Volume VII, n°03 (December 2021) Belhadj.

Customer Value Analysis Using Weighted RFM model: Empirical Case

Study
Tarek BELHADJ*
Abdelhafid Boussouf University Center,
Mila, Algeria
[email protected]

Received: 11/03/2021 Accepted: 12/10/2021 Published: 16/12/2021

Abstract :
The aim of this article is to provide a comprehensive methodology to perform customer
value analysis. First, we review the CLV concept, RFM model and other related
concepts. Then a case study for transactions database of a chosen company is presented.
We use Weighted RFM (recency, frequency and monetary) model to determine
customers’ lifetime value (CLV) and segment them based on k means clustering
approach. The relative weights of RFM model were determined by analytic hierarchy
process (AHP). Findings revealed that CLV analysis using RFM measures can help the
company better understand its customers and locate the most profitable customers.
Therefore, the company can apply marketing strategies to maintain customer
relationships more effectively.
Key Words: Analytical Hierarchy Process; Customer Lifetime Value; Customer
Relationship Management; K-Means Clustering; RFM model.

JEL Classification : M31, C38, C69.

* Corresponding author: Tarek Belhadj ([email protected])

Introduction:
Many studies have proved that attracting new customers is more costly than
retaining old ones (Larivière & Van den Poel, 2005; Hwang, Jung, & Suh, 2004). It
has been estimated that obtaining new customers costs five times more than
retaining an existing one (Rust & Zahorik, 1993). From the perspective of customer
relationship management, customers are not equal in terms of their values to the
company. Therefore, evaluating customers values in order to understand their
profitability and retain valuable customers became a critical parts of customer
relationship management activities (Shen & Chuang, 2009). Companies are
increasingly recognizing the importance of customer value analysis to identify
profitable customers and to develop strategies to target customers. The progress of
information technology and the growing availability of customer transaction data
enabled companies to have an important base for decision making. Such a huge
volume of data (Big Data) is useless unless implementing data analytics plans
which enable companies connecting customers to their marketing decisions
(Verhoef, Kooge, & Walk, 2016). The literature of customer relationship
management suggests diverse and rich approaches of customer value analysis.

932
Volume VII, n°03 (December 2021) Belhadj.T

Among it, Customer lifetime Value (CLV) approach based on Recency, Frequency,
Monetary (RFM) model represents an important alternative, because it requires a
very small number of variables, it is easy to understand and implement and it is
very effective in identifying valuable customers which can help the company
boosting its profit in a short time (Wei, Lin, & Wu, 2010). Recently, some
researchers proposed an enhanced Weighted RFM model. They devoted different
weights to each of RFM variables depending on business features and
characteristics of the industry (Shih & Liu, 2003; Shen & Chuang, 2009; Wei-
Jiang, Shu-Yong, Xue, & Xiao-Feng, 2011). The aim of this clustering and analytic
method is to identify the most profitable customers and enable decision makers to
allocate resources and make marketing strategy more effective.
Research questions: In view of the above discussion, the research questions are
as follows:
(i) What is customer value analysis and how is it applied?
(ii) Can RFM analysis be used to segment and identify the most profitable
customers?
(iii) What is customer CLV and how is it estimated using RFM value?
Research methodology: This article aims at implementing a CLV analysis based
on RFM model. For this purpose, the framework of the paper will be as follows.
First, we review the CLV concept, RFM model and other related concepts. Next,
the application of these concepts is presented using a case study for transactions
database of a C company. Then, the analysis outputs and result discussion are
presented. Finally, the paper ends with concluding remarks.

I. Literature Review:
This section mainly explores the study background. It reviews related studies
of customer value analysis, RFM model, classifying algorithm and Customer
Lifetime Value (CLV).
Customer value is one of the most important key concepts in customer
relationship management CRM. Today’s companies are very interested to know the
value of their deferent customers. Customer value analysis is "a kind of analytic
method for discovering customers’ characteristics and makes a further analysis of
specific customers to abstract useful knowledge from large data" (Cheng & Chen,
2009, p. 4177). From the CRM perspective, customers are not homogeneous.
Therefore, market segmentation is a necessary tool to deal with customer diversity.
It is a process of dividing customers into distinct and homogeneous groups in order
to develop differentiated marketing strategies based on their characteristics (Kadir
& Achyar, 2019). For a long time, customer segmentation models were based on
traditional criteria such as: demographic, geographic, and psychographic features
of customers. Despite its importance, traditional segmentation fails to consider
major shifts in today’s complicated business environment caused by the advance of
Information Technology. Thus, customer data collected by sophisticated
information systems and advanced analytics technique support more accurate
customer segmentation based on transactional and behavioural data (Lee & Park,

933
Volume VII, n°03 (December 2021) Belhadj.T

2005; Huseynov & Yıldırım, 2017). Methods such as Customer Lifetime Value
(CLV) analysis; Recency, Frequency and Monetary (RFM) model and Customer
Pyramid have been developed as important marketing tools that help companies
analyse the profitability of its customers and improve the customer segmentation to
customize its marketing strategies.
Over the past twenty years, several studies have incorporated RFM model
and Customer Lifetime Value (CLV) to improve customer value analysis and
develop accurate prediction and classification models. For example, Shih & Liu
(2003) presented a systematic approach in evaluating customer lifetime value
(CLV) by means of analytic hierarchy process (AHP) to determine the relative
weights of RFM variables. The proposed approach was applied to marketing
database of a hardware retailer. Clustering analysis was employed to segment
customers based on weighted RFM value. The study also discussed three
viewpoints for validating the proposed method.
Cheng & Chen (2009) with the aim to improve segmentation accuracy and
enhance customer classification rules using data-mining model, they suggested a
new procedure that includes quantitative value of RFM attributes and K-means
algorithm integrated with rough sets theory. Empirical case study was performed to
validate the proposed procedure. The finding revealed that the proposed procedure
is more efficient than the listed methods in terms of accuracy rate in classifying the
segmentation of customer value and the output of proposed procedure represent
understandable decision rules.
Kumar, Chaitanya, & Madhavan (2012) focused on clustering e-banking
customers using RFM approach model. The aim was to improve the relation
between marketing decision and customer segmentation. The study analyses
customer characteristics and behaviours with appropriated criteria: access time,
transaction access and RFM Analysis, Life Time Value of the customers (LTV),
demographic variables. The analysis procedures consisted of two phases. Firstly,
customers were segmented into clusters according to their RFM values using K-
Means clustering. Secondly, the resulted clusters were again divided into new
clusters based on demographic data. Finally, LTV analysis was used to identify
customer's profile.
He & Li (2016)proposed a new customer segmentation approach based on
three dimensions namely customer lifetime value, customer satisfaction and
customer activity. The appropriated variables were obtained using RFM model,
Kano model and BG/NBD model. The study concluded that, decision-makers can
use the output of customer segmentation to better identify market segments and
developed more effective marketing strategies.
Christy, Umamakeswari, Priyatharsini, & Neyaa (2018) have performed a
segmentation process on a transactional dataset of the customers of a company
based on RFM model. Then the clustering analysis was extended using K –Means
algorithms, Fuzzy C – Means and a new proposed algorithm RM K-Means. The
different approaches were compared with one another. The results revealed that the

934
Volume VII, n°03 (December 2021) Belhadj.T

new proposed algorithm is more effective in terms of its iterations and execution
time.
1. Customer Value Analysis:
As we have mentioned above, customer value analysis is an analytic method
for abstracting customers’ characteristics by using transaction database and then
enhance the customer relationship management. It is clear that customers are not
homogeneous in their purchasing behaviours. Therefore, they vary extensively in
the value they represent to the company (Cheng & Chen, 2009). Thus, companies
aim via using value analysis method to know their customers who spend more
money and make the most contribution in the profits. Since, the literature suggest
that the cost of customer retention is far less costly than acquisition cost, customer
value analysis is used as an effective tool to identify profitable customers thus
helping decision-makers to decide which customers to give particular marketing
strategies.
There are several methods which are used in customer value analysis. Pareto
analysis and the Pyramid model are discussed below. RFM and CLV will be
discussed further later.
1.1. Pareto analysis:
The Pareto Law, named after the well-known Italian economist and
sociologist Vilfredo Pareto (1848-1923), states that for many events, roughly 80%
of consequences come from 20% of the causes. Marketing literature follow
Pareto’s 80/20-law suggested that 20 % of customers generate 80% of revenue
(Ultsch, 2002). This can be used to prioritize customers which result optimal
output. Therefore, particular marketing strategies can be elaborated to retain the top
20% customers and boost the lowest 80% customers towards the top 20% .
1.2. Customer Pyramid :
One of the most used analysis methods of customer profitability is the
Customer pyramid. Curry & Curry (2002) introduce the customer pyramid as way
to segment clients according to revenue. The active customers are divided into four
segments: Top, Big, Medium, and Small as shown in Figure 1. The advantage of
Customer pyramid in comparison with traditional market segmentation comes from
that the pyramid model help to understand customers’ value and identify profitable
customers, therefor, attracting and retaining a company’s most valuable customers.
The pyramid model has been used and proven extremely useful to business
companies (Aggelis & Christodoulakis, 2005).
Curry & Curry (2002) suggested in their work that successful application of
the pyramid comes to those who follow three-step Customer Marketing Strategy:
(1) acquiring new customers into the pyramid; (2) boosting customers higher into
the pyramid; (3) keeping the customers in the pyramid.

935
Volume VII, n°03 (December 2021) Belhadj.T

Fig. 1: The Customer pyramid

Top
1% 80% of
Big Revenue
4%
Medium
15%
20% of
Small Revenue
80%
Inactive customers

Prospects

Suspect

The rest of the world

Source: Curry & Curry (2002, p. 9)

2. RFM Model:
2.1. RFM model definition:
The RFM model is proposed by Hughes (1994), and it is one of the most
popular customer value analysis methods. Its advantage is to extract characteristics
of customers by using fewer criterions (three variables – Recency, Frequency, and
Monetary) as cluster attributes so that reduce the complexity of model of customer
value analysis.
The definitions of RFM criterions are described as follows:
Recency (R) refers to the interval (number of days) between the last purchase
and the present time. Recency is often regarded as the most important measure
from the three criterions, because the most recent purchasers are likely to purchase
again, so they potentially create future value. Frequency (F) is the number of
transactions / purchases made by the customer in a particular period. The higher the
frequency is, the greater customer loyalty is supposed. Monetary (M) represent the
amount of money spent by a customer in a specific period of time (Wei, Lin, &
Wu, 2010; Mesforoush & Tarokh, 2013)
RFM is considered as one of the most important models used for market
segmentation that distinguish important customers and identify customer’s
purchase behaviour. With the RFM framework, customers can be grouped and
classified into segments according to their RFM scores. So, the aim of RFM
scoring is to drive better market segmentation that helps companies expect future

936
Volume VII, n°03 (December 2021) Belhadj.T

customer behaviour. Therefore, the quantification of consumer behaviour plays a

critical role in interpreting customer value through time.
2.2. Weighted RFM model:
Although the easy scheming of the RFM variables, whether by scores or
numeric values, there are two views to calculate a single RFM value. Hughes
(1994) and Stone (1995) suggested two opinions which differ, one from the other,
in regard to the weight of the three variables of RFM model. . Hughes (1994)
considered that the three variables are equal in term of their weights and they have
the same importance when calculating a combined score. In the contrast, Stone
(1995) stated that the weights of the three variables are not the same. They have
different importance depending on the product features and industry characteristics.
Thus, different weights must be given to each element of RFM for computing a
single RFM value. For instance, Stone (1995) proposed ordering the three variables
as follows: Frequency with highest weight, followed by the Recency, then
Monetary value with the lowest weighting. However, Khajvand, Zolfaghar,
Ashoori, & Alizadeh (2011) and Monalisa, Nadya, & Novita (2019) obtained an
RFM weights values that put frequency at the higher importance, followed by
monetary and then recency. Other studies found Monetary to have the most value
and Recency to have the least value (Shen & Chuang, 2009; Mesforoush & Tarokh,
2013).
Moreover, Miglautsch (2000) suggested a formula of calculating the total
RFM score as follows = (R×3) + (F×2) + (M×1). In contrast to the arbitrarily
formulas of Miglautsch (2000), Shih & Liu (2003) applied an approach to
determine the different weights of RFM variables using Analytic Hierarchy Process
(AHP). AHP is a "measurement theory through pairwise comparisons and depends
on expert judgment to get priority scales" (Saaty, 2008, p. 83).
2.3. Analytical hierarchy process:
In this study, hierarchical analysis process, one of useful tools in method
selection, is used for weighing the relative importance of RFM variables. AHP was
developed by Thomas L. Saaty in the 1970s and has been since then extensively
used for weighted scoring decision making in complicated problems where many
variables or criteria are considered in priority setting. Mainly, the AHP is based on
the subjective experience and knowledge of the decision makers to get priorities of
variables. Decision makers score the different variables using the comparison of
pairs to determine the best solution. This comparison can be done using the scale of
relative importance suggested by Saaty (2008). To determine the relative
importance of a variable, decision makers or users attribute values that vary from 1
to 9, when compared to another variable, as seen in Table 1.

937
Volume VII, n°03 (December 2021) Belhadj.T

Table 1: Scores for the importance of variable

Importance Definition of Importance Scale
Scale
1 Equally Important Preferred
2 Equally to Moderately Important Preferred
3 Moderately Important Preferred
4 Moderately to Strongly Important Preferred
5 Strongly Important Preferred
6 Strongly to Very Strongly Important Preferred
7 Very Strongly Important Preferred
8 Very Strongly to Extremely Important Preferred
9 Extremely Important Preferred
Source: Taherdoost (2017, p. 245)
The AHP is performed in three main steps (Shih & Liu, 2003; Shen &
Chuang, 2009) as follow:
(1) Perform pairwise comparisons: evaluators (decision makers) were asked
to do paired comparison of the relative importance of RFM variables, and give the
value of 1 to 9 to each indicator using the scale as shown in Table 2.

Table 2: Sample AHP Questionnaire

Idicator Idicator
Rcency 9 . 7 . 5 . 3 . 1 . 3 . 5 . 7 . 9 Frequency

Monetray 9 . 7 . 5 . 3 . 1 . 3 . 5 . 7 . 9 Frequency

Monetray 9 . 7 . 5 . 3 . 1 . 3 . 5 . 7 . 9 Rcency

(2) Assess the consistency of pairwise judgments: expert evaluators may

make inconsistent judgments when making pairwise comparisons. Before the
weights are computed based on the pairwise judgments, the degree of inconsistency
is measured by the inconsistency index that should be of less than less than 0.1.
Otherwise the pairwise judgments may be revised before the weights of RFM are
computed.
(3) Computing the relative weights: this step determines the weight of each
decision element according to the pairwise comparisons.
3. Cluster analysis:
Clustering is a statistical method that was used for for classifying physical or
abstract items into homogeneous segments where the items have similar
characteristics. A cluster is a group of data items that are homogeneous and related
to the items within this the cluster and , at the same time, are heterogeneous and
unrelated to the items in other clusters (Mesforoush & Tarokh, 2013). Cluster

938
Volume VII, n°03 (December 2021) Belhadj.T

analysis is largely used for market segmentation and has been applied in meany
studies along with RFM model.
The literature suggested several methods which are used in clustering such as
TwoStep algorithm and K-means algorithm. The latter is one of the well-known
algorithms for cluster analysis. The K-means algorithm has two major features.
Firstly, it is based on the mean value of the objects in the cluster, i.e., the algorithm
assigns each item to the cluster with the nearest centroid (mean). Secondly, K-
means clustering technique requires predetermination of the number of clusters by
the decision maker or the user (Cheng & Chen, 2009; Mesforoush & Tarokh, 2013;
Kadir & Achyar, 2019).
1.4. Customer Lifetime Value:
CLV is developed from the literature of customer relationship management
(CRM). The aim of CRM is to build and maintain strong relationships with
customers and to generate higher customer lifetime value to the company (Estrella-
Ramón, Sánchez, Swinnen, & VanHoof, 2013) . CLV can be defined as "the
present value of all future profits obtained from a customer over his or hers
relationship with a firm" (Gupta, et al., 2006, p. 141). It is essentially used as a
basic method to compute and express customer profitability to develop marketing
strategies to target the most profitable customers
Gupta, et al. (2006) disscussed six implementable modeling approaches that
are useful for CLV estimating: RFM Models, Probability Models, Econometric
Models, Persistence Models, Computer Science Models and Diffusion/Growth
Models. The most powerful and simplest model to implement CLV may be the
RFM model – Recency, Frequency, and Monetary value. Gupta, et al. (2006)
states that, despite the limitations of RFM-analysis the models remain a mainstay
of the industry because of their ease of implementation in practice. Fader, Hardie,
& Lee (2005) showed how the well-known RFM (recency, frequency, and
monetary value) model can be used with customer lifetime value (CLV) to build a
model that overcomes many of its limitations. Therefore, Several studies use RFM
model to estimate Customer lifetime value (Miglautsch, 2000; Shih & Liu, 2003;
Sohrabi & Khanlari, 2007).
One of the most significant benefit of CLV based RFM model is that the
model inputs are nothing more than each customer’s RFM values derived from the
same using transaction data. Moreover, there is no need to split the data into two
(or more) time periods, the model use the entire customer base as a single sample
to estimate the CLV (Gupta, et al., 2006). Calculating the CLV for all of a firm’s
customers allows them to categorize customers based on their individual
contribution to the organizations profits This helps to develop strategies to deal
with each customer differently, instead of treating every customer the same way
using the same marketing approaches. Although firms are interested in knowing the
current and predicted customer life time value of their customers, they also need to
identify the factors they can control that can potentially increase the value of
customer. It is not enough to know who are the most profitable customers, it is

939
Volume VII, n°03 (December 2021) Belhadj.T

even more important to determine how to convert currently less profitable

customers into more profitable ones (Savoie, 2014).

II. Study Design:

The case study of this article is about (C) company, an Algerian spare parts
wholesaler that provides automotive parts and accessories for Asiatic brands. The
company is based in Algiers and has two branches, one in Aïn M'lila and the other
in Touggourt. This company implemented a registration system of transaction
information of customers. The database of consumer transactions contains
information about name and customer ID, date of purchase, total price. For the
purpose of this study, six months of data on consumer transactions have been
selected (from 2019/07/01 to 2019/12/31 ). It represents 218 customers and 1542
purchase records. RFM values of the 218 customers are extracted from this
particular marketing database to measure the customers’ CLV following the
proposed framework shown in figure 2:

Fig. 2: The research framework

Customer Preparing and

transactions
preprocessing data
data

Extracting the RFM Pareto Analysis

variables

Clustering RFM / Data normalization

K-means method Min-max method

Weight of RFM
The customer pyramid
Indicators
AHP method

Clustering WRFM /
K means method

CLV ranking

The research steps can be described as follow:

 Data Pre-processing: First of all, we collect purchase information from records
to get accurate data for analysing. There are 218 raw data of customer

940
Volume VII, n°03 (December 2021) Belhadj.T

transactions of the given case. In each Raw, there is a unique customer with
her/his ID, number of transactions, latest transaction date and total amount of
money spent in all of her/his purchases at the end of the study period. Therefore,
each customer is assigned with three values for recency, frequency, and
monetary variables.
 Pareto Analysis: in this step, we divide customers into 5 equal parts (quantiles).
Then, we test the 20/80 principle (20% of customers contribute to 80% of total
revenue
 Segment customers by RFM model: The simple sum of RFM values are used
to clustering customers using K-Means algorithm. The resulting clusters
represent customer segments to be compared with the customer pyramid
introduced by Curry & Curry (2002). Furthermore, these customer segments
will be analysed in order to identify their different patterns. The RFM values of
each cluster is examined whether above or below the overall average value.
Based on that, marketing strategy according to cluster pattern can be suggested
to have more targeted marketing action.
 Data normalization: In order to increase the cohesion of entry types and
simplify the analysis the RFM data must be normalised. Min-max normalization
method is used. It performs a linear transformation on the original data and
makes it lie between 0 and 1 thus bringing all the values RFM model in the
dataset to a common scale.
 Calculating The Weight of RFM Indicators: this step employs the AHP to
evaluate the relative weight (importance) of each RFM variable, and
specifically asks decision makers to make intuitive judgments about ranking
order to make pairwise comparisons.
 Weighted RFM clustering: K-means clustering is then employed to group
customers according to the weighted RFM values.
 CLV ranking: Finally, the simple weighted sum of normalised RFM values are
used to derive CLV ranking and thus customer segments can be identified and
compared clearly.

III. Result Analysis:

1. Pareto Analysis:
By conducting Pareto analysis, customers are arranged in descending order
based on monetary value (from the best to the worst). The number of customers is
then divided into equal quintiles (20%). The results show that 74,48% of sales
volume is generated by 20% of the company customers and 80% of customers only
contribute to 25,52% of sales. This is a natural trend which is consistent with
“Pareto Principle,” states that approximately 80% of all effects come from roughly
20% of the causes. It is obvious that “Pareto’s principle” represents a basic rule in
classifying customers based on their profitability. However, the company needs to
more identifying its main customers in order to make marketing strategy more
effective and achieve the result with minimum effort and cost.

941
Volume VII, n°03 (December 2021) Belhadj.T

2. RFM evaluation:
In this respect, RFM is considered as one of the most important models used
for customer classification that help to identify important customers and their
purchase behaviour by three indicators Recency (R), Frequency (F), and Monetary
(M). Recency is the interval between the customer’s last transaction and the present
time reference. Since the R value –by this definition- negatively impacts the RFM
score, a reversed form is used to get more consistence. The new R is defined as the
number of days between the first date concerned (1/07/2019) and the date of the
last customer purchase. For example a date of last purchase on 20/09/2019 is
represented by R=82, while a date on 10/12/2019 is represented by R=163.
Frequency (F) refers to the number of transactions in a particular period of time.
Monetary (M) is the total amount of money spent by the customer over the same
period of time. To calculate the RFM Score we have used the formula:
RFM Score = R+F+M.
Table 3 illustrates a part of data input on which RFM Score is calculated:
Table 3: Sample data of RFM for each customer
Customer Recency Frequency Monetary (DZD) RFM Score
No (day)
00001 135 14 1 076 830,00 1 076 979,00
00002 174 7 45 580,00 45 761,00
00003 167 17 101 740,00 101 924,00
00004 46 2 111 500,00 111 548,00
……… .… .… …...…..… ……….…
……… .… .… …...…..… ……….…
00218 62 7 60 450,00 60 519,00
The next step is to classify customers using the K-means method. This
method requires pre-specifying the number of clusters (k). Since we follow the
pyramid model, the parameter k is set to 4. We use the K-means clustering
algorithm of SPSS to cluster the RFM values. The result of clustering analysis is
represented in table 4.
Table 4 shows four clusters with their corresponding number and percentage
of customers and the average of recency, frequency and Monetary values within
each cluster.

942
Volume VII, n°03 (December 2021) Belhadj.T

Table 4: The results of RFM K-means clustering for 218 customer

Clusters Nu. of Percent Recency Frequency Monetary Pattern
Customers (days) (DZD)
1 6 2,8 135,00 11 1221726,67 R↑ F↑
M↑
2 3 1,4 150,67 11 3878320,00 R↑ F↑
M↑
3 183 83,9 126,89 7 82389,61 R↓ F↓
M↓
4 26 11,9 131,85 7 568996,54 R↑ F↓
M↑
Total / 218 100 128 7,07 224020,78
Average

3. The customer pyramid:

Following the work of Curry & Curry (2002), we have made a comparison
between the above clustering results in table 2 and Curry's Customer Pyramid. The
comparison result is shown below:
Fig. 3: Clustering results and Customer Pyramid comparison

Top
Cluster2 (1,4%)
1%

Big
Cluster1 (2,8%) 4%
Spending
amount
Cluster4 (11,9%) Medium
15%

Small
Cluster3 (83,9%)
80%

Inactive customers

As seen in Figure 3, the similarity of the four clusters as derived from the K-
means algorithm to the pyramid model is apparent. From the total of 218
customers 183 (83.9% compared to 80%) are included in small segment, 26 (11,9%
compared to 15%) in medium segment, 6 (2,8% compared to 4%) in big segment
and 3 (1,4% compared to 1%) in top segment.
Furthermore, the resulting clusters represent customer segments to be
analysed in order to identify their different patterns. The last line in table 2 also

943
Volume VII, n°03 (December 2021) Belhadj.T

shows the total number of customers and the overall average RFM values for all
customers. The last column of the table also shows the RFM pattern of the four
clusters. The RFM pattern presents a comparison between the average RFM values
of each cluster and the total averages. The comparisons results can be obtained by
assigning upward or downward arrow, depending on whether the average RFM
values of a cluster is less or higher than the total average.
The pattern of cluster 3 is characterised by low RFM values (R↓ F↓ M↓).
This means it has average recency value (the interval between the customer’s last
transaction and the first date concerned) below the overall average, beside
frequency and monetary less than the overall average. This cluster includes
customers who had earlier visited the company and made very few transactions
with low monetary value. These customers generally make occasional buying and
switch companies. To attract such customers, the company should reduce prices
and provide extra services. As a result, the company reduces margins and achieves
less profit from these customers. So, the cluster 3 includes valueless customers.
Cluster 4 with the pattern (R↑ F↓ M↑) has average frequency below the
overall average along with recency and monetary higher than the overall average.
This cluster may include new customers who have recently dealt with the company
and make important monetary amount. The amount of spending shows that they are
tending to gradually increase their purchase frequency. Therefore, the company
have to strengthen its relationships with this cluster of customers. They should have
special pull marketing strategy to create an ongoing relationship with the company.
Finally, customers in clusters 1 and 2 represent the same pattern. They are
characterised by high RFM values (R↑ F↑ M↑) that exceeds the overall average.
They represent loyal customers who have a long-term relationship with the
company. They have a high purchase frequency and an important contribution to
company profitability. Clusters 1 and 2 include high valuable customers. Since,
the marketing practices indicates that retaining an existing customer is easier,
cheaper and more valuable than attracting a new one, the company should develop
marketing strategies to increase customer loyalty in this clusters.
4. Estimating CLV for clusters: To calculate CLV for each cluster, weighted
RFM method is used according to the assessments obtained by the AHP. Before
that, the RFM scores retrieved from the original database are normalized by the
min-max normalization before estimating CLV for clusters.
 Data normalization: In the RFM model, R, F, and M have different units. In
order to facilitate the analysis the RFM data must be normalized. Min-max
method is used to map indicators to the interval [0, 1]. The normalization formula
is (Sohrabi & Khanlari, 2007) as follows:
𝑅 − 𝑅𝑚𝑖𝑛 𝐹 − 𝐹𝑚𝑖𝑛 𝑀 − 𝑀𝑚𝑖𝑛
𝑅′ = , 𝐹′ = , 𝑀′ =
𝑅𝑚𝑎𝑥 − 𝑅𝑚𝑖𝑛 𝐹𝑚𝑎𝑥 − 𝐹𝑚𝑖𝑛 𝑀𝑚𝑎𝑥 − 𝑀𝑚𝑖𝑛

Where R', F', and M’ represent recency, frequency, and Monetary values after
normalization. R, F, and M represent the original recency, frequency, and monetary
values. Rmax, Fmax, and Mmax are the maximum values of R, F and M for the all

944
Volume VII, n°03 (December 2021) Belhadj.T

customers. Rmin, Fmin, and Mmin are the minimum values of R, F and M for the
all customers. All the normalized values are between 0 and 1 and the distribution
maintains the same. This process was used also in other studies, where normalized
RFM values of each customer were then multiplied by the relative importance of
RFM variable, wR, wF and wM (Shih & Liu, 2003; Shen & Chuang, 2009).
 Weight of Indicators: To assess the relative importance of RFM variables,
hierarchical analysis process is used. So, a company's manager and an academic
expert were asked to do paired comparison, and give the value of 1 to 9 to each
indicator (see table 2). Using AHP software, the assessments obtained for the
relative weights of the RFM variables are mentioned in table 5. The inconsistency
value of 0,086 is less than 0.1. The results are reliable.
Table 5: Relative Weight of RFM
Variable R F M
Weight values 0,128 0,087 0,785

 CLV ranking: To rank the CLV values we proceed the following steps. The
RFM values of each customer are normalised, as described above. The K-means
method is then applied to cluster the customers into four groups, according to the
weighted RFM values. Then, the CLV value of each cluster is calculated based on
the following formula:
CLV= WR* R’ + WF* F’+ WM* M’
So, the normalised RFM values of each cluster are multiplied by the relative
importance of RFM variable, wR, wF and wM, which are determined by the AHP.
The value obtained for each cluster is shown in Table 6.
After that, the CLV ranking of the clusters is determined according to the
CLV value of each cluster.
The results in table 6 show that cluster 1 has the highest ranking, it includes
the most valuable customers; followed by cluster 3; then cluster 2 in third rank;
whereas cluster 4 has the lowest ranking, this later may include the most valueless
customers. According to CLV ranking of each cluster, managers of the company
have to develop different strategies to attract and maintain customers.
Table 6: The results of RFM K-means clustering for 218 customer
Clusters Recency Frequency Monetary CLV CLV
(days) (DZD) ranking
1 0,831 0,208 0,974 0,889 1
2 0,844 0,114 0,049 0,156 3
3 0,845 0,768 0,060 0,222 2
4 0,381 0,082 0,027 0,077 4
Total 0,705 0,126 0,055 - -
Average
The aim of cluster analysis is to get low similarity (maximising variance)
between the different segments while getting high similarity (minimising variance)
within these segments. Therefore, analysis of variance (ANOVA) is used to
945
Volume VII, n°03 (December 2021) Belhadj.T

examine whether RFM variables significantly discriminates between the four

cluster. The result show significant p-values (p<0.01) indicating that RFM
variables can use to distinguish between these four clusters.

Conclusion:
The development of information technology enables companies to collect
large and important information about their customers. Companies, particularly
with increasing sophistication in modelling, can benefit significantly by analysing
customer transaction data to determine the most profitable customers and
accordingly, adopt the appropriate marketing strategies. Customer lifetime value
analysis represents a comprehensive model of customer profitability in customer
relationship management (CRM) research. Many methods have been developed to
evaluate customer lifetime value.
This article follows the recency, frequency and monetary (RFM) model.
Customers are divided into different segments by means of clustering method
based on their lifetime value expressed in terms of RFM. The customers data for an
Algerian automotive spare parts company has been presented as case study to
demonstrate how customer value analysis can be performed using CLV based RFM
model. We normalised the extracted RFM values using min-max normalization
method and K-Means algorithm was used to cluster customers into four segments
(clusters) based on RFM attributes. Then, we employed the analytic hierarchy
process (AHP) to evaluate the relative weights (importance) of the RFM variables.
Finally, the weighted RFM values were used to calculate customer lifetime values
and derive CLV ranking and thus the resulted clusters can be compared clearly and
used essentially to explain marketing decisions and CRM strategies by the
company.
The paper concluded that a company's adoption of customer value analysis
would help better in developing more effective marketing decisions.

Bibliography:
 Aggelis, V., & Christodoulakis, D. (2005). Customer clustering using rfm
analysis. In Proceedings of the 9th WSEAS International Conference on
Computers (p. 2). 1-5.
 Cheng, C. H., & Chen, Y. S. (2009). Classifying the segmentation of customer
value via RFM model and RS theory. Expert systems with applications, 36(3),
4176-4184.
 Christy, A. J., Umamakeswari, A., Priyatharsini, L., & Neyaa, A. (2018). RFM
ranking – An effective approach to customer segmentation. Journal of King
Saud University –Computer and Information Sciences. (In press).
 Colombo, R., & Jiang, W. (1999). A stochastic RFM model. Journal of
Interactive Marketing, 13(3), 2-12.

946
Volume VII, n°03 (December 2021) Belhadj.T

 Curry, A., & Curry, J. (2002). The customer marketing method: How to
implement and profit from customer relationship management. Simon and
Schuster.
 Estrella-Ramón, A. M., Sánchez, P. M., Swinnen, G., & VanHoof, K. (2013). A
marketing view of the customer value: Customer lifetime value and customer
equity. South African Journal of Business Management, 44(4), 47-64.
 Fader, P. S., Hardie, B. G., & Lee, K. L. (2005). RFM and CLV: Using iso-
value curves for customer base analysis. Journal of marketing research, 42(4),
415-430.
 Gupta, S., Hanssens, D., Hardie, B., Kahn, W., Kumar, V., Lin, N., . . . Sriram,
S. (2006). Modeling customer lifetime value. Journal of service research, 9(2),
139-155.
 He, X., & Li, C. (2016). The research and application of customer segmentation
on E-commerce Websites. In 2016 6th International Conference on Digital
Home (ICDH) (pp. 203-208). IEEE.
 Hughes, A.M. (1994). Strategic database marketing.Chicago:Probus Publishing.
 Huseynov, F., & Yıldırım, S. O. (2017). Behavioural segmentation analysis of
online consumer audience in Turkey by using real e-commerce transaction data.
International Journal of Economics and Business Research, 14(1), 12-28.
 Hwang, H., Jung, T., & Suh, E. (2004). An LTV model and customer
segmentation based on customer value: a case study on the wireless
telecommunication industry. Expert systems with applications, 26(2), 181-188.
 Kadir, M. A., & Achyar, A. (2019). Customer Segmentation on Online Retail
using RFM Analysis: Big Data Case of Bukku. id. ICEASD 2019, April 01-02,
Indonesia.
 Khajvand, M., Zolfaghar, K., Ashoori, S., & Alizadeh, S. (2011). Estimating
customer lifetime value based on RFM analysis of customer purchase behavior:
Case study. Procedia Computer Science, 3, 57-63.
 Kumar, M. V., Chaitanya, M. V., & Madhavan, M. (2012). Segmenting the
banking market strategy by clustering. International Journal of Computer
Applications, 45(17), 10-15.
 Larivière, B., & Van den Poel, D. (2005). Predicting customer retention and
profitability by using random forests and regression forests techniques. Expert
systems with applications, 29(2), 472-484.
 Lee, J. H., & Park, S. C. (2005). x Intelligent profitable customers segmentation
system based on business intelligence tools. Expert systems with applications,
29(1), 145-152.
 Mesforoush, A., & Tarokh, M. J. (2013). Customer profitability segmentation
for SMEs case study: network equipment company. International Journal of
Research in Industrial Engineering, 2(1), 30-44.
 Miglautsch, J. R. (2000). Thoughts on RFM scoring. Journal of Database
Marketing & Customer Strategy Management, 8(1), 67-72.

947
Volume VII, n°03 (December 2021) Belhadj.T

 Monalisa , S., Nadya, P., & Novita, R. (2019). Analysis for customer lifetime
value categorization with RFM model. Procedia Computer Science, 161, 834-
840.
 Rust, R., & Zahorik, A. (1993). Customer Satisfaction, Customer Retention and
Market Share. Journal of Retailing, 69(2), 193-215.
 Saaty, T. L. (2008). Decision Making With The Analytic Hierarchy Process. Int.
J. Services Sciences 1 (1). 83–98.
 Savoie, P. P. (2014). A study of customer profitability in automotive
retail.Master of Business Administration. (N. Saint Mary's University (Halifax,
Ed.)
 Shen, c. c., & Chuang, H. M. (2009). A study on the applications of data mining
techniques to enhance customer lifetime value. WSEAS Transactions on
Information Science and Applications, 6(2), 319-328.
 Sheshasaayee, A., & Logeshwari, L. (2017). An efficiency analysis on the TPA
clustering methods for intelligent customer segmentation. . In: 2017
International Conference on Innovative Mechanisms for Industry Applications
(ICIMIA), Bangalore, pp. 784–788.
 Shih, Y. Y., & Liu, C. Y. (2003). A method for customer lifetime value
ranking—Combining the analytic hierarchy process and clustering analysis.
Journal of Database Marketing & Customer Strategy Management, 11(2), 159-
172.
 Sohrabi, B., & Khanlari, A. (2007). Customer lifetime value (CLV)
measurement based on RFM model. Iranian Accounting & Auditing Review,Vol.
14 No. 47, pp 7- 20.
 Spring, P., Leeflang, P., & Wansbeek, T. (1999). The combination strategy The
combination strategy to optimal target selection and offer segmentation in direct
mail. Journal of Market-Focused Management, 4(3), 187-203.
 Stone, B. (1995). Successful Direct Marketing Methods, Lincoln-wood. IL:
NTC Business Books, 29-35.
 Taherdoost, H. (2017). Decision making using the analytic hierarchy process
(AHP); A step by step approach. International Journal of Economics and
Management Systems, 2.
 Ultsch, A. (2002). Proof of Pareto’s 80/20 law and Precise Limits for ABC-
Analysis. Data Bionics Research Group University of Marburg/Lahn, Germany.
1-11.
 Verhoef, P., Kooge, E., & Walk, N. (2016). Creating value with big data
analytics: Making smarter marketing decisions. Routledge.
 Wei, J. T., Lin, S. Y., & Wu, H. H. (2010). A review of the application of RFM
model. African Journal of Business Management, 4(19), 4199-4206.
 Wei-Jiang, Shu-Yong, D., Xue, Y., & Xiao-Feng. (2011). Determination of
customer value measurement model RFM index weights. African Journal of
Business Management, 5(14), 5567-5572.

948