Combining RFM Model and Clustering

Techniques for Customer Value Analysis of a

Company selling online

Rachid AIT DAOUD Belaid BOUIKHALENE

Department of Computer Science
Department of physics Sultan Moulay Slimane University
Sultan Moulay Slimane University Beni Mellal, Morocco
Beni Mellal, Morocco [email protected]
[email protected]
Rachid LBIBB
Abdellah AMINE
Department of physics
Department of Mathematics and Applications Mathematics
Sultan Moulay Slimane University
Sultan Moulay Slimane University
Beni Mellal, Morocco
Beni Mellal, Morocco
[email protected]
[email protected]

Abstract²A case study of applying RFM (recency, frequency, transactional data consist of 730 customers who have
and monetary) model and clustering techniques in the sector of purchased the website of the company from November 2013 to
HOHFWURQLFFRPPHUFHZLWKDYLHZWRHYDOXDWLQJFXVWRPHUV¶YDOXHV January 2015. The profile for each customer includes the
is presented. Self-organizing maps method (SOM) is first used to customer identifier, gender, birth date, city, shopping
determine the best number of clusters and then K-means method frequency, date of first transaction, date of last purchase and
is applied to classify 730 customers into eight clusters when R, F, the total spending at the online store.
and M are the segmenting variables, and then developing The transaction recorded for each customer must be
effective marketing strategies for each cluster. The average transformed to a usable format for the RFM model in this
values of RFM are computed for each cluster and the overall
study. Therefore, customer values of different clusters can be
customers. The values of RFM variables for each cluster greater
measured by the use of clustering techniques and RFM model.
than those of the overall average are identified. The results show
that the cluster 7 is the most important cluster because the The remainder of the paper is as follows: Section 2
average values of R, F and M are higher than the overall average provides the literature review on RFM (recency, frequency and
value. In summary, the purpose of this case study is customer monetary) model. In Section 3, Cluster analysis is depicted. A
segmentation using RFM model and clustering algorithms (SOM case study of applying RFM model and clustering analysis is
and K-means) to specify loyal and profitable customers for summarized in Section 4 and last section presents a brief
achieving maximum benefit and a win-win situation. conclusion.

Keywords²RFM model, Customer value, loyalty, Cluster analysis, II. LITERATURE REVIEW OF RFM MODEL
Self-Organizing Maps method (SOM), K-means algorithm.
Recency, frequency and monetary (RFM model) is an
I. INTRODUCTION effective method of segmenting and it is likewise a behavioral
A real case study for an online selling company in Morocco analysis that can be employed for market segmentation [1,
is employed by combining RFM (recency, frequency and 2].Hughes [1] described that the main asset of the RFM
monetary) model and data mining techniques (cluster analysis) PHWKRG LV RQ WKH RQH KDQG WR REWDLQ FXVWRPHUV¶ behavioral
to achieve better market segmentation and improve customer analysis in order to group them into homogeneous clusters,
satisfaction. Data mining techniques such as Self Organizing and on the other hand, to develop a marketing plan tailored to
Maps and K-means are used in this study to divide all each specific market segment. RFM analysis improves the
customers into an appropriate number of clusters. On the other market segmentation by examining the when (recency), how
hand, the customers are segmented into similar clusters often (frequency), and the money spent (monetary) in a
according to their RFM values. Therefore, the characteristics of particular item or service [3]. Yang [3] has summarized that
each cluster are examined in order to determine and retain customers who had bought most recently, most frequently, and
profitable and loyal customers and then develop effective had spent the most money would be much more likely to react
marketing strategy for each cluster of customers. The to the future promotions.

978-1-5090-0478-2/15/$31.00 ©2015 IEEE

The advantage of RFM model resides in its relevance as long K-means method is used for grouping n vectors based on
as it operates on several variables which are all observable and attributes into k partitions, where k < n, depending to some
REMHFWLYH 7KH\ DUH DOO DYDLODEOH DW WKH RUGHU¶V SDVW IRU HDFK measures. The name comes from the fact that k clusters are
customer. These variables are classified according to three identified and the center of a cluster is the mean of all vectors
independent criteria, namely recency, frequency and within this cluster. The algorithm starts with choosing k
monetary. Recency is the time interval between the last random initial centroids, then assigns vectors to the nearest
purchase and a present time reference; a lower value centroid using Euclidean distance and recalculates the new
corresponds to a higher probability that a customer will make centroids as means of the assigned data vectors. This process is
repeated many times until vectors no longer altered clusters
a repeat purchase. Frequency is the number of transactions
between iterations [13]. The K-means method is arguably a
that a customer has made in a particular time period and
non-hierarchical method.
monetary means the amount of money spent in this specified
time period [4]. The traditional approach to adopt RFM model However, SOM has a few disadvantages. For example, with
LV WR VRUW WKH FXVWRPHUV¶ GDWD YLD HDFK YDULDEOH RI 5)0 DQG the result generated by SOM technique, it is difficult to detect
then divide them into five equal quintiles [5, 6].The process of clustering boundaries, a fact which limits their application to
segmentation begins with sorting all customers based on automatic knowledge [10]. Furthermore, in the k-means
recency, then frequency and monetary. For recency, the technique, the number of clusters and the initial starting point
customer database is sorted in an ascending order (most recent are randomly selected, which means that the algorithm has to
purchasers at the top). Customers are then sorted for frequency turn several times to identify strong forms, because the final
and monetary in a descending order (most frequently and had result depends on the initial starting points (different initial k
spent the most money were at the top).The customers are then objects may produce different clustering results). Due to the
split into quintiles (five equal groups), and given the top weakness of SOM and k-means method, the integration of
20%segment is assigned as a value of 5, the next 20% segment these methods becomes desirable. Punj and Steward (1983)
is coded as a value of 4, and so on. Therefore, all customers [14] took this view, adopting a two-staged clustering method
are represented by one of 125 RFM cells, namely, 555, 554, by integrating the hierarchical method into the non-
553, . . . , 111. hierarchical. Kuo and al [15] have pointed out that it is
Customers who have the most score are profitable. In this preferable to use iterative partitioning methods instead of the
study, we adopt another approach proposed by chang et al. hierarchical methods if the initial centroid and number of
(2010) [7], it consistsof using the original data rather than the clusters are provided. If the information is provided, the
coded number. The definitions are as follows: recency is the iterative method consistently finds better clusters and higher
time interval between the first day of study period and the last accuracy than the hierarchical methods and yields faster
purchase; frequency is the number of transactions that a results because the initialization procedure that ultimately
customer has made in a particular time period and monetary determines the number of iteration is already executed.
means the amount of money spent in this specified time Therefore, self-organizing maps is used to determine the
period. number of clusters for K-Means method.

III. REVIEW OF CLUSTER ANALYSIS IV. A CASE STUDY

Data mining techniques have been widely employed in In this section the proposed model to determine loyal and
different domains. As the transactions of an organization profitable customers is described. Fig. 1 shows the required
become much larger in size, data mining techniques, steps for the proposed model.
particularly the clustering technique, can be applied to divide Dataset used in this case study was provided by a company
all customers into several clusters based on some similarities selling online in Morocco and collected through its e-
in these customers [8]. Clustering techniques are used to commerce website. The customer purchase database consists of
identify a set of groups that both minimize within-group 730 customers who purchased directly from the company
variation and maximize between-group variation according to a website from November 2013 to January 2015. The profile for
distance or dissimilarity function [9]. each customer includes the customer identifier, gender, birth
date, city, shopping frequency, date of first transaction, date of
The SOM (Self-Organizing Map) is an unsupervised neural last purchase and the total expense. Specifically, the notations
network methodology, which need only the input, is used to of male and female are m and f, respectively.
clustering for problem solving [10] and market screening [11].
The network is formed by an unsupervised competitive The definition of the RFM model used in this study is
learning algorithm, which can detect for itself (which means shown in Table I.
that no human intervention is needed during the learning The descriptive statistics for R, F and M are presented in
process) patterns, strong features, and correlation in the large Table II.
input data and code them in the output. The patterns of SOM in
a high-dimensional input space are originally very complicated. 7KHPD[LPXPDQG PLQLPXP UHFHQF\V¶YDOXHVLQWHUPVRI
When projected on a graphical map display, its structure, after days are calculated to be 438 and 1, the larger value
clustering, turns out to be not only understandable but more demonstrates that the customer has shopped recently. In what
transparent as well[12]. concerns the frequency, the maximum and minimum values are
calculated to be 19 and 1, respectively. For monetary, the
highest and lowest total amount spent are MAD 13731.60 and and M for different genders, where the percentage of male
MAD 69, respectively (calculated in Moroccan Dirham). customers is twice as high as that of females (67,9% versus
32,1%).
Therefore, it is necessary to further decompose the dataset
for more details. Table III present the characteristics of R, F

Fig. 1. Framework for customer segmentation based on RFM model and clustering techniques

TABLE I. AN EXAMPLE DATASET FROM CUSTOMERS, PRODUCTS AND TRANSACTIONS TABLES

Customer table

CID Gender Birth_date Marital_S address City «

00025 M 17/09/1983 1 /RW%ORF« Casa «
00107 M 03/04/1990 0 Avenue Hass... Rabat «
00115 f 11/08/1985 1 1DSW« Casa «
« « « « « « «
Product table

PID Brand Category Subcategory Description Price «

A00140 17/09/1983 6 64 VibroActio.. 209.00 «
I00077 03/04/1990 3 32 CovrePhone.. 100.00 «
A00125 11/08/1985 6 61 SlimSport.. 79.99 «
« « « « « « «
Transaction table

TID CID PID Date Qte Total «

T00044 00107 A00140 15/09/2014 1 209.00 «
T00045 00025 I00077 07/09/2014 2 200.00 «
T00046 00107 A00125 01/10/2014 1 79.99 «
« « « « « « «
 TABLE II. EXAMPLE R-F-M VALUES OF SOME CUSTOMERS According to the proposed model, we use the Self-
AFTER DATA PREPROCESSING
organizing maps and k-means methods of Tanagra to
perform cluster analysis. First step: By applying self-
organizing maps (SOM) method to cluster 730 customers,
CID Recency Frequency Monetary we found that the number eight is the best number of
00025 311 7 1600.99 clustering based on the characteristics of recency, frequency
00107 335 11 3052.74 and monetary. The result of this method is shown in Fig. 2.
00115 42 2 225.00 Later the number eight of cluster (k) generated by SOM can
« « « « be used as a parameter for the second step. In this step, K-
means method is applied to find the final solution. Therefore,
Table IV summarizes the information of these eight clusters
in terms of R, F, and M by K-means method when the
TABLE III. DATA FORM WITH FOUR ATTRIBUTES. number of clusters is set to eight.
Attribute name Data content
Gender Gender of the customers
Recent Refers to the number of days between the first day
transaction time of study period and the last day of purchase
(R)
Frequency (F) Refers to the total number of purchases
from November 2013 to January 2015
Monetary (M) Refers to the total amount spent by customers
from November 2013 to January 2015.
(Moroccan Dirhams or MAD)

TABLE IV. THE DESCRIPTIONS OF RECENCY, FREQUENCY AND

MONETARY.
Fig. 2. Eight Clusters generated by SOM technique
Variables Max Min Average Standard
deviation By observing Table IV, most of the customers are
R 438 1 296,81 113,93 grouped in clusters 6 and 7. More importantly, Cluster 7 is
F 19 1 10,16 6,17
the most important cluster because the average values of L,
M 13731,60 69,00 3947,18 3405,75
R, F and M are higher than the overall average value, which
might indicate that these customers purchase recently and
frequently with a high money spent. It is worth investing lots
of resources to sustain good relationships with these
TABLE V. CHARACTERISTICS OF R, F AND M FOR DIFFERENT customers from the online store viewpoints. In addition to
GENDERS. Cluster 7, customers in cluster 4 have high R and M values,
Description of "Gender"
which might indicate that they shopped the store recently
with high money spent. So, they could be the customers with
Gender=f Gender=m profit potential in the near future. Cluster 4 is called big
Number of [32,1 %] 234 Number of [67,9 %] 496
spender customers.
customers customers For Clusters 1 and 5, customers have the lowest values of R,
Variables Group Overral Variables Group Overral F and M compared to other clusters. In terms of recency,
these customers have recently joined the online store, and in
Average of 242,53 287,90 Average of 309,31 287,90 terms of frequency and monetary, these customers are not
R R those who spend a lot of money, and even not those who
Average of 6,99 10,16 Average of 11,65 10,16 often purchase. This cluster is called uncertain new
F F customers. In addition to Clusters 1 and 5, customers in the
Average of 2772,98 3986,63 Average of 4559,20 3986,63 Cluster 6 are new customers; the only difference between the
M M two is that customers in Cluster 6 have recently shopped at
the merchant site. They could become more important
customers if the store could increase their frequency and
monetary.
Cluster 2 has the characteristics of high F and M values but
low R value. The low R value indicates that these customers
have not shopped the store recently. Thus, the marketing
strategy is to invite them to come back. Finally, Cluster 3 and
8, the values of R and F are above the average values. They
have the characteristics of high recency, high frequency, it
can be said that these two clusters are loyal, but they do not marketing strategy is to encourage these customers to spend
have the right profile to become profitable customers, much more money.
because their contribution to the company is still low. The

TABLE VI. DESCRIPTIVE STATISTICS OF EIGHT CLUSTERS BASED ON K-MEANS METHOD.

Cluster Number of Male Female Average of R Averageof F Average of M pattern
customers
Cluster n1 89 50 39 243,68539 4,764045 1560,1137 ***

Cluster n2 96 70 26 175,41667 13,104167 3987,0856 *FM

Cluster n3 79 67 12 376,37975 17,341772 10691,82 RF*

Cluster n4 44 30 14 374,86364 5,386364 8797,2724 R*M

Cluster n5 98 38 60 67,897959 3,173469 923,72276 ***

Cluster n6 119 71 48 376,2605 4,97479 1302,9202 R**

Cluster n7 107 87 20 365,90654 17,233645 5890,529 RFM

Cluster n8 98 83 15 355,41837 14,061224 2867,7194 RF*

Total 730 497 235 287.90 10.16 3 986.63

VI. ACKNOWLEDGMENT
V. CONCLUSION
We would like to thank the online store specialized in
The main purpose of this paper is customer segmentation and electronics, fashion, home appliances, and children's items in
measuring their loyalty by combining RFM model and data Morocco for providing us with the data.
mining techniques (Clustering analysis). First, behavioral
variables, recency, frequency and monetary were obtained
using RFM model, then customers were segmented by
applying two different methods, in the first self-organizing REFERENCES
maps is applied to determine that eight might be the best
number of clusters in this study. Later, K-means method is [1] Hughes AM. Boosting response with RFM. Marketing Tools 1996;5:4-
used to classify 730 customers into eight clusters in 7.
accordance with R, F, and M variables. [2] Marakas GM. Decision Support Systems in the 21st Century, Second
The results have demonstrated that group 7 is the most Edition. Prentice Hall, Upper Saddle River, NJ, 2003.
important cluster, as the average values of R, F, and M are [3] Yang AX. How to develop new approaches to RFM segmentation.
superior to the overall average value. In summary, with the Journal of Targeting, Measurement and Analysis for Marketing
applications of RFM model and clustering techniques, the 2004;13(1):50-60.
online store can easily identify high-value and profit [4] Wang CH. Apply robust segmentation to the service industry using
potential customers and then design different marketing kernel induced fuzzy clustering techniques. Expert Syst. Appl 2010., 37:
8395-8400.
strategies to maximize its profits for different types of
clusters. >@2¶&RQQRU*& 2¶.HHIH%9LHZLQJWKHZHEDVDPDUNHWplace: the
case of small companies. Decision Support Systems, 21(3), 1997, pp. 171±
183.
>@-7:HL6</LQDQG++:X³$UHYLHZRIWKHDSSOLFDWLRQ5)0
PRGHO´$IULFDQ-RXUQDORI%XVLQHVV0DQDJHPHQWYROQRSS
4199±4206.
[7] Chang, E. C., Huang, H. C., & Wu, H. H. Using K-means method and
VSHFWUDO FOXVWHULQJ WHFKQLTXH LQ DQ RXW¿WWHU¶V YDOXH DQDO\VLV 4XDOLW\ 
Quantity, 44(4), 2010, pp 807±815.
[8] Huang, S., Chang, E. C., & Wu, H. H. A case study of applying data
mining techniques LQ DQ RXW¿WWHU¶V FXVWRPHU YDOXH DQDO\VLV  ([SHUW
Systems with Applications, 36, 2009, pp 5909±5915.
[9] Jo-Ting Wei, Shih-Yen Lin, Chih-Chien Weng, Hsin-Hung Wu..
Customer relationship management in the hairdressing industry: An
application of data mining techniques. Expert Systems with Applications,
40, 2013, pp 7513±7518.
[10] Wang, S. Cluster analysis using a validated self-organizing method:
Cases of problem identification. Intelligent Systems in Accounting, Finance
and Management, 10(2), 2001, pp 127±138.
[11] Fish, K. E., & Ruby, P. An artificial intelligence foreign market
screening method for small businesses. International Journal of
Entrepreneurship, 13, 2009, pp 65±81.
[12] Churilov, L., Bagirov, A., Schwartz, D., Smith, K., & Michael, Data
mining with combined use of optimization techniques and self-organizing
maps for improving risk grouping rules: Application to prostate cancer
patients. Journal of Management Information Systems, 21(4), 2005, pp 85±
100.
[13] Derya Birant. Data Mining Using RFM Analysis, Knowledge-Oriented
Applications in Data Mining, Prof. Kimito Funatsu (Ed.), ISBN: 978-953-
307-154-1, InTech, DOI: 10.5772/13683, 2011. Available from:
http://www.intechopen.com/books/knowledge-oriented-applications-in-
data-mining/data-mining-using-rfm-analysis.
[14] Punj, G., & Steward, D. W. Cluster analysis in marketing research:
review and suggestions for applications. Journal of Marketing Research,
20, 1983, pp 134-148.
[15] Kuo, R. J. Ho, L. M. Hu, C. M. Integration of self-organizing feature
map and K-means algorithm for market segmentation, Computers &
Operations Research, Vol.29, No.11, 2002, pp. 1475-1493.