MIT SCALE RESEARCH REPORT
The MIT Global Supply Chain and Logistics Excellence
(SCALE) Network is an international alliance of
leading-edge research and education centers, dedicated
to the development and dissemination of global
innovation in supply chain and logistics.
The Global SCALE Network allows faculty, researchers,
students, and affiliated companies from all six centers
around the world to pool their expertise and collaborate
on projects that will create supply chain and logistics
innovations with global applications.
This reprint is intended to communicate research results
of innovative supply chain research completed by
faculty, researchers, and students of the Global SCALE
Network, thereby contributing to the greater public
knowledge about supply chains.
For more information, contact
MIT Global SCALE Network
Postal Address:
Massachusetts Institute of Technology 77
Massachusetts Avenue, Cambridge, MA 02139 (USA)
Location:
Building E40, Room 267
1 Amherst St.
Access:
Tel: +1 617-253-5320
Fax: +1 617-253-4560
Email: scale@mit.edu
Website: scale.mit.edu
Research Report: ZLC-2010-7
Classification of SKUs as a Way to Improve Company Performance:
A Case Study from the Telecommunication Industry
Yuliya Kasirava and HanCheol Lee
MITGlobalScaleNetwork
For Full Thesis Version Please Contact:
Marta Romero
ZLOG Director
Zaragoza Logistics Center (ZLC) Edificio
Náyade 5, C/Bari 55 – PLAZA 50197
Zaragoza, SPAIN
Email: mromero@zlc.edu.es
Telephone: +34 976 077 605
MITGlobalScaleNetwork
Classification of SKUs as a Way to Improve
Company Performance: A Case Study from
the Telecommunication Industry
By Yuliya Kasirava and HanCheol Lee
Thesis Advisors: Dr. Mozart Menezes
Dr. Mustafa Çagri Gürbüz
Summary:
This thesis offers classification as one of the methods to lower the cost associated with spare parts inventory
for a company operating in the telecommunication industry. Several classification methods were examined
and compared in terms of total relevant costs. Further simulations were carried out in order to examine the
impact on total costs.
University Degree in
Economics with Major
in World Economy,
Grodno State
University, Belarus
Bachelor of Arts in
Geography
Education, Seoul
National University,
South Korea
Master of Economic
Science in European
Economic and Public
Affairs, University
College Dublin, Ireland
Supply Chain
Manager, LG
Telecom, Seoul,
South Korea
KEY INSIGHTS
1. Inventory classification is able to provide
benefits for the company via treating
heterogeneous items in a different way.
2. Different classifications produce diverse results,
which affect company performance in terms of
costs associated with inventory
3. There is no the ¨best¨ classification method. As
assumptions and conditions change, different
classifications outperform each other.
Introduction
Inventory being an integral component of operations
of virtually every company has been a matter of
great concern and increasing attention in the
academic and business world for the last few
decades. Indeed, carrying high levels of inventory
incurs financial and operating costs undesirable for
any company. However, for the companies
functioning in certain industries, the issue becomes
a matter of vital importance.
The present Thesis Project is sponsored by
company named TeleCo (a fictitious name was used
in order to hide the company‟s true identity) that
operates mainly as an Internet, mobile network and
cable television provider. The nature of these
operations implies rendering repair services, and
therefore, carrying a certain level of spare parts
inventory.
Companies operating in the telecom industry are
required to have as a high service level as possible.
Indeed, the outcomes of shortage of important spare
parts can be disastrous for telecom companies,
especially if institutions such as banks or hospitals
are affected. This service constraint leads to carrying
extremely high levels of spare parts, which in
general are expensive. The resulting high capital
cost moves companies in the opposite direction –
towards lowering inventory level.
The dilemma described above poses the following
challenge to TeleCo:
How could the Company optimize the level of spare
parts inventory given the service level in the
telecommunication industry it operates?
TeleCo holds a substantial number of spare parts,
which are extremely heterogeneous in many ways.
Naturally, this diversity requires applying different
inventory policies, which is not realistic without
appropriate spare parts classification. Hence, the
approach to inventory optimization without hurting
customer service level would be multicriteria
classification of spare parts.
First, we applied four classification methods to the
data sample obtained from TeleCo and illustrated
the disparities of classification results across
different approaches. Our next step was the
investigation of the impact that different
classifications have on the total cost associated with
inventory. Further, the sensitivity of total cost was
examined while changing different assumptions. The
final step was determining the optimum classification
for the data sample under study.
Classification
We applied seven criteria in order to classify the
sample of 109 spare parts obtained from TeleCo.
The criteria included price of spare part, total
demand, variation of demand across different
regions, mean of lead repair time, coefficient of
variation of lead repair time, criticality, and accuracy
of replacement.
Four classification methods were employed,
specifically, ABC analysis, multicriteria ABC-based
classifications using Analytic Hierarchy Process
(AHP) and weighted linear optimization (so called Ng
model), and cluster analysis.
ABC analysis is a simple and widely applied
technique in the managerial world, which is based on
80/20 Pareto law. The law states that 20% of SKUs
(class A items) generate 80% of dollar usage, which
is a product of price and annual demand. However, it
is often criticized for employing only a single criterion,
namely, annual dollar usage, and therefore ignoring
other crucial aspects worth being considered.
Multicriteria ABC-based classifications fill the gap,
introducing several criteria into the analysis. One of
the main issues with the methodology is criteria
prioritizing or assigning weights to each criterion. The
problem was solved by using the AHP approach,
which is based on pair wise comparisons of criteria
by decision makers. The alternative solution was
obtained by applying the linear weighted optimization
model in order to get weights for each criterion.
Clustering is a statistical method that is often
regarded as a form of classification. The goal is that
the objects within the groups be similar to each other,
and different from the objects in other groups. We
used the SPSS software package in order to obtain
three clusters which were further approximated to A,
B and C groups.
We found out that all classifications produce different
results, which are illustrated on the Figure below.
ABC
A B
A
5 8
Cluster B 13 8
C
7 14
sum 25 30
C sum
24 37
13 34
17 38
54 109
AHP
A B
A 12 6
B
7 4
ABC
C 6 20
sum 25 30
C sum
7 25
19 30
28 54
54 109
AHP
A B
A 17 17
Cluster B
8 9
C
0 4
sum 25 30
C sum
3
37
17 34
34 38
54 109
Ng
A B
A 16 7
B
6 5
ABC
C 3 18
sum 25 30
C sum
2 25
19 30
33 54
54 109
C sum
7
37
16 34
31 38
54 109
A
A 14
B
7
AHP
C 4
sum 25
Ng
B
11
15
4
30
C sum
0 25
8 30
46 54
54 109
A
A 12
Cluster B
6
C
7
sum 25
Ng
B
18
12
0
30
Total relevant cost (TRC)
In this paper, we considered inventory holding cost,
shortage cost, review management labor cost, and
transportation cost as a total relevant cost (TRC) to
determine the impact of each classification method
on management performance. Also, several
assumptions were made, including both endogenous
and exogenous assumptions. Under current
assumptions, the total relevant costs are seen below.
Current
ABC
AHP
Cluster Ng Model
Inventory Holding Cost
5,710
7,533
9,574
9,784
Review Management Cost
1,635
2,220
2,220
2,920
8,400
2,220
Transportation Cost
25,533
17,500
16,033
15,267
14,767
Shortage Cost
25,663
25,663
25,663
25,663
25,663
Total Relevant Cost
58,542
52,916
53,490
53,634
51,050
TRC simulations with endogenous assumptions
The simulations about endogenous assumptions,
which managers can change, allow managers to
easily find where the total TRC has the lowest value.
The endogenous assumptions include review periods,
service level, and transportation modes. First of all,
the review periods for group A items should be 2 or 3
weeks depending on the classification methods,
while those for group B should be 3 or 4 weeks, and
those for group C 3 or 4 months. (Green shaded cells
represent the lower cost, while red cells the higher
cost.)
Ng
Less frequent
Group B review period (months)
3.50
3.75
4.00
51,970
52,083
52,196
51,717
50,949
50,779
50,758
50,795
50,862
50,945
51,037
51,137
51,240
51,347
51,456
51,566
51,678
51,791
51,904
0.75
51,902
51,134
50,964
50,943
50,980
51,047
51,130
51,223
51,322
51,425
51,532
51,641
51,751
51,863
51,976
52,089
1.00
52,194
51,426
51,256
51,235
51,273
51,339
51,422
51,515
51,614
51,718
51,825
51,933
52,044
52,156
52,268
52,381
1.25
52,522
51,754
51,585
51,563
51,601
51,668
51,750
51,843
51,942
52,046
52,153
52,262
52,372
52,484
52,596
52,710
0.25
More0.50
1.50
Group A
review
period
(months)
Group B
More frequent
(Short review period)
0.25
0.50
0.75
1.00
1.25
1.50
1.75
2.00
2.25
2.50
2.75
3.00
3.25
52,009
51,241
51,071
51,050
51,087
51,154
51,237
51,329
51,429
51,532
51,639
51,748
51,858
52,096
51,926
51,905
51,942
52,009
52,092
52,185
52,284
52,387
52,494
52,603
52,714
52,825
52,938
53,051
1.75
53,209
52,441
52,272
52,250
52,288
52,355
52,437
52,530
52,629
52,733
52,840
52,949
53,059
53,171
53,283
53,397
2.00
52,864
53,555
52,787
52,617
52,596
52,634
52,700
52,783
52,876
52,975
53,079
53,186
53,294
53,405
53,517
53,629
53,742
2.25
53,899
53,131
52,962
52,940
52,978
53,045
53,127
53,220
53,319
53,423
53,530
53,639
53,749
53,861
53,973
54,087
2.50
54,241
53,473
53,303
53,282
53,319
53,386
53,469
53,562
53,661
53,765
53,871
53,980
54,091
54,202
54,315
54,428
2.75
54,579
53,811
53,642
53,620
53,658
53,725
53,807
53,900
53,999
54,103
54,210
54,319
54,429
54,541
54,653
54,767
3.00
54,915
54,147
53,977
53,956
53,993
54,060
54,143
54,235
54,335
54,438
54,545
54,654
54,764
54,876
54,989
55,102
3.25
55,247
54,479
54,309
54,288
54,325
54,392
54,475
54,567
54,667
54,770
54,877
54,986
55,096
55,208
55,321
55,434
3.50
55,575
54,807
54,638
54,616
54,654
54,721
54,803
54,896
54,995
55,099
55,206
55,315
55,425
55,537
55,649
55,763
3.75
55,901
55,133
54,963
54,942
54,980
55,046
55,129
55,222
55,321
55,425
55,531
55,640
55,751
55,862
55,975
56,088
56,224
55,456
55,286
55,265
55,302
55,369
55,452
55,545
55,644
55,748
55,854
55,963
56,074
56,185
56,298
56,411
Group
A
Less4.00
When it comes to transportation modes, once
classification methods are introduced, then it is
easier to apply aggregate orders because different
review periods and policies can be applied to the
company‟s inventory. As an example of this, the Ng
model can have the lowest cost if group A items are
delivered by ground parcel and group B and C items
by less-than-truckload.
TRC simulations with exogenous assumptions
The simulations about exogenous assumptions,
which managers cannot choose, can make it easier
to predict what classification method will be the best
under given situations, which accordingly allows
managers to properly cope with future changes in
advance. These exogenous assumptions include the
inventory holding unit cost, the labor salary, and the
transportation cost.
There can be a trade-off between inventory holding
cost and salary for those in charge of inventory
review activity, because frequent review usually
lowers the inventory level and increases the labor
cost. If other assumptions stay the same, the Ng
model has the best management performance and a
quite stable position regarding changes in the salary
level. Also, the Ng model yields the lowest cost
regardless of the inventory holding cost.
Unlike that of the salary and the inventory holding
cost, the simulation results of transportation costs
show unstable patterns of the best classification like
the Figures below. First of all, if the variable cost of
the air parcel service is low, then the current method,
in which TeleCo orders spare parts whenever one
item is used, is the best. This is because there is no
reason to aggregate procurement orders. However,
as the variable cost of the air service increases, the
probability that the current situation would be the
best decreases.
Variable Cost for Air Parcel
Low
High
Current
Current
Current
Current
Current
Current
Current
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
ABC
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
ABC
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Current
Current
Current
Current
Current
Ng
Ng
Ng
ABC
Current
High
Ng Model
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Cluster
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Current
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Current
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Variable Cost for Air Parcel
Low
Low
High
Ng
Current
ABC
Cluster
Fixed Cost for LTL
Low
Variable cost
for Ground Parcel
As for the service level, under current assumptions,
the spare parts with „H+‟ and „NBD/E‟ criticality
should have 99% of service level while those with
„NBD‟ 96% under current assumptions. This is mainly
because the impact of penalty constant currently
assumed is much higher than that of inventory
holding cost, which is resulting from the attribute of
telecom industry – “always on service”. When the
assumptions about penalty constants are relaxed
while increasing the inventory holding cost, the
optimal set of service levels can be changed.
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Cluster
Cluster
Cluster
Current
Current
Current
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Cluster
Current
Current
Current
ABC
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Current
Current
Current
ABC
ABC
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Current
Current
Current
Current
ABC
ABC
Ng
Ng
Ng
Ng
Ng
Current
Current
Current
Current
ABC
ABC
ABC
Ng
Ng
Ng
Current
Current
Current
Current
ABC
ABC
ABC
ABC
Ng
Ng
Current
Current
Current
Current
Current
ABC
ABC
ABC
ABC
Current
Current
Cluster
High
Current
ABC
Cluster
Cluster
Cluster
Cluster
Cluster
Cluster
Ng
Ng
Cluster
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng
Ng Model
Current
Current
Current
Current
ABC
ABC
ABC
ABC
ABC
Ng
Ng
Ng
Ng
Ng
Ng
Current
Current
Current
Current
Current
Current
ABC
ABC
ABC
ABC
ABC
Ng
Ng
Ng
Ng
Ng
Current
Current
Current
Current
Current
Current
ABC
ABC
ABC
ABC
ABC
ABC
Ng
Ng
Ng
Ng
Current
Current
Current
Current
Current
Current
ABC
ABC
ABC
ABC
ABC
ABC
ABC
Ng
Ng
Ng
Current
Current
Current
Current
Current
Current
Current
ABC
ABC
ABC
ABC
ABC
ABC
ABC
Ng
Current
Current
Current
Current
Current
Current
Current
ABC
ABC
ABC
ABC
ABC
ABC
ABC
ABC
Current
Current
Current
Current
Current
Current
Current
ABC
ABC
ABC
ABC
ABC
ABC
ABC
ABC
Ng
Current
Current
Current
Current
Current
Current
Current
Current
ABC
ABC
ABC
ABC
ABC
ABC
ABC
ABC
Ng
Ng
Conclusions
Our research started with the belief that
classifications of spare parts can provide TeleCo with
the opportunity to reduce the total relevant cost, like
the transportation cost, without hurting the service
level because different decisions on the inventory
can be made by the classification segment.
This paper introduced several classification methods
to TeleCo, including not only the ABC method but
also the multicriteria classification methods such as
the AHP, the Cluster, and the linear optimization
model. The reason why multicriteria classification
methods were introduced is that the ABC method
cannot perfectly reflect the attribute of the telecom
industry – „always on‟.
After that, the impact of each classification method
on the management performance was examined to
see which one can have the lowest total relevant cost.
On top of that, several simulations about how the
total relevant cost and the best classification method
vary were conducted by applying different levels of
assumptions.
One of implications from the findings of this research
is that there is not „one best‟ classification method
that can be applied for all industries or for all
situations. Therefore, managers have to revise the
company‟s classification of spare parts and to
examine this impact regularly. In addition, many
papers regarding classification usually compare only
the different grouping results of classifications, but do
not consider the management impact. In this sense,
this paper can make a contribution to the field of
classification research.