A GENERALIZED MODEL FOR EVALUATING SUPPLY CHAIN

DELIVERY PERFORMANCE



Maxim Bushuev
Department oI Management and InIormation Systems
Kent State University
Kent, Ohio 44242 USA
mbushuevkent.edu

AlIred L. GuiIIrida*
Department oI Management and InIormation Systems
Kent State University
Kent, Ohio 44242 USA
aguiIIrikent.edu
Voice: 716/954-3504

Murali Shanker
Department oI Management and InIormation Systems
Kent State University
Kent, Ohio 44242 USA
mshankerkent.edu



*Please direct all correspondence to AlIred L. GuiIIrida.


Submitted to:
The 47
th
Annual MBAA International ConIerence, Chicago, March 23-25, 2011

Division: Marketing Management Association
Track: Supply Chain Management/Logistics












2
A GENERALIZED MODEL FOR EVALUATING SUPPLY CHAIN
DELIVERY PERFORMANCE

Abstract
Supply chain management has evolved as a core component in an organization`s overall
strategy Ior gaining and maintaining competitive advantage. The supply chain
management philosophy serves as a Ioundation Ior integrating and eIIective managing
value-adding activities such as raw material acquisition, production processing and
physical distribution. In support oI this philosophy, quantitative models Ior evaluating
delivery perIormance to the end customer in the supply chain are used to assist managers
in assuring that high levels oI customer service and satisIaction are maintained.

In this paper we present a generalized delivery perIormance model that overcomes two
existing limitations Iound in delivery perIormance models that have been reported in the
literature. The model presented herein uses the gamma probability density Iunction (pdI)
in the construction oI a cost based delivery perIormance model. The model is illustrated
using a numerical example.

Key Words: Deliverv performance, Supplv chain management, Cost based modeling

Introduction

Supply chain management serves as the Ioundation oI an organization`s overall
competitive strategy Ior attaining and maintaining competitive advantage. Under the
supply chain management philosophy, value-adding activities such as raw materials
acquisition, production processing and physical distribution are all coordinated to insure
that customer demand is met with a correct order quantity that is delivered in a timely
manner.
The need Ior perIormance measurement and evaluation in supply chain management is
well recognized in the literature. Detailed discussions on the importance oI perIormance
measurement in supply chains may be Iound in Gunasekaran and Kobu (2007) and
Lockamy and McCormack (2004). A taxonomy oI useIul supply chain perIormance
metrics that spans the strategic, tactical and operational levels oI supply chain operation
has been eIIectively summarized in Bhagwat and Sharma (2007) and Gunasekaran et al
3
(2004). Within the hierarchy oI supply chain perIormance metrics, delivery perIormance
is acknowledged as a key metric Ior supporting operational excellence oI supply chains
(Gardner, 2010) and is classiIied as a strategic level perIormance measure by
Gunasekaran et al (2004).
In today`s competitive business environment customer dissatisIaction resulting Irom
untimely delivery is oI high concern to managers. Numerous empirical studies have
documented the high level oI perIormance that on time delivery plays in the operation oI
the supply chain (see Ior example, da Silveira and Arkader, 2007 and Iyer et al. 2004). In
support oI this concern, models Ior evaluating delivery perIormance to the Iinal customer
within multi-stage supply chains have been proposed by several researchers. These
models and the delivery perIormance measures that they contribute serve as an integral
precursor Ior managing improvements in delivery perIormance.
In this paper we develop a cost based delivery perIormance model in which delivery
times are modeling using the gamma pdI. Our model contributes to the literature in that
it: i) overcomes a common limitation that is inherent to delivery models Iound in the
literature, and ii) represents a generalized modeling approach to the evaluation oI supply
chain delivery perIormance.
This paper is organized as Iollows. In Section 2 we provide a review oI the literature on
models oI delivery perIormance and identiIy a common limitation inherent to these
models. In Section 3 we introduce the mathematical Iorm oI a generalized model Ior
evaluating supply chain delivery perIormance that overcomes the limitation oI models
currently Iound in the literature and demonstrate the model. In Section 4, we summarize
our Iindings and discuss directions Ior Iuture research.
4
Literature Review
Models Ior evaluating delivery perIormance within supply chains can be categorized into
two groups: i) index based models, and ii) cost based models. Table 1 provides an
overview oI this literature.
Table 1: ClassiIication oI Supply Chain Delivery PerIormance Models.
Index Based Models Cost Based Models
Nabhani and Shokri (2009) Shin et al. (2009)
Wang and Du (2007) GuiIIrida and Jaber (2008a)
Choudhary et al. (2006) GuiIIrida et al. (2008b)
Garg et al. (2006) GuiIIrida and Nagi (2006a)
GuiIIrida and Nagi (2006b)

Both categories oI models are similar in that delivery timeliness to the Iinal customer is
analyzed with regard to the customer`s speciIication oI an on time delivery window.
Under the concept oI a delivery window, contractually agreed upon benchmarks in time
are used to classiIy deliveries as being early, on time, and late (see Figure 1).


1
c
2
c
Early delivery
1
c
Late delivery
b c A
2

On-time
Delivery
2 1
c c A A

Figure 1. Illustration oI Delivery Window.


Legend: random variable deIining delivery time

2 1
, c c benchmark times deIining early, on-time, and late delivery

5
All the models deIined in Table 1 use the delivery window to assess the probabilities oI
early, on time and late delivery. The models diIIer in how they report delivery
perIormance in terms oI an overall metric. Index based models translate the probability oI
untimely (early and late delivery) into a 'delivery capability index measure which is
similar to the Iamily oI process capability indexes that have been traditionally used in
statistical process control activities in manuIacturing. Cost based models translate the
probability oI untimely delivery into an expected cost measure.
The two classes oI models are elegant in their application oI statistical theory to the task
oI evaluating supply chain delivery perIormance; however, these models have two major
limitations. First, the models deIine delivery time as a Gaussian random variable. By
deIinition, a Gaussian random variable is deIined over the range oI negative inIinity to
positive inIinity. In reality, a random variable Ior representing delivery time can only be
deIined Irom zero to positive inIinity. As demonstrated in GuiIIrida and Jaber (2006),
using the Gaussian to model a non-negative random variable can lead to signiIicant
underestimation when the coeIIicient oI variation exceeds 0.25. Second, by using the
Gaussian pdI the models are restricting the delivery distribution to be symmetric. We
argue that a modeling delivery time using the Gamma pdI oIIers a more exact and
generalized representation Ior measuring supplies chain delivery perIormance. A gamma
random variable can only take on positive values hence it is naturally suited Ior
measuring delivery times which can not be negative. Since the gamma pdI is not
constrained to be symmetric it supports a more generalized modeling approach Ior a
delivery distribution.


6
Model Development

Consider an n-stage serial supply chain where an activity at each stage contributes to the
overall delivery time to the Iinal customer. Delivery time to the Iinal customer is deIined
to be the elapsed time Irom the dispatch oI an order by the originating supplier in the
supply chain to the receipt oI the product ordered by the Iinal customer in the supply
chain. Delivery lead time is composed oI a series oI activities (manuIacturing,
processing and transportation) at each stage oI the supply chain. The activity duration oI
stage i,
i
W , is deIined by pdI 7 , f
W
that is reproductive under addition with respect to
parameter set 7 . Delivery time to the Iinal customer,

n
i
i
W X
1
, is deIined by the
resulting n-Iold convolution oI 7 , f
W
which has pdI
'
+

'

=
n
i
i X
f
1
, 7 . We assume that
activity durations (manuIacturing, processing and transportation time) at each stage oI the
supply chain independent.

For stage i n i , , 1 let
i i
k Gamma W , ~ - with pdI


k
e
f
k k
W

=
1 -
-
(1)

Ior 0 , scale parameter 0 - , and shape parameter 0 k .


The mathematical Iorm oI the pdI governing delivery time
n
W W W X ...
2 1
can be
determined using moment generating Iunctions.
For the gamma distribution the parameter set = k , - 7 and the corresponding moment
generating Iunction is
7

k
W
t
t M
'
+

'

=
-
-
. (2)

Hence, Ior
n
W W W X = ...
2 1
,


n
k k k
X
t t t
t M
'
+

'

'
+

'

'
+

'

=
-
-
-
-
-
-

2 1

'
+

'

=
=
n
i
i
k
t
1
-
-
. (3)
Examining t M
X
, we recognize,
'
+

'

=
n
i
i
k Gamma X
1
, ~ - .

The general structure oI the cost based delivery model as deIined by GuiIIrida and Nagi
(2006a) is
d f c d f c QH Y
X
c
X
c

C
C

2
1
2 1
) ( (4)
where Y expected penalty cost per period Ior untimely (early and late) delivery
Q constant delivery lot size
H supplier`s inventory holding cost per unit per unit time
penalty cost per time unit late (levied by the buyer)

1
c ,
2
c parameters deIining the on time delivery window
f
X
the probability density Iunction oI delivery time.

We redeIine (4) when f
X
is gamma distributed thereby overcoming the two model
limitations that were identiIied in the literature review.
Consider a two-stage supply chain in which each individual stage activity time is gamma
distributed with common scale parameter- and shape parameters
1
k and
2
k respectively.
Using (3), the pdI oI delivery time
2 1
W W X = is gamma distributed and is deIined by
the random variable
2 1
, ~ k k Gamma X - . Restating (4) with f
X
deIined in terms
oI the gamma pdI leads to (see Appendix I Ior derivation)

8

k
c k
k
c k c
QH Y
-
- -
1 1 1
, 1 ,

k
c k c
k
c k

2 2 2
, , 1 -
-
-
. (5)

The advantage oI using the gamma based delivery perIormance model deIined by (5)
over the Gaussian model is illustrated in the two Iigures below. First, as seen in Figure 2,
the gamma based model is more general than the Gaussian model (G) in its application
since it does not require that the delivery distribution exhibit symmetry. Second, as
illustrated in Figure 3, the gamma model can be applied to evaluate the cost oI untimely
delivery Ior all possible values oI the mean and variance oI delivery time where the
Gaussian model is not. For the range oI values Ior the mean and variance displayed in
Figure 3, a Gaussian based delivery model would underestimate the true total cost oI
untimely delivery as a result oI the high probabilities oI negative delivery times that
would occur when 9 3 4 .

Figure 2. Comparison oI Gamma (G) and Gaussian Based Models in the Skewness
1

and Excess Kurtosis Plane
2
.




9
Figure 3. Cost SurIace Ior Gamma Model Ior K/QH 2, 5
1
C and 10
2
C .




Summary and Directions for Future Research

In this paper we have presented a model Ior evaluating the expected cost oI untimely
delivery to the end customer in a supply chain when delivery times are gamma distributed.
The model presented herein overcomes two limitations in the Gaussian based models
Iound in the literature: i) the gamma model is applicable Ior a wide range oI delivery
distributions and is not constrained to be symmetric, and ii) the gamma model is deIined
Ior only positive delivery values and hence can be used to model delivery distributions in
which a Gaussian model can not be applied.
There are several aspects oI this research that can be extended. First, we can expand the
scope oI the model to include multiple products. Second, we could relax the assumption
oI independence among stage activity times. Lastly, we could determine the optimal
width and placement oI the delivery window within the delivery distribution.
10
References:
Bhagwat, R. and Sharma, M. K. (2007) 'PerIormance measurement oI supply chain
management: A balanced scorecard approach, Computers and Industrial Engineering,
53(1), 43-62.

Choudhary, A. K., Singh, K. A. and Tiwari, M. K. (2006). 'A statistical tolerancing
approach Ior design oI synchronized supply chains, Robotics and Computer-Integrated
Manufacturing, 22, 315-321.

da Silveira, G. J. C. And Arkader, R. (2007). 'The direct and mediated relationships
between supply chain coordination investments and delivery perIormance, International
Journal of Operations & Production Management, 27(2), 140-158.

Garg, D., Naraharai, Y. and Viswanadham, N. (2006). 'Achieving sharp deliveries in
supply chains through variance reduction, European Journal of Operational Research,
171(1), 227-254.

Gartner.com (2010) 'The AMR Supply Chain Top 25 Ior 2010
http://www/gartner.com/resources/201200/201212/theamrsupplychaintop2520.pdI.

GuiIIrida, A. L. and Jaber, M. Y. (2008a). 'Managerial and economic impacts oI
reducing delivery variance in the supply chain, Applied Mathematical Modeling, 32(10),
2149-2161.

GuiIIrida, A. L. and Jaber, M. Y. (2006). 'The expected quantity short per replenishment
cycle Ior truncated normal lead time demand, Journal of Academv of Business and
Economics, 6(2), 88- 95.

GuiIIrida, A. L., Jaber, M. Y. and Rzepka, R. A. (2008b). 'An economic model Ior
justiIying the reduction oI delivery variance in an integrated supply chain, INFOR.
Information Svstems and Operational Research, 46(2), 147-153.

GuiIIrida, A. L. and Nagi, R. (2006a). 'Cost characterizations oI supply chain delivery
perIormance, International Journal of Production Economics, 102(1), 23-26.

GuiIIrida, A. L. and Nagi, R. (2006b). 'Economics oI managerial neglect in supply chain
delivery perIormance, The Engineering Economist, 51(1), 1-17.

Gunasekaran, A. and Kobu, B. (2007) 'PerIormance measures and metrics in logistics
and supply chain management: a review oI recent literature (1995-2004) Ior research and
application, International Journal of Production Research, 45(12), 2819-2840.

Gunasekaran, A., Patel, C. and McGaughey (2004) 'A Iramework Ior supply chain
perIormance measurement, International Journal of Production Economics, 87(3), 333-
347.
11

Iyer, K. N. S., Germain, R. and Frankwick, G. L. (2004). Supply chain B2B e-commerce
and time-based delivery perIormance, International Journal of Phvsical Distribution &
Logistics Management, 34(7/8), 645-661.

Lockamy, A. and McCormack, K. (2004) 'Linking SCOR planning practices to supply
chain perIormance, International Journal of Production Research, 47(9), 2449-2460.

Nabhani, F. and Shokri, A. (2008). 'Reducing the delivery lead time in a Iood
distribution SME through the implementation oI six sigma methodology, Journal of
Manufacturing Technologv, 20(7), 957-974.

Shin, H., Benton, W. C. and Jun, M. (2009). 'QuantiIying suppliers` product quality and
delivery perIormance: A sourcing policy decision model, Computers and Operations
Research, 36, 2462-2471.

Wang, F. K. and Du, T. (2007). 'Applying capability index to the supply chain network
analysis, Total Qualitv Management, 18(4), 425-434.


Appendix I

When delivery time is gamma distributed the expected penalty cost Ior late delivery is

d f c Y
X
c
late

C
=
2
2
(A1)

C C
2 2
1
2
c c
k k k
d
k
e
c d
k
e
-
-
-
-
-
- -
. (A2)

Substituting u - , d du - = and changing the lower limit oI the integral Irom
2
c =
to
2
c u - - = = yields

C C
2 2
1
2
c c
u k u k
late
du
k
e u
c du
k
e u
Y
- -
-
. (A3)

Examining (A3), it is noted that the integrals are oI the Iorm oI the incomplete gamma
Iunction
dt t e a

a t

C


1
, . (A4)

12
For the Iirst integral in (A3), the parameters oI (A4) are deIined as k a = 1 and
2
c - ,
which gives
2
, 1 c k - . For the second integral in (A3), 1 1 = k a and
2
c - ,
which gives
2
, c k - .

Hence,

=
k
c k c
k
c k
Y
late
2 2 2
, , 1 -
-
-
. (A5)

The expected earliness cost is deIined as
d f c QH Y
X
c
earlv

C

1
1
(A6)

1 1
0 0
1 1
1
c c
k k k k
d
k
e
d
k
e
c QH
- -
- -
. (A7)

Substituting u - = , d du - = and changing the upper limit oI the integral Irom
1
c
to
1
c u - - yields

=

1 1
0 0
1
1
c c
u k u k
earlv
du
k
e u
du
k
e u
c QH Y
- -
-
. (A8)

Examining (A8), the integrals are oI the Iorm oI the incomplete gamma Iunction
dt t e a

a t

=
0
1
, . (A9)
For the Iirst integral in (A8), the parameters oI (A9) are deIined as 1 1 = k a and
1
c - = , which gives
1
, c k - . For the second integral in (A8), k a = 1 and
1
c - = ,
which gives
1
, 1 c k - . Hence,

=
k
c k
k
c k c
QH Y
earlv
-
- -
1 1 1
, 1 ,
. (A10)

Combining (A5) and (A10) gives (5).