Decision Support Systems 54 (2012) 206–214

Contents lists available at SciVerse ScienceDirect

Decision Support Systems

journal homepage: www.elsevier.com/locate/dss

A hybrid MCDM methodology for ERP selection problem with interacting criteria
Tuncay Gürbüz a, S. Emre Alptekin a, Gülfem Işıklar Alptekin b,⁎
a
Department of Industrial Engineering, Galatasaray University, Çırağan Cad. No:36 34357 Ortaköy, İstanbul, Turkey
b
Department of Computer Engineering, Galatasaray University, Çırağan Cad. No:36 34357 Ortaköy, İstanbul, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: An enterprise resource planning (ERP) system is an information system to plan and integrate all of an
Received 5 August 2011 enterprise's subsystems including purchase, production, sales and finance. Adopting such a comprehensive
Received in revised form 20 February 2012 framework may result in the great savings in both costs and man hours. This research explores the applica-
Accepted 10 May 2012
tion of a hybrid multicriteria decision making (MCDM) procedure for the evaluation of various ERP alterna-
Available online 22 May 2012
tives. The proposed evaluation framework integrates three methodologies: Analytic Network Process (ANP),
Keywords:
Choquet integral (CI) and Measuring Attractiveness by a Categorical Based Evaluation Technique (MAC-
ERP BETH). ANP produces the priorities of alternatives with respect to the interdependent evaluation criteria.
Supplier selection The conjunctive or disjunctive behaviors between criteria are determined using MACBETH and CI. Numerical
MCDM application of the proposed methodology is implemented on the decision making problem of a firm that faces
ANP with four ERP projects. The final ranking is compared to the one obtained by ignoring the interactions among
Choquet integral criteria. The results demonstrate that the ignorance of the interactions may lead to erroneous decisions.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction
Due to severe market competition, companies have directed to
consider alternative business environments in order to reduce total
cost, maximize return on investment, shorten lead times and be
more responsive to customer requirements. An enterprise resource
planning (ERP) system can be considered as a solution for inefficient
business processes. Organizations think about purchasing it when
they deal with a number of complex and interrelated business trou-
bles, such as achieving company's financial goals, managing and
streamlining the company's operational processes, better forecasting
features or obtaining the benefits of improved information manage-
ment by reducing data duplication. An ERP system typically imple-
ments a common enterprise-wide database together with a range of
application modules [13]. It standardizes processes and stores infor-
mation as well as recalls that data when it is required in real time en-
vironment. Implementing an ERP system may be costly and time-
consuming. Companies spend billions of dollars and use numerous
amounts of man-hours for installing elaborate ERP software systems.
However, the benefits of a successful ERP project are worthwhile.
The offered ERP software packages cannot provide a once-for-all busi-
ness model for every process of all industries. In other words, no single
ERP packaged software can meet all company functionalities or all special
⁎ Corresponding author.
E-mail addresses: tgurbuz@gsu.edu.tr (T. Gürbüz), ealptekin@gsu.edu.tr
(S.E. Alptekin), gisiklar@gsu.edu.tr (G. Işıklar Alptekin).
business requirements [33]. Thus, companies must choose a flexible ERP
system that is responsive to customer requirements. The major reason for
ERP implementation failures is stated as the inappropriate system selec-
tion [13,22]. An inappropriate selection process can significantly affect
not only the implementation but also the performance of the company
[9]. Therefore, the importance of selecting a suitable ERP system cannot
be overemphasized [19].
This paper introduces a hybrid multicriteria decision making (MCDM)
model for ERP selection based on Analytic Network Process (ANP),
Choquet Integral (CI) and Measuring Attractiveness by a Categorical
Based Evaluation Technique (MACBETH). MCDM refers to find the
best opinion from all of the feasible alternatives in the presence of mul-
tiple, usually conflicting, decision criteria. It is a branch of a general class
of operations research models that deals with the decision making
problems under the presence of a number of decision making alter-
natives described by their attributes [8]. Priority based, outranking,
distance-based and mixed methods could be considered as the primary
classes of the current methods [26]. One of the most outstanding MCDM
approaches is ANP which is a generalization of one of the most known
MCDM methodologies: Analytic Hierarchy Process (AHP) [30]. While
AHP represents a framework with a unidirectional hierarchical relation-
ship, ANP allows for complex interrelationships among decision levels
and criteria. ANP is used in many decision support systems of various
types. Verdecho et al. [32], have provided a methodology based on
ANP to prioritize and manage inter-enterprise performance at both
the strategic and the process level. Another research has explored the
application of ANP approach for the evaluation of R&D projects that
are elements of programs with heterogeneous objectives [21]. Although
there are various methods, such as mathematical programming, MCDM

0167-9236/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.dss.2012.05.006
 T. Gürbüz et al. / Decision Support Systems 54 (2012) 206–214 207

analysis or scoring, which is applied to the ERP software selection, ANP framework has three basic features which are useful in mul-
we focus on studies including MCDM analysis. Wei et al. [33] have intro- ticriteria decision making problems:
duced their selection framework that is based on AHP method. Erensal
and Albayrak [14] have suggested AHP to effectively enhance adoption • Define the goal and criteria (and sub-criteria),
of macroergonomics and to improve management decision perfor- • Determine the interdependencies and the network,
mance in measuring and comparing the overall performance of differ- • Build the supermatrix and synthesize.
ent management styles based on macroergonomical criteria. The most
similar work to our proposed framework is the work of Yazgan et al.
In this approach, comparison matrices, prioritization and the
[34]. However, they have designed an artificial neural network model
weights while considering the interdependencies are formed between
and trained with using ANP results in order to calculate ERP software
various attributes of each level with the scale of 1–9 suggested by
priorities.
Saaty [29]. Also the consistencies of the pairwise comparisons, made
In our paper, we have categorized ERP selection criteria into three
by the experts or decision makers (DMs), have to be checked in order
main sets: vendor related criteria (VRC), customer related criteria
to make the necessary changes, if there is any inconsistency above the
(CRC), and software related criteria (SRC). Each one consists of its own
allowed limit. Once the pairwise comparison matrices are formed,
sub-criteria set. Since these criteria and sub-criteria have both inner
weight vectors for all the matrices are calculated. The concept of super-
and outer dependencies, we have made use of the ANP to determine
matrix is used to obtain the composite weights that overcome the exis-
these dependencies and relative priorities of all criteria. MACBETH is
ting interrelationships. The synthesizing step is to rate the alternatives
both an approach and a set of techniques that have the goal of providing
according to all the criteria, compute the overall score for the alterna-
an overall ordering of options, and that aid on the construction of inter-
tives and make the final decision as to choose the best alternative or
val numerical scales based on qualitative (non-numerical) pairwise
to obtain the final ranking of the alternatives.
comparison judgments [4,12].
In this research, we have used CI to determine conjunctive or dis-
2.2. Choquet Integral (CI)
junctive behaviors between criteria and integrated MACBETH in CI in
order to define the parameters of CI. The last phase of the proposed
The CI, which has been introduced in the fuzzy measure community
methodology involves ranking the given ERP alternatives according
by Murofushi and Sugeno [25] is a fuzzy integral proposed by Choquet
to their final performance scores. We have shown the feasibility of
[11] and considers the interactions between k out of n criteria of the
the framework on the decision making problem of a company that
problem, which is called the k-additivity property. Basic notations and
needs to evaluate four ERP software alternatives and select the most
definitions on the CI can be analyzed in Appendix A. In this research,
suitable one according to its requirements. Furthermore, we have
2-additive CI is used.
obtained another ranking by using the same evaluation values; how-
Letting ti, i = 1, …, n be the scores on the criteria, by using only the
ever this time we have ignored the interactions among criteria. The
interaction index, it is possible to express CI in the case of 2-additive
comparison has shown that the final ranking may change dramatical-
measures as follows [15]:
ly with the ignorance of the interactions and therefore it may lead us
to erroneous decisions. n
     X  1  
The remaining part of the paper is organized as follows: Section 2 C μ ðt 1 ; …; t n Þ ¼ ∑ t i ∧t j I ij þ ∑ t i ∨t j Iij  þ t i φi − ∑ Iij 
   
Iij>0 I ij b0 2 j≠i
describes the methodologies that constitute the proposed framework. i¼1
ð1Þ
The steps and details of the proposed decision support framework are 1  
with φi − ∑ Iij ≥0; ∀i ¼ 1; …n

introduced in Section 3. The implementation into the ERP selection
problem is presented in Section 4. Finally, Section 5 gives the con-
cluding remarks of the study.
2. Preliminaries
In this section, three methodologies underlying the proposed frame-
work (ANP, CI and MACBETH) along with their integration will be
presented.
2.1. Analytic Network Process (ANP)
ANP, as well as AHP, incorporates both qualitative and quantita-
tive approaches to a decision problem [10]. It is also capable of cap-
turing the tangible and intangible aspects of relative criteria that
have some bearing on the decision making process [30]. AHP is lim-
ited to relatively static and unidirectional interactions with little
feedback among decision components and alternatives [20]. How-
ever, many real life decision problems cannot be structured as a hi-
erarchy because of the interactions and dependence among criteria.
Therefore, the hierarchy becomes more like a network. On this con-
text, ANP and its supermatrix technique can be considered as an ex-
tension of AHP that can handle a more complex decision structure
as the ANP framework has the flexibility to consider more complex
interrelationships (outer dependence) among different elements
[30,31]. Hence, ANP is very useful in these kinds of situations pro-
viding a general framework without the assumptions of indepen-
dence of higher-level elements from lower ones, or independence
on the same level.
2 j≠i
Here, φi represents the relative importance of criterion i with
n
φi ¼ 1 and Iij, defined in the interval [− 1;1], is the interaction
P
i¼1
value between criteria i and j. Appendix B presents the definitions.
Three cases may be observed for interaction value:
• Positive values of Iij implies a conjunctive behavior between criteria
i and j. i.e. simultaneous satisfaction of both criteria is significant for
the global score.
• Negative values of Iij implies a disjunctive behavior between criteria
i and j. i.e. the satisfaction of either one is sufficient to have a signif-
icant effect on the global score.
• If Iij is null, then there is no interaction between criteria i and j. If for
all pairs of criteria, Iij are null then the φi value acts as a weight vector
in a weighted arithmetic mean. This represents the linear part of CI.
The use of CI will be demonstrated on a simple example from Saad et
al.'s [28] work. Suppose we were to consider three students, A, B, and C,
with respect to three subjects, mathematics, physics and literature in a
school that is more scientifically oriented. Hence, the weights for these
three subjects may be 3, 3 and 2, respectively. In that case, using weighted
sum method and marks given on a scale from 0 to 20, the average for
these students are calculated and given in Table 1.
However, if one desires to favor well equilibrated students with-
out weak points, the weighted sum may not be the best suited meth-
od to use since student A, although having a considerable weakness in
literature has been considered better than student C, who has no
208 T. Gürbüz et al. / Decision Support Systems 54 (2012) 206–214

Table 1 equation, where α is a coefficient necessary to meet the condition and

Students' grades. p k and p l ∈[0;1], will be obtained:
M P L Weighted sum

A 18 16 10 15.250
k h l k l
B 10 12 18 12.750 A ≻ A ⇔pi −pi ¼ hα ð2Þ
C 14 15 15 14.625

weak point. If one were to use CI in order to find the average for these
students, the following logic can be applied:
• Since the school is scientifically oriented, following weights can be
appropriate for the subjects: μ(M) = μ(P) = 0.45 and μ(P) = 0.30.
• Since mathematics and physics are redundant, the weight attribut-
ed to ‘mathematics and physics’ can be chosen to be less than the
sum of their individual weights: μ(M & P) = 0.5 b 0.45 + 0.45.
• With the same logic, to favor the students equally good in scientific
subjects and literature, the weight attributed to ‘mathematics and lit-
erature’ and ‘physics and literature’ can be chosen to be greater than
the sum of their individual weights as follows: μ(M & L) = μ(P & L) =
0.9 > 0.45 + 0.3.
• By definition μ(M & P & L) = 1 and μ(ϕ) = 0.
Using this fuzzy measure and Eq. (1), the average of these stu-
dents are calculated as: 13.9 for student A, 13.6 for student B and
14.9 for student C. The new results place student C to the top position
instead of student A.
2.3. Measuring Attractiveness by a Categorical Based Evaluation Technique
(MACBETH)
MACBETH is a multicriteria decision analysis approach which has
first been proposed in 1992 [2]. MACBETH has been used in various
fields such as individual's career choice [3], evaluation and com-
Therefore, a preference ranking of alternatives for a specific crite-
rion, including the ‘good’ and the ‘neutral’ alternatives, collected from
a DM with the strength of the comparisons will give us a system of
equation. After solving it, the individual performance values of the
alternatives for the criterion in question will be determined.
2.4. MACBETH and Choquet Integral
This sub-section includes the use of MACBETH in order to find the CI
parameters given in Eq. (2). The operators of the CI family can be written
under the form of a conventional weighted mean modified by effects
coming from interactions between elementary performances [12]. They
are stable under linear transformation [23] and thus consistent with
the performance expressions on an interval scale on the universe [0;1].
This property constitutes the theoretical basis of the extension of the
MACBETH procedure to the CI aggregation operators [12].
According to the MACBETH procedure, the DM is asked to provide
preferential ranking information on the criteria and the couples of
criteria including the strength of the preferences. Eq. (2) and the defini-
tions that will be given below will help us to build a system of equations
with the Shapley and the interaction parameters as variables.
In the situations where only one pi = 1 and all others are equal to
0 (i.e. only criterion i is satisfied), the aggregated performance expres-
sion will be as [12]:
i 1X n

parison of the technical performance of on-board hydrogen storage pAg ¼ φi − I ð3Þ

2 j¼1 ij
technologies [24], politics [27], supply chain management [6], and j≠i
earthquake risk mitigation [5]. The method requires only qualitative
judgments about differences of value to help an individual or a
group in quantifying the relative attractiveness of the elements of The aggregated performance expression of the situations where
a finite set A and to associate a real number φ(x) to each element x only one pi = 0 and all others are equal to 1 (i.e. all criteria except i
of A. is satisfied) will be as [12]:
Let X be the finite set of elements (alternatives) with at least two el-
ements and J the group of DMs who want to compare the relative attrac- 1X n
i
tiveness of these elements. Here, it is assumed that the DMs are able to pAg ¼ 1−φi − I ð4Þ
2 j¼1 ij
rank the elements of X either directly or through pairwise comparisons. j≠i
Each DM is first asked to provide a judgment about the relative attrac-
tiveness of two elements at a time to retrieve the ordinal judgment.
Then secondly, they are asked to provide a qualitative judgment on the The aggregated performance expression of the situations where
difference of attractiveness of those two elements, if they are not equally only two elementary performance expressions are equal to 1 (namely
attractive using the following linguistic terms: very weak (VW), weak i and j) and all others are equal to 0 (i.e. criterion i and j are satisfied)
(W), moderate (M), strong (S), very strong (VS) and extreme (E). will be as [16]:
MACBETH method presents a procedure to transform qualitative
preferences into coherent quantified elementary and aggregated per- !
i;j 1
formances. In order to solve the inter-criteria commensurability pAg ¼ φi þ φj − ∑ I ik þ ∑ Ijk ð5Þ
2 k∈ℵ 1;n jpk ¼0 k∈ℵ 1;n jpk ¼0
problem, it is sufficient to determine, for all interval scales, two com-
mon reference points namely the ‘good’ situation and the ‘neutral’ sit-
uation with the performance values 1 and 0, respectively.
Let us illustrate a demonstrative example of the integration of
Let pik be the performance expression of the k th alternative for cri-
MACBETH in CI. Consider a decision based on two criteria, namely
terion i. Suppose the DM prefers for the i th criterion the alternative k
C1 and C2, between two alternatives, namely A1 and A2. Suppose that
to the alternative l and in addition to that information, DM will char-
the DM has the following preferential rankings:
acterize the strength of his judgments with a level of strength that
can take values from 1 to 6 (from the least to the most strong level)
according to the six semantic categories of the difference of attrac-
C 1 &C 2 ≻ C 1 ≻ C 2 ≻ ≻ ″0″
M W VW S
tiveness explained above and 0 for a null strength. This level will be
C 1 →Good≻ A1 ≻ A2 ≻ Neutral
M M W
denoted with h. Therefore, if the DM prefers the alternative k to the
C 2 →Good≻ A2 ≻ A1 ≻ Neutral
S W VW
alternative l for the i th criterion, with a strength h, then the following
 T. Gürbüz et al. / Decision Support Systems 54 (2012) 206–214 209

The first preferential ranking results in the following system of In the use of CI method, if the experts (DMs) are able to provide
equations whose resolution gives the CI parameters: a fuzzy measure for their criteria set, then it is possible to define the
Shapley indices and interaction values using the Eqs. (B.1) and
(B.2). Or if, using their experience and expertise, they are able
1 to provide the weights of their criteria and information about
8
> pð1;1Þ −pð1;0Þ ¼ 3α ¼ 1−φ1 þ I12
2
>
interactions (being strongly positive, positive, null, negative or
>
>
< ð1;0Þ ð0;1Þ 1
−p ¼ 2α ¼ φ1 −φ2
>
p 1 2 1 strongly negative), then the interaction values can be retrieved
⇒α ¼ φ1 ¼ φ2 ¼ I ¼
ð0;1Þ ð0;0Þ 1 6 3 3 12 3
>p −p ¼ α ¼ φ2 − I12 using some other simplifying approaches, such as the one used in
>
>
2
>
Montignac et al. [24].
>
:
φ1 þ φ2 ¼ 1
In this case, only definition of individual performance values of the
The second and third preferential rankings give the following sys- alternatives according to those criteria has to be made. But, it has to
tems equations whose resolutions determine the performance values be noted that for the quantitative criteria, a simple normalization proce-
of alternatives for each criteria: dure will be required to obtain the values. On the other hand, in order to
define the performance values for the qualitative ones, a tool such as
1 AHP or MACBETH could be used. In this research, MACBETH is proposed
1−p1 ¼ 3α

f p11 −p21 ¼ 3α
p21 ¼ 2α
⇒α ¼
1
8
p11 ¼
5
8
p21 ¼
2
8
for that purpose.

8 3.2. Selection criteria

2
< 1−p2 ¼ 4α
>
1 1 1 2 3
2
p2 −p12 ¼ 2α ⇒α ¼ p2 ¼ p2 ¼
: 1 7 7 7
>
p2 ¼ α
Finally, using Eq. (1), the aggregated performance values are cal-
culated for A1 and A2 as 0.384 and 0.280, respectively. Hence, in this
example, A1 turns out to be a better alternative than A2.
3. ERP system evaluation framework
3.1. Evaluation procedure
The evaluation procedure of this study consists of seven steps as
follows Fig. 1:
1. Identify the ERP software selection/evaluation criteria that are
considered the most important for the users.
2. Once the model is built and the relations between criteria are
defined, decide the method to use. This is not an arbitrary choice.
3. If there is an outer-dependence between sub-criteria, then this is
something to be analyzed with ANP because of the simple fact that
CI cannot handle two elements that are connected to two different
points. In this case, two sub-criteria in question belong to two differ-
ent criteria. Hence, these dependencies will be handled with ANP.
4. Analyze sub-criteria of the same cluster in order to define the con-
junctive and disjunctive behavior between them. If there is such
relation, use CI in order to find the interaction values. In case of
no such interaction, handle the relations with ANP.
5. After handling the sub-criteria, take in consideration the upper
level, i.e. the criteria.
6. A preference ranking of the criteria given by the DMs will define the
conjunctive/disjunctive behavior between those. If there is not any
interaction of this kind between the criteria, then solve the model
with ANP. Make the final aggregation and obtain a ranking.
7. If there are conjunctive/disjunctive behavior between criteria, then
use the Shapley indices and the interaction values including the
weights of the sub-criteria and the alternatives' individual perfor-
mance values for each of those sub-criteria in order to perform
the final aggregation.
To resume this [17],
• Outer dependencies between sub-criteria+no conjunctive/disjunctive
behavior between them + NO conjunctive/disjunctive behavior
between criteria = ANP.
• No outer dependencies between sub-criteria + conjunctive/disjunctive
behavior between them + conjunctive/disjunctive behavior between
criteria= CI.
• Other than those situations = Hybrid ANP & CI.
Baki and Çakar [1] have summarized the ERP selection criteria in
their research after reviewing the related literature. We have used
the 16 criteria that they have proposed, but we have grouped them
under three main categories: vendor related, customer related and
software related (Table 2).
The problems that the firms face during the installation, the imple-
mentation and after the implementation periods can cause seven to
ten times the initial software cost. Hence, the firms need support and
service from the software firms in terms of IT expertise and domain
knowledge. One of the criteria that a buyer should consider is the
vendor's vision, specifically the modifications that the vendor plans to
make to its products and services over the next years. The reputation
of the vendor and its service infrastructure constitute its market posi-
tion. It is important that the software developer/vendor knows the in-
dustry of the customer. This characteristic is referred as the domain
knowledge of the vendor. The implementation of an ERP system to a
firm generally necessitates big changes in firms. Therefore, the software
vendor needs to introduce an effective methodology in order to elimi-
nate unnecessary activities. All these six criteria are gathered under
the title of vendor related criteria.
The ease of customization defines the ability of the ERP system vendor
to adopt its generic software solution to the company's specific needs. It is
important that the selected software can be implemented with current
organizational structure. Providing fitness with parent/allied organiza-
tions can affect both the decision process for some companies and
whole ERP project success. An ERP system consists of various modules.
The more an ERP system is better integrated, the more it becomes effec-
tive. These four criteria constitute customer related criteria.
Functionality is defined as the most important evaluation factor
[18]. The solution should have enough or even more modules related
to companies' core activities such as human resources, material man-
agement, project management, production planning, supply chain
management, etc. [7]. The selected ERP solution should address the
current trends in IT, which is referred as technical aspects. Setting re-
alistic expectations for the overall cost and attractive prices are essen-
tial in the buying process, as the ERP system cost is generally very
high.
The system reliability criterion consists of the answers to the ques-
tions such as “How long has the vendor been in the core ERP solution
business?” or “Have its current users been satisfied with the package?”.
In spite of the fact that there is not any single application that can do ev-
erything a company needs, the selected ERP solution has to be compat-
ible with the home-grown systems and other specialized software
products. The implementation time is closely related to the selected im-
plementation strategy. More customization requires more time and
hence, more cost.
210 T. Gürbüz et al. / Decision Support Systems 54 (2012) 206–214

Development of
BEGIN performance’s criteria &
sub-criteria

Outer Official
YES dependence between documents
sub-criteria +
Determination of the Literature
outer dependencies surveys
between sub-criteria NO +
DMs
Inner
NO dependence between
sub-criteria
Independent criteria
situation: Resolution YES
with an MCDM
method Determination of the
inner dependencies
between sub-criteria
Resolution of these END
dependencies and
determination of DMs
relative priorities of
the sub-criteria with Conjunctive or
NO disjunctive behavior
ANP
between sub-criteria

DMs
YES

Conjunctive or Resolution of these

YES disjunctive behavior dependencies and
between criteria determination of relative
priorities of the sub-criteria
with CI
NO
Resolution of these
interactions with CI Resolution of the
model with ANP
through the end.
Final aggregation in
order to find the
performance values

END

Fig. 1. Flowchart of the proposed methodology [17].

3.3. Proposed decision framework 4. Numerical application of the proposed framework

The proposed decision model consists of three levels. At the
highest level the objective of the problem (selecting the best ERP soft-
ware) is situated. In the second level, the criteria are listed. The low-
est level belongs to the alternatives. As alternatives, A1, A2, A3 and A4
are selected since they are in the same interval of price. Furthermore,
the interactions among criteria and sub-criteria are illustrated in
Fig. 2. In the proposed hierarchy, outer dependencies and inner de-
pendencies among sub-criteria as well as interaction between criteria
are assumed to be present.
4.1. Part 1: ANP
In the first part of the framework, pairwise comparison matrices
for all the sub-criteria have been prepared and filled out by the DM.
The consistency indexes of the matrices are all smaller than 0.10,
which proves their consistency [29]. The pairwise comparisons en-
able us to retrieve relative weights for the sub-criteria. The super-
matrix, which has the role of obtaining the composite weights, has
been constructed (Table 3).
 T. Gürbüz et al. / Decision Support Systems 54 (2012) 206–214 211

Table 2 As the next step, cluster/criteria comparison matrices have been

Selection criteria. prepared and filled out by the DM in order to normalize the super-
C1 Vendor related criteria (VRC) matrix (Table 3). Using the weights retrieved from these matrices,
C11 Support and service the cluster matrix is constructed and weighted supermatrix is calcu-
C12 Vision lated. The cluster matrix and the weighted supermatrix are represen-
C13 Market position
ted in Tables 4 and 5, respectively.
C14 Domain knowledge
C15 Reputation As the final step of ANP procedure, from the weighted supermatrix
C16 Methodology of software given in Table 5, the limit supermatrix has been calculated. The rela-
tive importance of the sub-criteria have been collected and then nor-
C2 Customer related criteria (CRC)
malized with respect to the criteria (Table 6).
C21 Ease of customization
C22 Better fit with organizational structure
C23 Fit with parent/allied organizational system
4.2. Part 2: CI
C24 Cross module integration
In order to find out criteria weights and conjunctive/disjunctive
C3 Software related criteria (SRC) interactions between criteria, a preferential ranking, including the
C31 Functionality
strength of the judgment, has been asked to the DM. The following
C32 Technical aspects
C33 Cost ranking has been obtained:
C34 System reliability
C35 Compatibility
C36 Implementation time M W M W S S
C 1 &C 3 ≻ C 1 &C 2 ≻ C 2 &C 3 ≻ C 1 ≻ C 3 ≻ C 2 ≻ ″0″

Using Eqs. (2)–(5), we have obtained the following equation system:

1
8
ð1;0;1Þ ð1;1;0Þ
>
> pAg −pAg ¼ 3α ¼ φ3 −φ2 − ½I12 −I13 Š
2
>
C1: Vendor
>
>
1
>
ð1;1;0Þ ð0;1;1Þ
pAg −pAg ¼ 2α ¼ φ1 −φ3 − ½I23 −I12 Š
>
>
>
related (VRC) 2
>
>
>
> ð0;1;1Þ ð1;0;0Þ
pAg −pAg
>
>
>
>
> ¼ 3α ¼ 1−2φ1
C11 C12 C13 1
<
ð1;0;0Þ ð0;0;1Þ
p −pAg ¼ 2α ¼ φ1 −φ3 − ½I12 −I23 Š
> Ag
> 2
>
ð0;0;1Þ ð0;1;0Þ 1
> pAg −pAg ¼ 4α ¼ φ3 −φ2 − ½I13 −I12 Š
>
C14 C15 C16
>
>
>
> 2
1
>
> ð0;1;0Þ ð0;0;0Þ
−p −
>
> p ¼ 4α ¼ φ ½ I þ I23 Š
Ag Ag 2
2 12
>
>
>
>
: pð1;1;1Þ ¼ 1 ¼ φ þ φ þ φ
>
Ag 1 2 3
C2: Customer C3: Software
related (CRC) related (SRC)
The resolution of this equation system has given the results shown
C21 C22 C31 C32 C33 in Table 7:
Table 7 shows that vendor related criteria is the most important
C23 C24 C34 C35 C36 cluster for the DM with a relative importance (φ1) of 0.4375, whereas
customer related criteria is the least important with a relative impor-
tance (φ2) of 0.2083. Furthermore, it is possible to state that an employ-
ee must be successful on both VRC and CRC in order to be considered
successful for the DM. The same situation is valid for CRC and SRC, as
Fig. 2. Interactions among criteria. the interaction values (I12 and I23) are positive for those couples of cri-

Table 3
Unweighted supermatrix.

S&S V MP DK REP MS EC BFOS FPAOS CMI F TA C SR CP IT

S&S 0 0 0.07 0 0.25 0 0.39 0.25 1 0 0 0 0.11 0 0 0.11

V 0.75 0 0.15 1 0.75 1 0.07 0 0 0 0.25 0 0 0.17 0.13 0
MP 0 0 0 0 0 0 0 0 0 0 0 0 0.64 0 0 0
DK 0.25 0 0.39 0 0 0 0.39 0.75 0 0 0.75 0 0 0.83 0.88 0.26
REP 0 0 0.39 0 0 0 0 0 0 0 0 0 0.26 0 0 0
MS 0 0 0 0 0 0 0.15 0 0 0 0 0 0 0 0 0.64
EC 0.64 0 0.43 0 0 0 0 0.43 0.25 1 0.08 0.19 0 0 0.15 0.38
BFOS 0.11 0 0.14 0 0 0 0.09 0 0 0 0.20 0.08 0 0 0.39 0.13
FPAOS 0 0 0 0 0 0 0.46 0.43 0 0 0.20 0.08 0 0.75 0.39 0.38
CMI 0.26 0 0.43 0 0 0 0.46 0.14 0.75 0 0.52 0.66 0 0.25 0.07 0.13
F 0 0 0.05 0 0 0 0.25 0 0 0 0 0 0.19 0.25 0 0.26
TA 0 0 0 0 0 0 0.75 0 0 1 0 0 0.07 0.75 1 0.11
C 0 0 0.48 0 0 0 0 0 0 0 0.73 0.64 0 0 0 0
SR 1 0 0.21 0 1 0 0 0 0 0 0 0.26 0.19 0 0 0
CP 0 0 0.05 0 0 0 0 0 1 0 0.18 0.11 0.07 0 0 0.64
IT 0 0 0.21 0 0 0 0 0 0 0 0.08 0 0.47 0 0 0
212 T. Gürbüz et al. / Decision Support Systems 54 (2012) 206–214

Table 4 Table 6
Cluster matrix. Sub-criteria weights.

VRC CRC SRC VRC S&S V MP DK REP MS

VRC 0.637 0.258 0.258 w 0.138 0.473 0.065 0.235 0.045 0.044
CRC 0.105 0.637 0.105 CRC EC BFOS FPAOS CMI
SRC 0.258 0.105 0.637
w 0.422 0.049 0.202 0.327
SRC F TA C SR CP IT

terion. On the other hand, there is no conjunctive/disjunctive interac- w 0.099 0.276 0.238 0.159 0.126 0.102
tion between VRC and SRC as the interaction value for these (I13) is null.

4.3. Part 3: relative performance scores of alternatives

For each sub-criterion, the DM has been asked to give us a prefer-

4.4. Part 4: final aggregation
ential ranking, including the strength of the comparisons. For VRC,
the following ranking has been obtained.
As the last step of the proposed procedure, the scores of the alter-
natives for each criterion have been calculated (Table 9).
C 11 →Good≻VW A4 ≻M A1 ≻M A2 ≻M A3 ≻M Neutral In order to determine the final performance scores of four alterna-
C 12 →Good≻ A1 ≻ A2 ≻ A3 ≻ A4 ≻ Neutral
VW W M S S
tives (Table 10), we have used the values in Tables 7 and 9 and
C 13 →Good≻ A4 ≻ A1 ≻ A2 ≻ A3 ≻ Neutral
VW W M S S
Eq. (1).
C 14 →Good≻ A1 ≻ A2 ≻ A4 ≻ A3 ≻ Neutral
VW W M M S

C 15 →Good≻ A1 ≻ A2 ≻ A4 ≻ A3 ≻ Neutral
VW W M S S 5. Discussion
C 16 →Good≻ A1 ≻ A2 ≻ A3 ≻ A4 ≻ Neutral
VW W M S S

The result indicates that the final performance score of alternative

For CRC, the following ranking has been obtained.
C 21 →Good≻W A4 ≻M A3 ≻M A2 ≻M A1 ≻M Neutral
C 22 →Good≻M A2 ≻M A1 ≻S A3 ≻M A4 ≻M Neutral
C 23 →Good≻ A2 ≻ A1 ≻ A3 ≻ A4 ≻ Neutral
M M S M M the difference between the performance values of those two with re-
C 24 →Good≻ A3 ≻ A1 ≻ A2 ≻ A4 ≻ Neutral
M W W S M
For SRC, the following ranking has been obtained.
C 31 →Good≻VW A2 ≻M A1 ≻M A3 ≻M A4 ≻S Neutral
C 32 →Good≻W A1 ≻W A2 ≻M A3 ≻M A4 ≻S Neutral
C 33 →Good≻M A4 ≻S A3 ≻S A2 ≻M A1 ≻VW Neutral
C 34 →Good≻W A1 ≻W A2 ≻S A3 ≻M A4 ≻S Neutral
C 35 →Good≻M A1 ≻W A2 ≻S A3 ≻M A4 ≻M Neutral
C 36 →Good≻M A4 ≻S A3 ≻S A2 ≻W A1 ≻W Neutral
The equation systems have been retrieved from these three pref-
erence rankings, using Eq. (2). The resolution of these equation sys-
tems are summarized in Table 8.
A1 is the highest (0.6867) and that of alternative A4 is the lowest
(0.4705). The fact that VRC has the greatest relative importance
(Φ1 = 0.4375) has played an important role for A1 and A2 to be ranked
first two in the final ranking and for A3 and A4 to be ranked last two.
Although A2 has greater performance values for CRC and SRC (albeit
spect to SRC is significantly low), VRC was the defining criteria for A1
to be ranked first. The same situation is present between A3 and A4: A4
has lower performance values with respect to CRC and SRC and great-
er performance value with respect to VRC. However in this case, A3 is
ranked before A4. The reason is the fact that the differences between
performance values with respect to CRC and SRC for those two alter-
natives are greater than that for A1 and A2.
If we were to ignore the interactions and solve the model with the
expert judgments given to us, for our numerical application, the final
ranking would be A4 ≻A1 ≻A2 ≻A3 with the respective relative weights
0.614, 0.596, 0.584 and 0.446. As it can be observed, the ranking has
dramatically changed. The worst performing alternative has become
the best performing one. This is not the proof that the result will be
different in each and every case or the changes will be as dramatic
as this example. However, this demonstrates the fact that the final
ranking may change with the ignorance of the interactions and there-
fore it may lead us to erroneous conclusions and decisions.

Table 5
Weighted supermatrix.

S&S V MP DK REP MS EC BFOS FPAOS CMI F TA C SR CP IT

S&S 0 0 0.043 0 0.178 0 0.101 0.072 0.258 0 0 0 0.030 0 0 0.027

V 0.478 0 0.097 1 0.534 1 0.018 0 0 0 0.07 0 0 0.043 0.033 0
MP 0 0 0 0 0 0 0 0 0 0 0 0 0.184 0 0 0
DK 0.159 0 0.248 0 0 0 0.101 0.216 0 0 0.19 0 0 0.215 0.226 0.067
REP 0 0 0.248 0 0 0 0 0 0 0 0 0 0.075 0 0 0
MS 0 0 0 0 0 0 0.039 0 0 0 0 0 0 0 0 0.165
EC 0.067 0 0.045 0 0 0 0 0.305 0.159 0.87 0.01 0.027 0 0 0.016 0.039
BFOS 0.011 0 0.015 0 0 0 0.058 0 0 0 0.02 0.011 0 0 0.041 0.013
FPAOS 0 0 0 0 0 0 0.290 0.305 0 0 0.02 0.011 0 0.079 0.041 0.039
CMI 0.027 0 0.045 0 0 0 0.290 0.102 0.478 0 0.05 0.093 0 0.026 0.007 0.013
F 0 0 0.013 0 0 0 0.026 0 0 0 0 0 0.138 0.159 0 0.165
TA 0 0 0 0 0 0 0.079 0 0 0.14 0 0 0.052 0.478 0.637 0.067
C 0 0 0.124 0 0 0 0 0 0 0 0.47 0.547 0 0 0 0
SR 0.258 0 0.055 0 0.289 0 0 0 0 0 0 0.222 0.138 0 0 0
CP 0 0 0.013 0 0 0 0 0 0.105 0 0.12 0.090 0.052 0 0 0.406
IT 0 0 0.055 0 0 0 0 0 0 0 0.05 0 0.332 0 0 0
 T. Gürbüz et al. / Decision Support Systems 54 (2012) 206–214 213

Table 7 Table 9
Choquet integral parameters. Performance values of alternatives with respect to criteria.

Parameter Φ1 Φ2 Φ3 I12 I13 I23 VRC CRC SRC

Value 0.4375 0.2083 0.3542 0.0417 0 0.0417 A1 0.8854 0.5042 0.5754

A2 0.7232 0.5950 0.5803
A3 0.4310 0.5760 0.4934
A4 0.4877 0.4788 0.4467

6. Conclusion Table 10
Final scores of alternatives.
ERP systems have a vital role in today's organizations, however A1 A2 A3 A4
they have high costs and high implementation risks. Therefore, eval-
Score 0.6867 0.6429 0.4786 0.4705
uating the offered ERP systems and selecting the most suitable one
among them is a challenging task. With the increase in the number
of requirements, the selection problem becomes more complex to Appendix B
solve, and the companies need to make their decisions in increasingly
complex environments. Shapley indices and interactions for the criteria can be calculated
Our work presents a comprehensive framework for selecting a using fuzzy measures with the following formula:
suitable ERP system based on a hybrid multicriteria decision analysis
process. The procedure consists of three methodologies: Analytic Net- ðn−jK j−1Þ!jK j!
work Process (ANP), Choquet Integral (CI) and Measuring Attractive- φi ¼ ∑ ½μ ðK∪figÞ−μ ðK ފ ðB:1Þ
K⊂X 5 fig n!
ness by a Categorical Based Evaluation Technique (MACBETH). We
have illustrated the applicability of the framework through a case
study of the ERP software selection of a company. The proposed deci- ðn−jK j−2Þ!jK j!
Iij ¼ ∑ ½μ ðK∪fi; jgÞ−μ ðK∪figÞ−μ ðK∪fjgÞ þ μ ðK ފ
sion making framework is flexible enough to fit other sectors with K⊂N 5 fi;jg ðn−1Þ!
some specific characteristic changes and to incorporate different ðB:2Þ
criteria in the evaluation process.
Besides proposing a hybrid MCDM model for ERP selection prob-
lem, what we would also like to underline with this study is the im-
References
portance of not ignoring the interactions among different elements
of a decision model. Moreover as it is discussed in the previous sec- [1] B. Baki, K. Çakar, Determining the ERP package-selecting criteria: the case of
tion, we can say that ignoring the interactions may lead us to obtain Turkish manufacturing companies, Business Process Management Journal 11
(1) (2005) 75–86.
a different final ranking, hence making a bad decision according to
[2] C.A. Bana e Costa, Structuration, Construction et Exploitation d'un Modèle à
the wrong conclusions. Multi-critère d'aide à la Décision, PhD Thesis, Technical University of Lisbon, Lis-
bon, 1992.
[3] C.A. Bana e Costa, M.P. Chagas, A career choice problem: an example of how to use
MACBETH to build a quantitative value model based on qualitative value judg-
Appendix A ments, European Journal of Operational Research 153 (2004) 323–331.
[4] C.A. Bana e Costa, J.M. De Corte, J.C. Vansnick, On the mathematical foundations of
Let μ be a non-monotonic fuzzy measure on X and f a function on X MACBETH, Working Paper LSEOR, 03, London School of Economics and Political
Science, 2004, p. 56.
with range {a1,a2,…,an,} where a1 ≤ a2 ≤ … ≤ an. The CI (C) ∫ f (x)dμ(x)
[5] C.A. Bana e Costa, C.S. Oliveira, V. Vieira, Prioritization of bridges and tunnels in
or simply (C) ∫ fdμ of f with respect to μ is defined as follows: earthquake risk mitigation using multicriteria decision analysis: application to
Lisbon, Omega 36 (2008) 442–450.
[6] L. Berrah, G. Mauris, J. Montmain, Monitoring the improvement of an overall in-
n dustrial performance based on a Choquet integral aggregation, Omega 36
ðai −ai−1 Þ:μ ðfxjf ðxÞ≥ai gÞwhere a0 ¼ 0
X
ðC Þ∫f dμ ¼ ðA:1Þ (2008) 340–351.
i¼1
[7] G. Brewer, On the road to successful ERP, Instrumentation & Control Systems 73
(5) (2000) 49–58.
[8] G. Campanella, R.A. Ribeiro, A framework for dynamic multiple-criteria decision
making, Decision Support Systems 52 (2011) 52–60.
Since a fuzzy measure defined on a set of n elements requires 2n [9] U. Cebeci, Fuzzy AHP-based decision support system for selecting ERP systems in
textile industry by using balanced scorecard, Expert Systems with Applications 36
real coefficients for its definition, k-additive measures have been in-
(2009) 8900–8909.
troduced by Grabisch [15] in order to decrease the exponential com- [10] E.W.L. Cheng, H. Li, L. Yu, The Analytic Network Process (ANP) approach to loca-
plexity of fuzzy measures in practical applications. In a k-additive CI tion selection: a shopping mall illustration, Construction Innovation 5 (2004)
83–97.
the interactions of higher order than k will not be taken in consider-
[11] G. Choquet, Theory of capacities, Annales de l'Institut Fourier 5 (1953) 131–295.
ation or they will be considered null. Therefore, k-additive measures [12] V. Clivillé, L. Berrah, G. Mauris, Quantitative expression and aggregation of perfor-
k mance measurements based on MACBETH multicriteria method, International
can be represented by a limited set of coefficients, at most C ni
P
Journal of Production Economics 105 (1) (2007) 171–189.
i¼1 [13] T.H. Davenport, Putting the enterprise into the enterprise system, Harvard Busi-
coefficients. ness Review 76 (4) (1998) 121–131.

Table 8
Performance values of alternatives with respect to sub-criteria.

C11 C12 C13 C14 C15 C16 C21 C22 C23 C24 C31 C32 C33 C34 C35 C36

A1 0.692 0.929 0.786 0.923 0.929 0.929 0.214 0.625 0.625 0.786 0.714 0.857 0.067 0.867 0.800 0.133
A2 0.462 0.786 0.571 0.769 0.786 0.786 0.429 0.813 0.813 0.643 0.929 0.714 0.267 0.733 0.667 0.267
A3 0.231 0.571 0.286 0.308 0.286 0.571 0.643 0.563 0.563 0.500 0.500 0.500 0.533 0.467 0.400 0.533
A4 0.923 0.286 0.929 0.538 0.571 0.286 0.857 0.188 0.188 0.214 0.286 0.286 0.800 0.267 0.200 0.800
214 T. Gürbüz et al. / Decision Support Systems 54 (2012) 206–214
[14] Y.C. Erensal, E. Albayrak, Successful adoption of macroergonomics in manufactur-
ing: using a multicriteria decision-making methodology-analytic hierarchy pro-
cess, Human Factors and Ergonomics in Manufacturing 14 (4) (2005) 353–377.
[15] M. Grabisch, k-order additive fuzzy measures, 6th International Conference on In-
formation Progressing and Management of Uncertainty, Fuzziness and Knowl-
edge Based Systems 5 (1997) 587–607.
[16] T. Gürbüz, Multiple criteria human performance evaluation using Choquet inte-
gral, International Journal of Computational Intelligence Systems 3 (3) (2010)
290–300.
[17] T. Gürbüz, Multicriteria Decision Making Approach to Human Resources Perfor-
mance Evaluation. Ph.D. Thesis, Galatasaray University, Turkey, 2010.
[18] B. Hecht, Choose the right ERP software, Datamation 43 (3) (1997) 56–58.
[19] C.W. Holsapple, M.P. Sena, ERP plans and decision-support benefits, Decision
Support Systems 38 (2005) 575–590.
[20] G. Işıklar Alptekin, G. Büyüközkan, An integrated case-based reasoning and
MCDM system for web based tourism destination planning, Expert Systems
with Applications 38 (3) (2011) 2125–2132.
[21] U. Jung, D.W. Seo, An ANP approach for R&D project evaluation based on interde-
pendencies between research objectives and evaluation criteria, Decision Support
Systems 49 (2010) 335–342.
[22] E.E. Karsak, C.O. Ozogul, An integrated decision making approach for ERP system
selection, Expert Systems with Applications 36 (2009) 660–667.
[23] C. Labreuche, M. Grabisch, The Choquet integral for the aggregation of interval
scales in multicriteria decision making, Fuzzy Sets and Systems 137 (1) (2003)
11–26.
[24] F. Montignac, I. Noirot, S. Chaudourne, Multi-criteria evaluation of on-board hy-
drogen storage technologies using the MACBETH approach, International Journal
of Hydrogen Energy 34 (10) (2009) 4561–4568.
[25] T. Murofushi, M. Sugeno, An interpretation of fuzzy measure and the Choquet in-
tegral as an integral with respect to a fuzzy measure, Fuzzy Sets Systems 29
(1989) 201–227.
[26] J.C. Pomerol, S. Barba Romero, Multicriterion Decision in Management: Principles
and Practice, 1st edition Kluwer Academic Publishers, Norwell, 2000.
[27] M. Roubens, A. Rusinowska, H. de Swart, Using MACBETH to determine utilities of
governments to parties in coalition formation, European Journal of Operational
Research 172 (2006) 588–603.
[28] I. Saad, S. Hammadi, M. Benrejeb, P. Borne, Choquet integral for criteria aggrega-
tion in the flexible job-shop scheduling problems, Mathematics and Computers in
Simulation 76 (5–6) (2008) 447–462.
[29] T.L. Saaty, The Analytic Hierarchy Process, McGraw-Hill, New York, 1980.
[30] T.L. Saaty, Decision Making with Dependence and Feedback: The Analytic Net-
work Process, RWS Publications, Pittsburgh, 1996.
[31] R.W. Saaty, Decision Making in Complex Environments, Creative Decisions Foun-
dation, Pittsburgh, 2003.
[32] M.J. Verdecho, J.J. Alfaro-Saiz, R. Rodriguez-Rodriguez, Prioritization and manage-
ment of inter-enterprise collaborative performance, Decision Support Systems 53
(1) (2012) 142–153.
[33] C.C. Wei, C.F. Chien, M.J.J. Wang, An AHP-based approach to ERP system selection,
International Journal of Production Economics 96 (2005) 47–62.
[34] H.R. Yazgan, S. Boran, K. Göztepe, An ERP software selection process with using
artificial neural network based on Analytic Network Process approach, Expert
Systems with Applications 36 (2009) 9214–9222.
Tuncay Gürbüz, completed his B.Sc. on Industrial Engineering in the Industrial Engi-
neering Department of Galatasaray University in 2000, his M.Sc. and Ph.D. on Industrial
Engineering in the Institute of Science and Engineering of Galatasaray University in
2003 and 2010 respectively. His research interests and focus are in the areas of multi-
ple criteria decision making, decision support systems, fuzzy logic and performance
evaluation. He is currently a research assistant in Industrial Engineering Department
of Galatasaray University.
S. Emre Alptekin received his B.Sc. degree in Industrial Engineering in1999 from
Istanbul Technical University, his M.Sc. and Ph.D. degrees in Industrial Engineering
from Galatasaray University in 2001 and 2006, respectively. He is currently an assis-
tant professor in Galatasaray University. His research interests contain multi-criteria
and expert decision analysis systems.
Gülfem Işıklar Alptekin received her B.Sc. degree in Computer Science in 2001, her
M.Sc. degree in Industrial Engineering from Galatasaray University in 2003, and her
Ph.D. degree in Computer Engineering from Boğaziçi University in 2010. She is current-
ly an assistant professor in Galatasaray University. Her research interests contain pric-
ing and resource allocation models in next generation wireless networks and
multi-criteria decision support systems in real life applications.