A Comparative Study on Multicriteria ABC Analysis

in Inventory Management
Ye Chen

Kevin W. Li

Si-feng Liu

College of Economics and Management

Odette School of Business
College of Economics and Management
Nanjing Univ. of Aeronautics and Astronautics
University of Windsor
Nanjing Univ. of Aeronautics and Astronautics
Nanjing, 210016, China
Windsor, ON, N9B 3P4, Canada
Nanjing, 210016, China
[email protected]
[email protected]
[email protected]

AbstractABC analysis is a well-known approach to classifying inventory items. Traditionally, a single criterion, the annual
dollar usage of stock-keeping units (SKUs), is employed in the
classication process. However, in many cases, other criteria may
also play a signicant role in classifying SKUs. Therefore, a recent
trend is to develop various procedures to conduct ABC analysis in
terms of multiple criteria. This article examines several multiple
criteria ABC analysis procedures and a comparative study of
their classication results is carried out by using 47 SKUs in a
hospital inventory system. Our results indicate that these different
approaches produce statistically consistent rankings.
Index TermsMulticriteria ABC Analysis, Inventory management, Comparative study

With the rapid growth of globalization, the economic activities in different countries are increasingly intertwined with
each other. For example, many retailers in US such as Walmart,
Sears, and Bestbuy, are now manufacturing most of their
products in Asia, especially in China. The economic incentive
behind such action is that the cost of manufacturing and
shipping of products in Asia is much cheaper than those made
in US. Hence, these retailers need to design effective inventory
management strategies to help them prevent possible shortage
of products and maintain competitive advantage, especially in
a time of accelerating globalization [23].
The number of stock-keeping units (SKUs) held by a
firm can easily reach tens of thousands. Clearly, it is not
economically feasible to design inventory management policies
at an individual SKU level. In addition, different SKUs may
play quite different roles in the firms business and, hence,
call for different levels of management attention. In order to
implement a sound inventory control scheme, it is necessary to
group SKUs into manageable and meaningful categories first,
and then design different policies for each group according
to the groups importance to the firm [7]. Thus, a generic
inventory management policy requiring a certain level of
effort and control from management is applied to all items in
each category. This aggregation process should dramatically
reduce the number of SKUs requiring extensive management
attention.
A. ABC Analysis in Inventory Management
ABC analysis is a widely used inventory management
technique that is designed to classify SKUs into varying groups
with different levels of importance to the inventory system. The

most important SKUs are placed in group A, which demand

the greatest effort and attention from management; the least
important SKUs fall into group C, where minimal effort is
applied; other SKUs belong to the middle group B. Figure 1
demonstrates the idea of ABC analysis.

. . ..
. . ..
.. . . . .
. .
SKUs

Fig. 1.

..
..

Group A

..

Group B

..
..

Group C

..
..

More Importance

Less Importance

An Illustration of ABC Analysis

The traditional ABC analysis is conducted based on a

single criterion, the annual dollar usage of SKUs, reflecting
the principle that a small proportion of SKUs accounts for a
majority of the dollar usage. The classification usually follows
the so-called 80-20 (or 90-10) Rule, roughly 80% (or 90%)
of the total annual usage comes from 20% (or 10%) of
SKUs. This rule suggests that the number of SKUs in A is
substantially smaller than the total number of SKUs. Although
exact values vary from an industry to another, the 80-20 rule
can be applied to many real-world situations. Figure 2 captures
the essence of this rule.
B. The Pareto Principle
The classical ABC analysis originates from the Italian
economist Vilfredo Paretos famous observation on the uneven
distribution of national incomes in Italy that about 80% property was owned by 20% of the Italian population [18], and,
hence, is sometimes referred to as the Pareto principle or Pareto
analysis. There exist many practical applications or phenomena
of Pareto principle across different fields, and some examples
are listed below:
In software engineering, it is often observed that 90% of
the execution time of a computer program is spent on

3280
c 2008 IEEE
1-4244-2384-2/08/$20.00 

1.00

0.9251

Cumulative percentage of dollar usage

0.90
0.80
0.70

0.7357

0.60

0.2128

0.40
0.30
0.20
0.10
0.00
0.00

Group B
(14 SKUs)

Group A
(10 SKUs)

0.10

0.20

0.30

Fig. 3.

0.40

20%

0.50

0.60

Cumulative percentage of SKUs

0.70

0.80

0.90

1.00

The Dollar Usage Distribution Curve of Table I

Cumulative
percentage of
dollar usage
80%

Fig. 2.

Group C
(23 SKUs)

0.5106

0.50

Cumulative
percentage
of SKUs

An Example of the Dollar Usage Distribution Curve

10% of the code. Also, 80% of users use only 20% of a

programs features, even, for Microsoft Excel, the 95/5
rule is applied: 95% of Excel users use a mere 5% of the
programs features [16].
In business, it is seen that 20% of customers, activities,
products or processes contributing 80% of the overall
profit. For example, it is estimated that the sales of
December-January in US, the traditional Christmas shopping season, can account for more than 50% of the retails
annual sales and 70% of their profits.
In human society, a few cities or regions may dominate
the population of a nation. For example, with 10% of
the national land, the province of Ontario in Canada has
roughly 40% of the total population in Canada, and within
Ontario, the Greater Toronto Area solely hosts around
18% of the national population.

I. P ROCEDURES FOR ABC A NALYSIS

The application of the Pareto principle in inventory management began in early 1950s. Some of the pioneer research
includes [6], [8], [15], to name a few. The general procedure
for conducting ABC analysis can be summarized as follows:
Select a proper criterion for classifying SKUs. The annual
dollar usage (ADU) is a widely used criterion.
Collect required information and calculate ADU for each
SKU, where the ADU is obtained from the multiplication
of unit price and annual demand of an SKU.
Arrange SKUs in a descending order of ADU.
Calculate the cumulative percentage of SKUs and their
corresponding cumulative percentage of ADU, as shown
in Figure 2.
Determine the thresholds of classification for each group
and arrange SKUs into different groups according to the
established rule.
Figure 3 demonstrates the classical ABC analysis of a numerical example adpated from [12]. The detailed information
is shown in Table I in Section IV, in which the first and
second columns of the table represent the SKUs IDs and the
corresponding annual dollar usages.
II. M ULTIPLE C RITERIA ABC A NALYSIS
The classical ABC analysis has been criticized due to the
fact that the classification depends on a single criterion, the
annual dollar usage [12]. In many cases, other attributes of
an SKU often play a significant role in determining which
group an SKU should be classified. For instance, if two
SKUs are virtually identical except that one is easy to replace
while the other is unique and has only one specific supplier.
Understandably the SKU with higher substitutability should
receive less management attention. Other criteria that could
be accounted for include obsolescence, repairability, criticality,
and lead time [10], [11].

2008 IEEE International Conference on Systems, Man and Cybernetics (SMC 2008)

3281

To carry out multiple criteria classification of SKUs, a variety of approaches has been proposed. One of the first attempts
was the Flores and Whybarks bi-criteria matrix method [10].
This approach begins with selecting another critical criterion,
in addition to the annual dollar usage, depending on the nature
of the industry. Next, the model requires that SKUs be divided
into three levels of importance, A, B, and C, for each of the
two criteria, respectively. The model then reclassifies SKUs
into three categories, AA, BB, and CC, representing the three
new groups, according to some rules jointly determined by the
new criterion and the dollar usage. The structure of the model
can be conveniently represented as a joint criteria matrix as
shown in Figure 4, adapted from [10]. A general guideline as
indicated by the arrows is to regroup AB and BA as AA, AC
and CA as BB, and BC and CB as CC.

Another Critical Criterion

TABLE I
L ISTING OF SKU S WITH MULTIPLE CRITERIA , ADAPTED FROM [12]
SKUs

Criteria
ADU ($)

AUC ($)

CF

S1

5840.64

49.92

LT (week)
2

S2

5670.00

210.00

S3

5037.12

23.76

S4

4769.56

27.73

0.01

S5

3478.80

57.98

0.5

S6

2936.67

31.24

0.5

S7

2820.00

28.20

0.5

S8

2640.00

55.00

0.01

S9

2423.52

73.44

S10

2407.50

160.50

0.5

S11

1075.20

5.12

S12

1043.50

20.87

0.5

S13

1038.00

86.50

S14

883.20

110.40

0.5

AA

AB

AC

Dollar
Usage B

BA

BB

BC

S15

854.40

71.20

S16

810.00

45.00

0.5

CA

CB

CC

S17

703.68

14.66

0.5

S18

594.00

49.50

0.5

S19

570.00

47.50

0.5

S20

467.60

58.45

0.5

Some other approaches to multiple criteria ABC analysis

(MCABC) developed recently are summarized below:
Multiple criteria decision analysis (MCDA)-based approaches: the analytic hierarchical process (AHP) approaches [12], [19], a case-based distance approach [3],
and a weighted linear model [17].
Data envelopment analysis (DEA)-based approaches: a
weighted linear optimization model [21] and its extension,
[24] and a flexible multiple criteria sorting method [2].
Artificial intelligence-based approaches: the artificial neural network approach [20], the genetic algorithm [14], and
the dominance-based rough set approaches [4], [5].

S21

463.60

24.40

S22

455.00

65.00

0.5

S23

432.50

86.50

S24

398.40

33.20

S25

370.50

37.05

0.01

S26

338.40

33.84

0.01

S27

336.12

84.03

0.01

S28

313.60

78.40

0.01

S29

268.68

134.34

0.01

S30

224.00

56.00

0.01

S31

216.00

72.00

0.5

S32

212.08

53.02

III. A N UMERICAL E XAMPLE

S33

197.92

49.48

0.01

A. Background
A case study to compare some of the recently proposed
MCABC methods is carried out based upon data provided
by [12] on a hospital inventory management problem. In
this example, 47 disposable SKUs used in a hospital-based
respiratory therapy unit are classified using an AHP [22] based
MCABC method. Table I lists the 47 disposable SKUs referred
to as S1 through S47. Four criteria are identified for the
MCABC analysis: (1) annual dollar usage ($), ranging from
$25.38 to $5840.64; (2) average unit cost ($), varying from
$5.12 to $210.00; (3) critical factor, 1, 0.50, or 0.01 is assigned
to each of the 47 disposable SKUs, where a value of 1 indicates
very critical, a value of 0.50 means moderately critical, and a
value of 0.01 stands for for non-critical; (4) lead time (weeks)
is the time that it takes to receive a replenishment order after
it is placed, ranging from 1 to 7 weeks.

S34

190.89

7.07

0.01

S35

181.80

60.60

0.01

S36

163.28

40.82

S37

150.00

30.00

0.01

S38

134.80

67.40

0.5

S39

119.20

59.60

0.01

S40

103.36

51.68

0.01

S41

79.20

19.80

0.01

S42

75.40

37.70

0.01

S43

59.78

29.89

0.01

S44

48.30

48.30

0.01

S45

34.40

34.40

0.01

S46

28.80

28.80

0.01

S47

25.38

8.46

0.01

Fig. 4.

3282

A Joint Matrix for Two Criteria

2008 IEEE International Conference on Systems, Man and Cybernetics (SMC 2008)

B. Result Comparison of Different Approaches

As mentioned before, in addition to the initial work of
Flores et al. [12], several MCABC methods, including [3],
[17], [21], and [24], will be applied to this data set to conduct
a comparative study. The detailed results of ranking and
classification group of these 47 SKUs from approaches of [3],
[12], [17], [21], and [24] are shown in Table III. Note that in
the last two columns of the table, Chen et al.1 and Chen et al.2
represent the ideal and anti-ideal-based distance approaches
proposed in [3], respectively. Note that the case-based distance
approach requires a training set with representative SKUs from
each group to calibrate the model. In this illustration, the
representative SKUs are listed in Table II, in which TA , TB
and TC stand for the representative case sets of A, B and C,
respectively.

the Friedman tests demonstrate that the rankings are not

significantly different and there exists overall consistency in
the rankings among the aforementioned six methods. From
an application perspective, it is possible that an appropriate
ranking aggregation algorithm such as [9] may be employed
to integrate the rankings from different approaches and obtain
a final result.
TABLE IV
F RIEDMAN T EST OF F ULL R ANKINGS OF 47 SKU S
Q (observed value)

10.242

Q (critical value)

12.592

DF

One-tailed p-value

0.115

Alpha

0.050

TABLE II
R EPRESENTATIVE T RAINING S ET
Group
TA

TB

TC

SKUs
S1
S2
S13
S10
S29
S36
S45
S4
S25
S27
S34

ADU($)
5840.64
5670.00
1038.00
2407.50
268.68
163.28
34.40
4769.56
370.50
336.12
190.89

Criteria
AUC($)
CF
49.92
1.00
210.00
1.00
86.50
1.00
160.50
0.50
134.34
0.01
40.82
1.00
34.40
1.00
27.73
0.01
37.05
0.01
84.03
0.01
7.07
0.01

TABLE V
F RIEDMAN T EST OF G ROUP C LASSIFICATIONS OF 47 SKU S
LT (Week)
2.00
5.00
7.00
4.00
7.00
3.00
7.00
1.00
1.00
1.00
7.00

Assume that these 47 SKUs are a sample from all SKUs

in the inventory system, after the MCABC models here
are calibrated using these SKUs, these models can then be
employed to categorize remaining SKUs in the inventory
system. Before any of the approaches is applied in practice, a
natural question is whether the ranking results from them are
consistent from a statistical perspective. Given the nature of the
data, a Friedman test [13] is conducted to test whether there
exists any significant difference among the rankings across the
six evaluation approaches. In this case, our two hypotheses are:
H0 : The rankings of SKUs for the six MCABC approaches are the same.
H1 : At least two rankings are different.
The software, Xlstat [1], is employed to carry out the
calculation. Two Friedman tests are entertained: the first test
examines the full rankings of the 47 SKUs in different approaches and the result is listed in Table IV; the second tests
the group classifications of the 47 SKUs and the result is
listed in Table V. Both tests show that at the significance
level Alpha = 0.050, we should not reject H0 in favour of H1 .
In the context of these tests, there is no significant statistical
evidence to infer that the ranking results from the six MCABC
approaches are different. In other words, although different
approaches generate varying ranking orders and classifications,

Q (observed value)

12.486

Q (critical value)

12.592

DF

One-tailed p-value

0.052

Alpha

0.050

IV. C ONCLUSIONS
This paper compares several multiple criteria ABC analysis
techniques by using a sample of 47 SKUs from a hospital
inventory system. Two Friedman tests confirm that these different approaches produce overall consistent ordinal rankings of
SKUs. Some issues remain open. In the current literature, the
number of classification groups is always assumed to be three,
i.e. A, B, and C group only. For firms in different industries,
are three groups always the best classification? Is it possible
that four or five groups can be a better classification scheme
than three? It is a worthwhile topic to determine the optimal
number of groups first and, then, adapt MCABC approaches
accordingly.
R EFERENCES
[1] Addinsoft, Inc., Xlstat, http://www.xlstat.com, cited on October 24, 2007.
[2] Y. Chen, K.W. Hipel, and D.M. Kilgour, A flexible multiple criteria
sorting method with application in inventory management, the Proceedings of the 2006 IEEE International Conference on Systems, Man and
Cybernetics, pp. 954-959, 2006.
[3] Y. Chen, K.W. Li, D.M. Kilgour and K.W. Hipel, A case-based distance
model for multiple criteria ABC analysis, Computers and Operations
Research, vol.35, pp. 776-796, 2008.
[4] Y. Chen, K.W. Li, J. Levy, D.M. Kilgour and K.W. Hipel, RoughSet multiple-criteria ABC analysis, Springer Lecture Notes in Artificial
Intelligence Series (LNAI) vol.4259, S. Greco et al. (Eds.), pp. 328-337,
2006.
[5] Y. Chen, K.W. Li, J. Levy, D.M. Kilgour and K.W. Hipel, A rough set
approach to multiple criteria ABC analysis, Springer LNCS Transactions
on Rough Sets, in press, 2008.
[6] K.L. Campbell, What comes after the ABCs, Production and Inventory
Management, vol. 109, pp. 20-26, 1968.

2008 IEEE International Conference on Systems, Man and Cybernetics (SMC 2008)

3283

TABLE III
L ISTING OF C LASSIFICATION R ESULTS FROM D IFFERENT A PPROACHES
SKUs

3284

Flores et al. [12]

Ramanathan [21]

Ng [17]

Zhou and Fan [24]

Chen et al.1 [3]

Chen et al.2 [3]

Ranking

Group

Ranking

Group

Ranking

Group

Ranking

Group

Ranking

Group

Ranking

Group

S1

13

24

S2

29

S3

46

10

S4

25

33

20

S5

23

39

21

19

S6

21

32

10

27

25

S7

38

11

28

10

26

S8

15

42

12

13

25

31

S9

18

43

S10

24

37

14

S11

36

44

37

43

24

11

S12

14

47

23

19

13

22

S13

10

45

S14

31

13

18

S15

32

27

28

14

B
C

S16

19

31

33

23

35

S17

11

12

31

32

21

30

S18

12

15

12

15

S19

29

10

20

13

16

23

S20

22

20

26

24

19

29

S21

20

21

31

30

15

S22

17

22

25

21

18

28

S23

45

15

20

19

11

S24

25

37

35

20

10

S25

34

36

45

47

46

47

S26

26

37

35

40

41

S27

23

31

41

45

43

S28

28

14

29

24

S29

16

28

13

S30

38

31

43

46

47

45

S31

40

13

19

13

17

21

S32

39

35

43

41

26

12

S33

33

14

24

16

34

33

S34

37

18

17

32

17

S35

43

28

36

34

39

39

S36

47

30

37

35

22

S37

16

27

21

36

36

S38

35

27

31

35

27

34

S39

44

16

20

18

35

32

S40

26

11

17

11

30

27

S41

46

40

47

44

44

46

S42

41

45

44

43

44

C
C

S43

42

16

29

26

37

37

S44

41

32

37

39

41

40

S45

27

15

12

31

16

S46

30

32

42

40

42

42

S47

25

16

30

30

38

38

2008 IEEE International Conference on Systems, Man and Cybernetics (SMC 2008)

[7] A.K. Chakravarty, Multi-item inventory aggregation into groups, Journal of Operational Research Society, vol.32(1), pp. 19-26, 1981.
[8] H.F. Dickie, ABC inventory analysis shoots for dollars, not pennies,
Factory Management and Maintance, vol. 109, pp. 92-94, 1951.
[9] C. Dwork, R. Kumar, M. Naor and D. Sivakumar, Rank aggregation
methods for the Web, the Proceedings of the 10th international conference on World Wide Web, pp. 613-622, May 01-05, 2001.
[10] B.E. Flores and D.C. Whybark, Multiple criteria ABC analysis,
International Journal of Operations and Production Management, vol.
6, pp. 38-46, 1986.
[11] B.E. Flores and D.C. Whybark, Implementing multiple criteria ABC
analysis, Journal of Operations Management, vol. 7, pp. 79-84, 1987.
[12] B.E. Flores, D.L. Olson and V.K. Dorai, Management of multicriteria
inventory classification, Mathematical and Computer Modeling, vol.16,
pp. 71-82, 1992.
[13] M. Friedman, The use of ranks to avoid the assumption of normality
implicit in the analysis of variance, Journal of the American Statistical
Association, Vol. 32 (200), pp. 675-701, 1937.
[14] H.A. Guvenir and E. Erel, Multicriteria inventory classification using a genetic algorithm, European Journal of Operational Research,
vol.105(1), pp. 29-37, 1998.
[15] J.M. Juran, Jurans Quality Handbook, 5th ed. (first edition published in
1951), New York: McGraw-Hill Professional, 1998.
[16] P. McFedries, Formulas and Functions with Microsoft Excel 2003,
Indianapolis, Indiana: Que Publishing, 2004.
[17] W.L. Ng, A simple classifier for multiple criteria ABC analysis,
European Journal of Operational Research, vol. 177, pp. 344353, 2007.
[18] V. Pareto, Mannual of Political Economy (English translation), New
York: A. M. Kelley Publishers, 1971.
[19] F.Y. Partovi and W.E. Hopton, The analytic hierarchy process as
applied to two types of inventory problems, Production and Inventory
Management Journal, vol.35(1), pp. 13-19, 1994.
[20] F.Y. Partovi and M. Anandarajan , Classifying inventory using an
artificial neural network approach, Computer and Industrial Engineering,
vol. 41, pp. 389-404, 2002.
[21] R. Ramanathan, ABC inventory classification with multiple-criteria
using weighted linear optimization, Computer and Operations Research,
vol. 33, pp. 695-700, 2006.
[22] T.L. Saaty, Analytic hierarchy process, New York: McGraw Hill, 1980.
[23] E.A. Silver, D.F. Pyke and R. Peterson, Inventory Management and
Production Planning and Scheduling, 3rd edition, New York: Wiley, 1998.
[24] P. Zhou and L. Fan, A note on multi-criteria ABC inventory classification using weighted linear optimization, European Journal of Operational Research, vol. 182, pp. 1488-1491, 2007.

2008 IEEE International Conference on Systems, Man and Cybernetics (SMC 2008)

3285