
tinuous review perishable inventory models.
case of stochastic demand its distribution Oper Res 1980;28:365–374.
is known), (ii) there is no seasonality, (iii)
12. Williams CL, Patuwo BE. A perishable inven-
demand is independent of pricing, shelf tory model with positive order lead times.
space allocation, and inventory level, (iv) Euro J Oper Res 1999;116:352–373.
order yield is perfect, and (v) there is 13. Berk E, Gurler U. Analysis of the (Q,
no substitutability of items (both within r) inventory model for perishables with
items of different ages and among different positive lead times and lost sales. Oper Res
products). For nonperishable items, all of 2008;56(5):1238–1246.
these assumptions have been relaxed to some 14. Graves S. The application of queuing theory
extent. However, the vast majority of papers to continuous perishable inventory system.
on managing inventory of perishables make Manag Sci 1982;28:400–406.
these assumptions. (Several initial attempts 15. Nahmias S, Perry D, Stadje W. Actuarial valu-
at relaxing some of these assumptions are ation of perishable inventory systems. Probab
referenced in Section 5 of Ref. 2.) Eng Infor Sci 2004;18:219–232.
16. Kaspi H, Perry D. Inventory system of
perishable commodities. Adv Appl Probab
REFERENCES 1983;15:674–685.
1. Grocery manufacturers of America. 2006 17. Pal M. The (s-1,S) inventory model for dete-
Unsaleables benchmark report. 2006. riorating items with exponential lead time.
2. Karaesmen I, Scheller-Wolf A, Deniz B. Calcutta Stat Assoc Bull 1989;38:83–91.
Managing Perishable and Aging Invento- 18. Perry D, Posner M. An (S-1;S) inventory sys-
ries: Review and Future Research Directions. tem with fixed shelf-life and constant lead
Kempf K, Keskinocak A, Uzsoy P, editors. times. Oper Res 1998;46:65–71.
Handbook of Production Planning. Kluwer 19. Tekin E, Gurler U, Berk E. Age based vs stock-
Academic Publishers; 2009. To appear. level control policies for a perishable inventory
3. Baron O, Berman O, Perry D. Stochastic (τ , S) system. Eur J Oper Res 2001;134:309–329.
models for managing inventory of perishable 20. Gurlur U, Ozkaya BY. Analysis of (s, S) policy
items. Working paper. Rotman School of Man- for perishables with a random shelf-life. IIE
agement, University of Toronto; 2008. Trans 2008;410:759–781.
4. Rajan A, Steniberg R. Dynamic pricing and 21. Nahmias S. Perishable inventory theory: a
ordering decisions by a monopolist. Manag Sci review. Oper Res 1982;30(4):680–708.
1992;38:240–262. 22. Roundy R. A 98% effective integer ratio lot-
5. Brown GW, Lu JY, Wolfson RJ. Dynamic mod- sizing rule for a one warehouse multi-retailer,
eling of inventories subject to obsolescence. system. Manag Sci 1985;31:1416–1430.
Manag Sci 1964;11(1):51–63. 23. Prastacos GP. Allocation of a perishable prod-
6. Nahmias S, Pierskalla W. Optimal ordering uct inventory. Oper Res 1981;29:95–107
policies for a product that perishes in two