INTERNATIONAL JOURNAL OF COMPUTERS COMMUNICATIONS & CONTROL
ISSN 1841-9836, 10(2):238-247, April, 2015.

Multicriteria Supplier Classification for DSS:
Comparative Analysis of Two Methods

J.M. Sepulveda, I.S. Derpich

Juan M. Sepulveda*, Ivan S. Derpich
University of Santiago
3769 Ecuador Ave., Santiago, Chile, CP 7254758
*Corresponding author: juan.sepulveda@usach.cl
ivan.derpich@usach.cl

Abstract: In this paper the analysis of two multicriteria decision making (MCDM)
methods for sorting suppliers in industrial environments is presented. The MCDM
methods correspond to Electre and FlowSort and both are applied to the classification
of providers in an actual case of the local softdrink bottling industry in Chile. The
results show that Electre as an outranking method it may well classify suppliers in a
similar manner as FlowSort does. Nevertheless, due to the intrinsic underlying fuzzy
multicriteria nature of the problem, FlowSort is found to be more suitable method for
building a rule-based system based on preference functions for automating the process
of suppliers clustering when developing strategies of relationship management in the
sense of the Kraljic categories in supply chain management.
Keywords: Decision Support Systems, Supply Management, Electre, FlowSort.

1 Introduction

Supplier management is crucial in order to improve the benefits that a company can have
at the operational, functional, economic, and financial levels and in terms of the supply chain
it is embedded. Supplier evaluation is a regular process of operational management in many
companies; the need arises from finding suppliers for new products, parts, or materials, or for
assessing performance of the current supplier base in order to decide continuation of their services.
However, in supplier management a more strategic task is classifying suppliers into categories
for deciding the relational approach to be followed. Kraljic [1] in his pioneering work established
categories of suppliers according to the economic impact for the purchasing company and the
risks the suppliers may experiment in their respective market. As recently reviewed by Monczka
et. al. [2] Kraljic’s categories are defined by a matrix of four quadrants, namely: routine,
leverage, bottleneck and critical quadrant. Each category means different strategies ranging
from automated transactions by the use of ERP and/or EDI of commodity items of low total
purchasing expenses, i.e. routine category, up to the critical category in which for example the
expenses are very high and the suppliers situation in its market is risky due to uncertainties,
uniqueness or high dependence. See Figure 1 for an example of the Kraljic matrix. The number
in parenthesis is a level in a scale from 1 to 4, as explained below.

Level 1: Many alternative products and processes; abundant sources of supply; low value-
small individual transactions; routinary use, unspecified items; no specialized purchasing knowl-
edge required.

Level 2: Complex specifications requiring complex manufacturing or service processes; few
alternate production/sources of supply; big impact on operations / maintenance; new technology
or untested processes.

Level 3: High expenditures, commodity items; large marketplace capacity, big inventories;
many alternate products and services; many qualified providers; market/price sensitive demand.

Level 4: Critical to profitability and operations; few qualified sources of supply; large expen-
ditures; design and quality critical; complex and/or rigid specifications.

Multicriteria Supplier Classification for DSS:
Comparative Analysis of Two Methods 239

Figure 1: Kraljic Matrix 1 for strategic positioning

In this work the classification or categorization of suppliers in the case study, defines three
categories: (a) transactional, (b) collaborative, and (c) integrated. These categories are specific
for the company dealt with; however the method is general enough as to be applied to other
similar situations.

A transactional supplier is understood as the profile of routine suppliers combining low pur-
chasing expenses, easy-to-obtain commodity-type products or services, and with low mutual
attraction in the relationship. A collaborative supplier is one of importance to the company due
to high purchasing expenses. For this reason, close monitoring of performance is needed in order
to assure the service level agreements and to control the costs. The logistics complexity is of
medium range as well as the mutual attraction. An integrated supplier is a critical one, of such
an importance that a strategic alliance is needed, such as the case of a third party fully man-
ufacturing a component or a product for a given customer company’s brand (OEM strategy),
as it occurs in the food, pharmaceutical, or car industries. It combines criticality in product
positioning with high logistics complexity and high mutual attraction of the relationship.

Clearly, the three categories above can be understood as fuzzy sets where the implementation
process of the method allows the identification of strengths and weaknesses of using formalized
supplier selection models to tackle the supplier sorting problem. This highlights potential barriers
preventing the adoption of these types of methods. For this purpose, the paper presents a number
of relevant issues arising from the application and managerial implications for both customer and
suppliers, concerning differentiated management practices according to their classification.

Besides Matrix 1, two other matrices are used to classify suppliers. Matrix 2 values the
complexity of logistics, and Matrix 3 the mutual attraction of the SRM (supplier relationship
management).

In Matrix 2 (Lead Time and Stock Rotation) the lowest level of logistics complexity is for
supplies of fast replenishment and slow moving items. It the follows the case of suppliers of items
with short lead time (hours, days, within a week) but of high consumption or inventory turnover.
The third level is for slow moving items and long lead times (e.g. imported products); finally,
the most complex logistics is for suppliers of items with high rotation and long lead time due the
risk of inventory breakdown.

The numbers assigned as well as the criteria are depending on the specific application and are

240 J.M. Sepulveda, I.S. Derpich

Figure 2: Matrices 2 and 3 for logistics complexity and SRM attraction level

provided by experienced purchasing executives, without loss of generality. In Matrix 3 (Supplier
Relationship Attraction) the lowest level is assigned when both parties have little mutual interest.
It follows when the purchaser is important to the supplier (level 2). For the specific case of the
application, the managers gave the highest importance to suppliers with superior power in the
relationship (generally due to size).

2 Strategies by category

With the suppliers being categorized, the organization can apply differentiated approaches
to SRM. Next, a brief description of each category is given.

2.1 Transactional category

The transactional category has as main characteristics the existence of many alternative prod-
ucts and services, many sources of supply, low value small individual transactions, commodity-
type items, requiring little or no specialization of the purchasing executive. Hence, this category
is non-critical and besides the logistics aspects are simple because lead times are short and ro-
tation is low. The mutual relationship is of low level. The general strategy for this category is
the simplification of the acquisition process. The tactics relate to increase the role of computer
based systems and reducing the buying effort. Specific actions include rationalization of the
supplier base, the automation of the purchasing process by the use of automated requisitioning,
electronic data interchange, stockless procurement, minimization of administrative costs. Little
negotiation is needed.

Multicriteria Supplier Classification for DSS:
Comparative Analysis of Two Methods 241

2.2 Collaborative category

The collaborative category has as main characteristics the existence of many alternative
products and services, many sources of supply, commodity-type items, requiring little or no
specialization of the purchasing executive. However, the purchasing expenses and volumes are
higher. Hence, the optimization of operations serve as leverage for obtaining commercial advan-
tage. Logistics may be more complex due to longer lead times and higher rotation of items. The
type of relationship is of purchaser’s power. The general strategy for this category is maximizing
the commercial advantage. The tactics involve concentration of purchasing to increase power
and also to maintain competition. Specific actions involve the promotion of competitive bidding,
procurement coordination, adoption of industry standards to reduce dependency.

2.3 Integrated category

This category has as main characteristics a heavy control of the providers performance,
partnerships in the form of a strategic outsourcing, cooperation to optimize mutual benefits.
For instance, a strategic supplier performing outsourced implant critical tasks of production
and operations. The complexity of operations demands close control and information exchange.
The importance of the purchaser and the criticality of the supplier calls for relations of mutual
benefits. The attraction force of both parties is high.

3 Basics of the two methods

The ELECTRE methodology is based on the concordance and discordance indices. The
simplest method of the ELECTRE family is ELECTRE-I. See Figueira et al. [3].

This method starts from the data in the decision matrix, and assumes that the sum of the
weights of all criteria is equal to one. For an ordered pair of alternatives (Aj, Ak), the concordance
index cjk is the sum of all the weights for those criteria where the performance score of Aj is
least as high as that of Ak, as in formula (1). It can be seen that cjk lies between 0 and 1.

cjk =
∑

i:aij≥aik

wi j, k = 1, .., n, j ̸= k (1)

The computation of the discordance index djk is more elaborated: djk = 0 if aij > aik ,
i = 1, ..., m, that is, the discordance index is zero if Aj performs better than Ak on all of the
criteria. Otherwise, djk is calculated as in (2). That is, for each criterion where Ak outperforms
Aj, the ratio is calculated as the difference in performance level between Ak and Aj and the
maximum difference in score on the criterion concerned between any pair of alternatives.

djk = maxi=1,..,m
aik − aij

maxj=1,..,naij − minj=1,...,naij
j, k = 1, .., n, j ̸= k (2)

The maximum of these ratios (which must lie between 0 and 1) is the discordance index. A
concordance threshold c∗ and discordance threshold d∗ are then defined such that 0 < d∗ < c∗ <
1. Then, Aj outranks Ak if the cjk > c∗ and djk < d∗, i.e. the concordance index is above and
the discordance index is below its threshold, respectively.

FlowSort is based on the Promethee ranking methodology (Figueira et al., 2005) [3], this
new sorting method by Nemery and Lamboray [4] is utilized for assigning actions to completely
ordered categories; these categories are defined either by limiting profiles (i.e., min and max
values) or by central profiles (or centroids). This method has also been applied by (Sepulveda

242 J.M. Sepulveda, I.S. Derpich

et al., 2010) [5] in another management decision problem in the innovation field for diagnosing
capabilities in small enterprises.

In what follows, limiting profiles will be used. The assignment of an action (i.e., an object to
be sorted) into a category is based on the relative position of this action with respect to the defined
reference profiles in terms or incoming or outgoing net flows. Let A = (a1, a2, ..., an) be the set of
n actions or alternatives to be sorted. These actions are evaluated on q criteria gj(j = 1, ..., q);
all criteria are supposed to be maximized in the decision making problem. The categories to
which the actions must be assigned are denoted by C1, C2, ..., Ck. Let R = (ri, ..., rK+1) be the
set of limiting profiles in the case when a category is defined by an upper and lower limit. Let
π(x, y) be the preference of an action x over an action y, as in the Promethee method.

Figure 3 shows typical shapes of preference functions where the x-axis is the degree of differ-
ence between actions x and y. Thus, the positive, negative and net flows ϕ of each action x in
R, are computed by equations (4) (5) (6) where Ṙi = R U {ai} is the extended set of profiles
either for the limiting profile case or the central profiles. The rules for assigning actions ai to a
category Ch are given by equations (7) and (8) in the case of limiting profiles.

π(x, y) =

q∑
j=1

wjP(x, y) (3)

ϕ+
Ṙi

=
1∣∣∣Ṙi∣∣∣ − 1

∑
y∈Ṙi

π(x, y) (4)

ϕ−
Ṙi

=
1∣∣∣Ṙi∣∣∣ − 1

∑
y∈Ṙi

−

π(x, y) (5)

ϕ
Ṙi

= ϕ+
Ṙi

− ϕ−
Ṙi

(6)

Cϕ+(ai) = Ch, if ϕ
+
Ṙi
(rh) ≥ ϕ+Ṙi(a1) > ϕ

+
Ṙi

(7)

Cϕ−(ai) = Ch, if ϕ
−
Ṙi
(rh) < ϕ

−
Ṙi
(a1) ≤ ϕ−

Ṙi
(rh+1) (8)

4 Results of the two methods

First, the sorting will be made by the Electre method. Table 1 shows the score combination
for six randomly chosen suppliers and the resulting category according to the values predefined
for the combination by using Electre. The six cases are for illustrative purposes and show the
type of results without loss of generality.

Table 2 shows the limiting profiles for the FlowSort method.
The scores for each criteria in FlowSort are the same as the combination values in Table 1.
Table 3 shows the results for FlowSort. In the table, the six extended sets R correspond to

suppliers A,B,C,D,E,F, respectively. The category is obtained by applying the rules defined by
equations (7) and (8). It can be observed that for the six suppliers, with the exception of C, the
assigned categories are the same.

This is encouraging since the basis of each method is very different. It can be said that in
case C this provider was better classified by FlowSort. The difference obtained in one of the
suppliers (Supplier C), is mainly because of the assessment data that show that this provider

Multicriteria Supplier Classification for DSS:
Comparative Analysis of Two Methods 243

Figure 3: Types of Preference functions

Table 1: Results by Electre
Supplier Score combination Category

A "4-3-3" Integrated
B "3-2-2" Collaborative
C "1-2-3" Transactional
D "4-3-4" Integrated
E "4-3-4" Integrated
F "4-3-2" Integrated

Table 2: Limiting profiles chosen for FlowSort
Profile C1 C2 C3

r1 4.5 4.5 4.5
r2 3.0 3.0 3.0
r3 1.5 3.0 3.0
r4 0 0 0

244 J.M. Sepulveda, I.S. Derpich

Figure 4: Graphical Representation of FlowSort Results for Two Suppliers

Multicriteria Supplier Classification for DSS:
Comparative Analysis of Two Methods 245

Table 3: Results for FlowSort
Flow Sort Results r1 r2 r3 r4 A(i) Category

ϕ+ 1.25 0.75 0.5 0 0.83 Integrated
Ṙ1 ϕ− 0 0.33 0.75 1 0.25 Integrated

ϕnet 1.25 0.42 -0.25 -1 0.58 Integrated
ϕ+ 1.25 0.92 0.5 0 0.75 Collaborative

Ṙ2 ϕ− 0 0.25 0.75 1 0.42 Collaborative
ϕnet 1.25 0.67 -0.25 -1 0.33 Collaborative
ϕ+ 1.25 0.92 0.58 0 0.67 Collaborative

Ṙ3 ϕ− 0 0.25 0.67 1 0.5 Collaborative
ϕnet 1.25 0.67 -0.08 -1 0.17 Collaborative
ϕ+ 1.25 0.75 0.5 0 0.92 Integrated

Ṙ4 ϕ− 0 0.42 0.75 1 0.25 Integrated
ϕnet 1.25 0.33 -0.25 -1 0.67 Integrated
ϕ+ 1.25 0.75 0.5 0 0.92 Integrated

Ṙ5 ϕ− 0 0.42 0.75 1 0.25 Integrated
ϕnet 1.25 0.33 -0.25 -1 0.67 Integrated
ϕ+ 1.25 0.83 0.5 0 0.83 Integrated

Ṙ6 ϕ− 0 0.33 0.75 1 0.33 Integrated
ϕnet 1.25 0.5 -0.25 -1 0.5 Integrated

has some lower scores in some criteria (matrix- criterion 1) and high in other ones ( 2 and 3 )
and the final class depends on the decision maker.

However, note that FlowSort is a method that reflects the in and out flows generated by
the alternatives, while the classification made in the Electre is arbitrary at some extent by the
threshold values chosen for each category. Nevertheless, both methods solve the problem and are
suitable for automating the classification process as part of a decision support system for supply
management.

Figure 4 shows a graphical representation of the sorting for Suppliers 1 and 2, as given by
Table 3.

In Table 1 the category is chosen according to the values in Table 4. This shows the results
of the Electre method for the 64 possible combinations of the three matrices with the aggregated
dominance index (concordance minus discordance). The score values of Table 4 in columns Score
Matrix 1,2,3 (indicated as SM1, SM2, SM3) for each supplier score combination (SC) were given
by experienced supply managers same as in the FlowSort method. Current work is addressing
the obtention of these numbers from the key performance indicators in the database of a business
intelligence module connected to the ERP system.

The categories are obtained by sorting in increasing order of the aggregated dominance index
(A.D.I.) and by defining thresholds for each category. The weights were chosen as 1/3 each
matrix (w.l.g). For instance, as shown in Table 1, Supplier A with scores "4-3-3" corresponds
to the combination "a60"in the Appendix, Supplier B with scores "3-2-2" corresponds to "a35",
and so on. In this case, the same cases were used for both the Electre and FlowSort methods.

246 J.M. Sepulveda, I.S. Derpich

Table 4: Categories by Electre
SC SM1 SM2 SM3 A.D.I. Category SC SM1 SM2 SM3 A.D.I. Category
a1 1 1 1 0 Transactional a34 3 1 4 17 Collab.
a2 1 1 2 1 Transactional a37 4 1 3 17 Collab.
a3 1 2 1 1 Transactional a38 4 3 1 17 Collab.
a4 2 1 1 1 Transactional a30 1 3 4 18 Collab.
a5 1 1 3 2 Transactional a31 1 4 3 18 Collab.
a6 1 3 1 2 Transactional a36 3 4 1 18 Collab.
a7 3 1 1 2 Transactional a40 2 2 4 20 Integrated
a8 1 1 4 3 Transactional a43 4 2 2 20 Integrated
a9 1 2 2 3 Transactional a41 2 4 2 21 Integrated
a10 1 4 1 3 Transactional a46 3 2 3 21 Integrated
a11 2 1 2 3 Transactional a47 3 3 2 21 Integrated
a12 2 2 1 3 Transactional a45 2 3 3 22 Integrated
a13 4 1 1 3 Transactional a44 4 4 1 27 Integrated
a14 1 2 3 6 Transactional a39 1 4 4 28 Integrated
a15 1 3 2 6 Transactional a42 4 1 4 28 Integrated
a16 2 1 3 6 Transactional a50 3 2 4 31 Integrated
a17 2 3 1 6 Transactional a53 4 3 2 31 Integrated
a18 3 1 2 6 Transactional a48 2 3 4 32 Integrated
a19 3 2 1 6 Transactional a49 2 4 3 32 Integrated
a23 2 2 2 7 Collaborative a51 3 4 2 32 Integrated
a20 1 2 4 10 Collaborative a52 4 2 3 32 Integrated
a21 1 4 2 10 Collaborative a54 3 3 3 35 Integrated
a22 2 1 4 10 Collaborative a56 4 2 4 43 Integrated
a24 2 4 1 10 Collaborative a57 4 4 2 43 Integrated
a25 4 1 2 10 Collaborative a55 2 4 4 44 Integrated
a26 4 2 1 10 Collaborative a58 3 3 4 45 Integrated
a27 1 3 3 11 Collaborative a59 3 4 3 45 Integrated
a28 3 1 3 11 Collaborative a60 4 3 3 45 Integrated
a29 3 3 1 11 Collaborative a61 3 4 4 55 Integrated
a32 2 2 3 12 Collaborative a62 4 3 4 55 Integrated
a33 2 3 2 12 Collaborative a63 4 4 3 55 Integrated
a35 3 2 2 13 Collaborative a64 4 4 4 63 Integrated

5 Conclusions and future works

In this article, the comparison of two methods for supplier sorting for determining the sup-
plier management strategy in large organizations has been presented. Assigning a category to a
supplier is an important task within supply chain management since many types of suppliers are
commonly in place and differentiated management approaches are needed in order to accomplish
the efficiency and service objectives. The sorted categories combined three main dimensions:
strategic positioning (matrix 1), logistics complexity (matrix 2), and attraction of the mutual
relation (matrix 3). While the original Kraljic’s matrix is concerned only with strategic position-
ing, giving four categories of suppliers: non-critical or routine, leverage, bottleneck, and critical,
the contribution in this work is that the analysis has been extended to other dimensions, such
as logistics complexity (matrix 2) and the attraction of mutual relation (matrix 3).

By using the extended Kraljic’s matrix concept developed in this work, a portfolio of strategies
may be identified according to the defined categories: transactional, collaborative, and integrated
suppliers. Commonly, the analysis of suppliers is performed based on experienced managers over
a restricted number of cases. In large organizations, however, because of the high number of
suppliers, such manual method becomes difficult and prone to error. With multicriteria sorting

Multicriteria Supplier Classification for DSS:
Comparative Analysis of Two Methods 247

models it is possible to overcome this weakness and even automate this task. Flowsort was
compared to ELECTRE. A good coincidence was obtained between these two methods. However
Electre requires previously defined categories by using explicit enumeration and threshold values
entered manually by experts. In this sense, Flowsort requires less human input being more
adaptable for automated processing.

As direction for future works, further research is needed in order to examine the robustness
of the results and the effects of the scales used in the assessment. Also, in order to minimize
human data input, the scores in the matrices ideally should be obtained in a direct manner from
the key performance indicators (KPI) obtained from the ERP system of the organization, or
from a business intelligence (BI) module, among other aspects. Ongoing work of the authors is
addressing these issues.

Acknowledgments

The authors are very grateful to DICYT (Scientific and Tecnological Research Office), Project
Number 061117SS and the Industrial Engineering (IE) Department, both of the University of
Santiago of Chile for their support in this work. Also to IE graduates Marcos Melin and Stephanie
Sepulveda who helped in the data collection and model implementation.

Bibliography

[1] Kraljic, P. (1983); Purchasing must become supply management, Harvard Business Review,
September-October, 1983.

[2] Monczka, R.M., Handfield, R.B., Giunipero, K.C., Patterson, J.L. (2011); Purchasing And
Supply Chain Management, 5th Edition, Cengage Learning.

[3] Figueira, J., Greco S., Ehrgott M. (2005); Multiple Criteria Decision Analysis: State of the
Art Surveys, Springer-Verlag.

[4] Nemery P.; Lamboray C. (2008); FlowSort: a flow-based sorting method with limiting or
central profiles, TOP, 16, 90-113, Springer-Verlag.

[5] Sepulveda, J., Gonzalez, J., Alfaro, M. (2010); A Metrics-based Diagnosis Tool for Enhanc-
ing Innovation Capabilities in SMEs, International Journal of Computers Communications
& Control, 5(5):919-928.