A HIERARCHICAL VENDOR SELECTION OPTIMIZATION
TECHNIQUE FOR MULTIPLE SOURCING
Mariagrazia Dotoli
Dipartimento di Elettrotecnica ed Elettronica, Politecnico di Bari, Via Re David 200, 7015 Bari, Italy
Marco Falagario
Dipartimento di Ingegneria Meccanica e Gestionale, Politecnico di Bari, Via Re David 200, 7015 Bari, Italy
Keywords: Business Intelligence, Supply Chain Management, Supplier Evaluation and Selection, Decision Support
System, Data Envelopment Analysis, Analytic Hierarchy Process, Linear Programming.
Abstract: The paper addresses a crucial objective of the strategic function of purchasing in supply chains, i.e., vendor
rating, proposing a hierarchical model for supplier business intelligence. A three-level optimization process
for supplier selection in a multiple sourcing strategy context is proposed. First, the Data Envelopment
Analysis, the most widespread method for supplier selection, is used to evaluate the efficiency of suppliers.
Second, the well-known Analytic Hierarchy Process is applied to rank the efficient suppliers given by the
previous step. Third, a linear programming problem is solved to find the quantities to order from each
efficient supplier. We show the model effectiveness on a simulated case study of a C class component.
1 INTRODUCTION
A Supply Chain (SC) is a business network
interconnecting independent manufacturing and
logistics companies that perform critical functions in
the order fulfilment process (Dotoli et al., 2006).
The SC configuration is essential to pursue a
competitive advantage and meet the market demand.
This paper focuses on one of the strategic
purchasing function tasks in a private SC, i.e.,
vendor ranking (Costantino et al., 2009). Vendor
rating systems identify top suppliers, i.e., the
candidate partners that are best equipped to meet the
customer’s expected level of performance, and
check them periodically. Therefore, vendor selection
is a multi-objective decision problem, including
conflicting objectives such as, besides the obvious
goal of (low) price, quality, quantity, delivery,
performance, capacity, communication, service,
geographical location etc. (Degraeve et al., 2000).
Numerous multi-criteria decision making
approaches have been suggested to solve the vendor
evaluation and selection problem and, among these,
individual approaches and integrated ones can be
distinguished. The most important individual
methods are: the Data Envelopment Analysis
(DEA), mathematical programming, the Analytic
Hierarchy Process (AHP), case-based reasoning,
fuzzy decision making, genetic algorithms and many
more. The so-called integrated approaches join
together different techniques (e.g., integrated AHP,
DEA, and artificial neural networks, integrated AHP
and goal programming, etc.). Individual approaches
are more popular than integrated ones, with the most
widespread individual technique being DEA, due to
its robustness (Ho et al. 2010) and its ability to be
implemented also considering qualitative criteria: as
an example, Talluri et al. (2006) extend the classical
DEA technique considering risk evaluation.
However, DEA presents the drawback that its
efficient alternatives are by definition equally
optimal and no difference can be singled out with
respect to their different effectiveness.
In the private sector, the buyer can choose
between a single or multiple sourcing approach.
Single sourcing is defined as the fulfilment of all
corporate requirements for a particular product by a
selected supplier. On the other hand, multiple
sourcing is the splitting of an order among multiple
sellers, i.e., the company has two (dual sourcing) or
195
Dotoli M. and Falagario M..
A HIERARCHICAL VENDOR SELECTION OPTIMIZATION TECHNIQUE FOR MULTIPLE SOURCING.
DOI: 10.5220/0003088801950200
In Proceedings of the International Conference on Knowledge Management and Information Sharing (KMIS-2010), pages 195-200
ISBN: 978-989-8425-30-0
Copyright
c
2010 SCITEPRESS (Science and Technology Publications, Lda.)
more suppliers for the same component. Obviously,
each solution presents advantages and drawbacks.
In this paper we propose a hierarchical strategy
for optimal supplier evaluation and selection in
multiple sourcing supplies based on three levels.
First, we use the well-known DEA method to
evaluate the weights of input and output criteria and
divide suppliers into two categories: efficient and
inefficient ones. Second, we apply the widespread
decision making AHP technique (Saaty, 1990) to
rank the efficient alternatives and select the effective
ones. AHP is a multi-objective decision technique in
which all the elements of the decision problem
(overall goal, criteria, alternatives) are arranged in a
hierarchical structure and objectives are of varying
degrees of importance. Although in many cases
optimization methods lead to similar results, here we
select AHP because it relies on pairwise
comparisons of the solutions, providing an approach
to rank alternatives based on their reciprocal
assessment. Third, after ranking the efficient
solutions and identifying the most effective ones, a
linear programming problem is solved to calculate
the quantities of product to require from each
effective supplier in the multiple sourcing context.
Summing up, we provide a decision support tool for
supplier business intelligence, to rank vendors and
provide the buyer with a simple instrument to
determine the quantities to order from each effective
supplier in a multiple sourcing strategy context.
2 THE HIERARCHICAL
SUPPLIER SELECTION
TECHNIQUE
A vendor selection problem is defined by a set of
bidding suppliers
{}
12
, ,.....,
F
Sss s= and a set of
conflicting criteria
{}
12
, ,.....,
n
Ccc c= , according to
which vendors have to be ranked. The criterion set is
partitioned as
I
O
CC C=∪, with
{}
12
, ,.....,
IH
Ccc c= ,
{}
12
, ,.....,
OHH HK
Ccc c
++ +
= ,
and H+K=n respectively representing the input and
output criteria sets, and the criteria number.
The input criteria are defined as the supplier
attributes considered before the supply takes place
(e.g., price, geographical distance of the supplier,
ICT integration, etc.) while the output criteria are
connected to the supplier once the goods arrive at
the firm (e.g., quality, reliability, lead time, etc.).
Figure 1 shows the presented hierarchical integrated
approach to determine effective suppliers and the
requested product quantities.
2.1 The First Level of the Hierarchical
Optimization - the DEA Method
The first level of the supplier selection approach in
Fig. 1 employs the Data Envelopment Analysis
(DEA) (Charnes et al., 1978), a linear programming-
based technique for determining the efficiency of
different decision making units. As regards the
application of DEA to supplier selection, the
strength of this technique is the distinction between
input and output performance measures. Input
performance is given by the amount of resource used
by the vendor to carry out the supply process (for
instance, the purchasing price), while output
parameters express how good is the service provided
by the suppliers to the buyer (examples for these are
the quality of purchased product or the timeliness of
deliveries).
The efficiency of supplier
f
s
S is defined as:
1
1
K
kkf
k
f
H
hhf
h
uy
E
vx
=
=
=
with f=1,…,F,
(1)
where
y
kf
(x
hf
) is the k-th (h-th) output (input)
performance for the f-th actor and u
k
(v
h
) its weight.
Level 3
Query Linear Programming
Problem solution
Level 2
Query AHP
Level 1
Query DEA
Input Data
(input and output
performance values)
Input Data
for efficient suppliers
(input and output
performance values)
Input Data
for effective suppliers
(PI
i_AHP
AHP
performance values)
Hierarchical
Model
Design
Effectivesuppliers and required quantities
Efficient
suppliers
Ranking of
efficient
suppliers
Required
quantities
from
effective
suppliers
Suppliers set
Level 3
Query Linear Programming
Problem solution
Level 2
Query AHP
Level 1
Query DEA
Input Data
(input and output
performance values)
Input Data
for efficient suppliers
(input and output
performance values)
Input Data
for effective suppliers
(PI
i_AHP
AHP
performance values)
Hierarchical
Model
Design
Effectivesuppliers and required quantities
Efficient
suppliers
Ranking of
efficient
suppliers
Required
quantities
from
effective
suppliers
Suppliers set
Figure 1: The hierarchical supplier selection approach.
In the DEA method, the efficiency of each actor is
obtained by determining the set of coefficients u
k
and
v
h
which maximizes this value, taking into
account that for each actor it holds by definition
KMIS 2010 - International Conference on Knowledge Management and Information Sharing
196
1
f
E . Hence, the measure of supplier efficiency
can be obtained by solving the following
optimization problem for each considered vendor:
max
f
E with f=1,…,F,
(2)
subject to (s.t.):
1
1
1
K
kkj
k
H
hhj
h
uy
vx
=
=
with j=1,…,F,
(3)
,0
kh
uv for k=1,…,K and h=1,…,H.
(4)
Problem (2)-(3)-(4) can be linearized by
minimizing the inputs and keeping fixed output
values (input-oriented method) or maximizing the
outputs and keeping fixed input values (output-
oriented method) (Wang and Chin, 2010). Using the
latter solution, the problem is modified as follows:
1
max
K
fkkf
k
Euy
=
=⋅
with f=1,…,F,
(5)
s.t.:
11
0
KH
kkj hhj
kh
uy vx
==
⋅−
∑∑
with j=1,…,F,
(6)
1
1
H
hhf
h
vx
=
⋅=
with f=1,…,F,
(7)
and (4).
The efficiency of analyzed suppliers can be found
solving problem (5)-(6)-(7)-(4) for each
f-th supplier
for
f=1,2 ,…,F. Obviously, the f-th vendor is
maximally efficient if
E
f
=1. Therefore, suppliers can
be ranked based on their efficiency value
E
f
.
2.2 The Second Level of the
Hierarchical Optimization - the
AHP Approach
The Analytic Hierarchy Process (AHP) is a multi-
objective decision technique (Saaty, 1990) for
ranking a number of alternatives according to a set
of conflicting criteria of various degrees of
importance. This paper selects AHP to single out in
the second level of the optimization the effective
suppliers among efficient ones determined at the
first level since, being based on alternatives pairwise
comparison, AHP turns out to exhibit an enhanced
accuracy with respect to other decision making
techniques. AHP consists of the following steps.
Step 1. Structuring the decision problem as a
hierarchy.
Select the first level of the hierarchical
structure as the overall goal “Effectiveness”. Define
the second level, composed by the
n criteria
contributing to the goal. Determine the third level as
the
m alternative suppliers to be ranked in terms of
the criteria in the second level.
Step 2. Constructing the decision matrix.
Determine the decision matrix D of dimensions mxn,
where m is the number of alternatives (the efficient
suppliers),
n is the number of criteria, and element
d
ij
with i=1,…,m and j=1,…,n measures the i-th
supplier performance against criterion c
j
.
Step 3. Constructing the pairwise comparison
matrix
C
M
. Compare the n criteria with each other
and construct the nxn pairwise comparison matrix
C
M
by Saaty’s original AHP scale in Table 1. More
precisely, determine each element
ij
m
c of C
M
with
i,j=1,…,n, representing the relative importance of
the i-th criterion compared to the j-th one, by
interviewing the buyer evaluating the importance of
criterion
c
i
over c
j
and associating it an integer value
from 1 to 9 according to Table 1. Obviously, less
important criteria are defined by reciprocals
1
ij
j
i
m
m
c
c
= for each i,j=1,…,n.
Step 4. Determining the eigenvector associated
to the maximum eigenvalue of the comparison
matrix
. Calculate the eigenvalues set
{
λ
1
,
λ
2
,…,
λ
R
} of C
M
, where R is its rank. Let
λ
max
be the maximum eigenvalue of C
M
, then determine
its eigenvector v
max
. Compute the priority vector:
1
[...]
T
n
npp=⋅=
max
Pv
.
(8)
where each element p
j
with j=1,…,n of P represents
the importance degree of the j-th performance index
associated to the j-th column of D’: the greater p
j
,
the more important the j-th performance index.
Step 5. Raising alternatives to the criteria power.
Determine the alternative values associated to each
j-th performance index as follows:
1
[ ... ]
mj
dd
=
j
CRIT .
(9)
A HIERARCHICAL VENDOR SELECTION OPTIMIZATION TECHNIQUE FOR MULTIPLE SOURCING
197
Table 1: Saaty’s AHP scale of comparisons.
Intensity of importance Definition
1 Equal importance
3 Moderate importance
5 Strong importance
7 Very strong importance
9 Extreme importance
2,4,6,8 Intermediate values between
the two adjacent judgments
for each j=1,…,n. Determine the following vectors.
G
j
=[g
1j
… g
mj
]=
1
[...]
jjj
ppp
jmj
dd=
j
CRIT .
(10)
for each j=1,…,n.
Step 6. Determining the decision model. For each
alternative i with i=1,…,m, determine:
(
)
_1
min ,...,
iAHP i in
PI g g=
(11)
so that PI
i_AHP
provides information about the
satisfaction of alternative s
i
with respect to the
performance indices and their importance degree.
Step 7. Ranking the alternatives. Suppliers are
ranked according to index PI
i_AHP
: the best supplier
is the one showing the highest index PI
i_AHP
.
2.3 The Third Level of the Hierarchical
Optimization: the Linear
Programming Methodology
Linear programming is a mathematical optimization
process in which a single objective function states
mathematically what is being maximised, e.g.,
profit, or minimized, e.g., cost.
With the aim of determining the quantities to
require from the most effective suppliers singled out
in the previous and second level of the hierarchical
supplier evaluation procedure, we define the Supply
Evaluation Index (SEI) as follows:
_
1
iiAHP
i
SEI q PI
μ
=
=⋅
(12)
that is an overall index measuring the efficiency on
the supply considering the μm most effective
suppliers obtained by the second-level AHP
optimization among the m efficient vendors obtained
by the first-level DEA optimization. In particular,
variables q
i
with i=1,.., μ are the percentage
quantities of product with values ranging from 0 to 1
to request from each vendor to obtain the supply.
Hence, the linear programming problem is:
()
M
ax SEI
(13)
s.t.:
1
1
i
i
q
μ
=
=
,
(14)
ii
q
γ
with 01
i
γ
and i=1,…, μ,
(15)
ii
q
δ
with
1
1
i
i
μ
δ
=
.
(16)
In particular, δ
i
is a parameter measuring the
minimum percentage quantity (eventually equal to
zero) that the buyer decides to buy from each
effective supplier independently from its ranking to
keep the long-term partnership. In addition, γ
i
is the
given production capacity (expressed in percentage
values in a 0-1 range) of the i-th effective supplier
with i=1,…,μ. Hence, (14) guarantees that the whole
requested quantity is supplied, constraints (15) are
connected to the quantities each supplier is able to
deliver, (16) models the buyer will of requiring
products from each efficient supplier independently
from the ranked position.
3 THE CASE STUDY
To show the effectiveness of the presented
hierarchical approach, we consider a simulated case
study requiring the supply of C class components
under multiple sourcing and assuming that the
number of existing suppliers equals F=15. We
remind that spare parts in inventory are usually
divided in the literature into three classes according
to their money usage (Krajewski and Ritzman,
2002): class A items typically represent only about
20% of the items but account for 80% of the money
usage; class B components account for additional
30% of the items but only for 15% of the money
usage; finally, 50% of the items falls in class C,
representing a mere 5% of financial usage. While for
A and B components a strategic partnership between
buyer and seller is typically created (so that often
single sourcing is applied), C components are such
that an increasing competition among suppliers
usually allows the buyer to obtain a better price:
hence, it is important to rank suppliers and decide
the quantities to request them by different criteria.
The case study vendor efficiency is estimated
using H=2 input criteria, namely:
KMIS 2010 - International Conference on Knowledge Management and Information Sharing
198
price - This attribute measures the price offered
by each supplier. It is evaluated as
max
f
f
p
p
p
=
with f=1,…,F, where p
f
is the offered price and
max
1,2,...,
max ( )
f
fF
pp
=
= the maximum offered price;
geographical distance - this criterion expresses
the geographical distance of the supplier from
the buyer. The nearer the supplier, the lower the
transportation costs. The normalized
performance of the f-th supplier with f=1,…,F is
max
f
f
d
d
d
=
, where d
f
is the vendor distance and
max
1,2,...,F
max ( )
f
f
dd
=
=
the maximum distance.
The K=2 considered output criteria are:
quality - this criterion is strictly related to the
number of accepted products: indeed, a high
number of defects means high costs of
restoration. Hence, we define the index quality of
the f-th supplier with f=1,…,F as
,,
,,
af af
f
vf vf
pc lot
IQ
pc lot
=⋅
, where pc
a,f
(pc
v,f
) is the
number of accepted (verified) items and lot
a,f
(lot
v,f
) is the amount of accepted (verified) lots.
The normalized quality index is hence
max
f
f
I
Q
IQ
IQ
=
, with
max
1,2,...,F
max ( )
f
f
I
QIQ
=
=
;
lead time - This criterion is related to the supplier
manufacturing capability and flexibility. The
lead time is defined as the time span between the
placing of an order and the receipt of goods.
Obviously, the shorter the lead time, the better
the supplier in term of flexibility, production
capability and internal organization. Given the
lead time index LT
f
of the f-th supplier with
f=1,…,F, the normalized lead time is
max
1
f
f
LT
LT
LT
=−
, with
max
1,2,...,
max ( )
f
fF
LT LT
=
=
.
The normalized input performance values of each
supplier are collected in Table 2 (second and third
column). In the second-last and last column of Table
2 the output indices are reported.
Applying the DEA approach, problem (5)-(6)-
(7)-(4) is defined and solved, so that the results in
Table 3 are obtained. Analysing Table 3, the
efficient suppliers are supplier 3, 5, 10, 11, and 14,
so that m=5 suppliers are singled out. For example,
supplier 3 is efficient by weighting the price
criterion u
1
=0.389, the normalized geographical
distance u
2
=3.591, the quality index v
1
=0.517, and
the lead time index v
2
=0.942.
The next step is to rank the efficient suppliers 3,
5, 10, 11, and 14 in order to calculate the quantities
to require for a supply. The results of the AHP
optimization are shown in Table 4, collecting the
performance values of vendors s
f
with
f=3,5,10,11,14. The second column reports
performance index PI
i_AHP
and the last column ranks
the five efficient suppliers: the best supplier is s
10
,
showing a high value of lead time and a low value of
price (the lowest), together with an intermediate
geographical distance and a high quality index.
Following are suppliers s
14
, s
11
, s
3
, and s
5
.
Table 2: The data for the DEA input and output criteria.
Supplie
r
Input 1 Input 2 Output 1 Output 2
F
f
p
f
d
f
I
Q
f
L
T
1
0.689 0.456 0.894 0.237
2
1.000 0.538 0.998 0.347
3
0.798 0.192 0.985 0.521
4
0.790 0.594 0.946 0.125
5
0.589 0.066 0.902 0.000
6
0.487 0.987 0.945 0.568
7
0.897 1.000 0.976 0.625
8
0.657 0.456 0.928 0.544
9
0.984 0.732 1.000 0.875
10
0.123 0.450 0.756 0.757
11
0.235 0.200 0.912 0.359
12
0.357 0.759 1.000 0.915
13
0.573 0.417 0.350 0.830
14
0.233 0.350 0.870 0.765
15
0.467 0.897 0.910 0.935
Table 3: The first-level DEA optimization data and results.
Supplie
r
Weight Weight Weight Weight Efficiency
f u
1
u
2
v
1
v
2
E
f
1
0.520 1.407 0.443 0.000 0.396
2
0.168 1.547 0.223 0.406 0.363
3
0.389 3.591 0.517 0.942 1.000
4
0.417 1.129 0.355 0.000 0.336
5
1.303 3.525 1.109 0.000 1.000
6
1.358 0.343 0.425 0.000 0.402
7
0.122 0.890 0.094 0.338 0.303
8
0.251 1.831 0.193 0.695 0.557
9
0.090 1.245 0.000 0.597 0.522
10
1.695 1.759 0.000 1.321 1.000
11
2.442 2.131 0.949 0.374 1.000
12
1.095 0.802 0.330 0.325 0.628
13
0.158 2.181 0.000 1.046 0.868
14
1.463 1.883 0.499 0.740 1.000
15
0.715 0.742 0.000 0.558 0.521
A HIERARCHICAL VENDOR SELECTION OPTIMIZATION TECHNIQUE FOR MULTIPLE SOURCING
199
Table 4: The second-level AHP optimization results.
Efficient supplierPerf. index Position
f
PI
iAH
P
i
3
0.132 4
5
0.117 5
10
0.273 1
11
0.219 3
14
0.258 2
Table 5: The third-level linear programming problem data.
Supplier Minimum requested quantity Capacity
f δ
i
γ
i
10 0.200 0.600
11 0.200 0.400
14 0.200 0.700
Table 6: The third-level linear programming results.
Supplier Required quantity
f q
i
10 0.400
11 0.200
14 0.400
After rnking the efficient suppliers, the linear
programming problem (13)-(14)-(15)-(16) is defined
and solved for a multiple sourcing strategy with μ=3.
The values of minimum required quantities δ
i
and
percentage capacities γ
i
of suppliers are collected in
Table 5. Table 6 shows the results of the linear
programming problem solution, i.e., the required
quantities from the three effective suppliers s
10
, s
11
,
and s
14
. Results show that 40% of the supply will be
provided in turn by each of the two most effective
suppliers, i.e., s
10
and s
14
, whereas the remaining
20% of the requested product will be provided by
the third-ranked supplier s
11
.
4 CONCLUSIONS
The paper focuses on a crucial issue of purchasing in
supply chains, i.e., vendor evaluation and selection.
A novel three-step methodology based on the Data
Envelopment Analysis (DEA) approach, the
Analytic Hierarchy Process (AHP), and linear
programming is presented. At the first level of the
hierarchical technique, a vendor rating technique
based on DEA is devised to obtain efficient and
inefficient suppliers. Hence, the AHP process is
applied to rank the efficient vendors based on the
overall performance index. Finally, a linear
programming problem is solved to split the supply
adopting a multiple sourcing strategy. To the best of
the authors’ knowledge, no one in the related
literature has ever joined these three approaches for
supplier selection in such a context. A numerical
case study shows the effectiveness of the presented
three step method for a C class component. Future
perspectives are identifying a real case study to
further verify the approach flexibility and simplicity
of use by the firm purchasing manager.
ACKNOWLEDGEMENTS
This work was supported by the TRASFORMA
“Reti di Laboratori” network funded by Apulia
Italian Region.
REFERENCES
Charnes A., Cooper W.W., Rhodes E., 1978. Measuring
the efficiency of decision making units; European
Journal of Operational Research, 2, 429–444.
Costantino N., Dotoli M., Falagario M., Fanti M.P.,
Iacobellis G., 2009. A Decision Support System
Framework for Purchasing Management in Supply
Chains, Journal of Business and Industrial Marketing,
24(3/4): 278-290.
Degraeve Z., Labro E., Roodhooft F., 2000. An evaluation
of vendor selection models from a total cost of
ownership perspective; European J. of Operational
Research, 125(1), 34-58.
Dotoli M., Fanti M.P., Meloni C., Zhou M.C., 2006.
Design and optimization of integrated e-supply chain
for agile and environmentally conscious
manufacturing, IEEE Transactions on Systems Man
and Cybernetics, part A, 36(1), 62-75.
Ho W., Xu X., Dey P.K., 2010. Multi-criteria decision
making approaches for supplier evaluation and
selection: A literature review, European Journal of
Operational Research, 202, 16-24.
Krajewski L., Ritzman L., 2002. Operation Management:
Strategy and Analysis, Prentice Hall, London, UK, 6th
ed.
Saaty T.L., 1990. How to Make a Decision: the Analytic
Hierarchy Process, European Journal of Operational
Research, 48:9-26.
Talluri, S., Narasimhan, R., Nair, A., 2006. Vendor
performance with supply risk: A chance-constrained
DEA approach; International Journal of Production
Economics 100(2), 212–222.
Wang Y.-M., Chin K.-S., 2010. A neutral DEA model for
cross-efficiency evaluation and its extension; Expert
Systems with Applications, Elsevier, 37, 3666-3675.
KMIS 2010 - International Conference on Knowledge Management and Information Sharing
200