Handbook of Research on Novel Soft Computing Intelligent Algorithms:

finickyontarioAI and Robotics

Oct 29, 2013 (3 years and 10 months ago)

89 views

Handbook of Research
on Novel Soft Computing
Intelligent Algorithms:
Theory and Practical
Applications
Pandian M. Vasant
PETRONAS University of Technology, Maylaysia
Volume I
A volume in the Advances in
Computational Intelligence and Robotics
(ACIR) Book SeriesManaging Director: Lindsay Johnston
Production Manager: Jennifer Yoder
Publishing Systems Analyst: Adrienne Freeland
Development Editor: Allyson Gard
Assistant Acquisitions Editor: Kayla Wolfe
Typesetter: Christina Henning
Cover Design: Jason Mull
Published in the United States of America by
Information Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue
Hershey PA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: cust@igi-global.com
Web site: http://www.igi-global.com
Copyright © 2014 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in
any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or
companies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark.
Library of Congress Cataloging-in-Publication Data
Handbook of research on novel soft computing intelligent algorithms : theory and practical applications / Pandian Vasant,
editor.
volumes cm
Includes bibliographical references and index.
Summary: “This book explores emerging technologies and best practices designed to effectively address concerns inherent
in properly optimizing advanced systems, demonstrating applications in areas such as bio-engineering, space exploration,
industrial informatics, information security, and nuclear and renewable energies”--Provided by publisher.
ISBN 978-1-4666-4450-2 (hardcover) -- ISBN 978-1-4666-4451-9 (ebook) -- ISBN 978-1-4666-4452-6 (print & perpetual
access) 1. Soft computing--Industrial applications. 2. Intelligent control systems. 3. Intelligent agents (Computer soft-
ware). I. Vasant, Pandian.
QA76.9.S63H35 2014
006.3--dc23
2013017606

This book is published in the IGI Global book series Advances in Computational Intelligence and Robotics (ACIR) (ISSN:
2327-0411; eISSN: 2327-042X)
British Cataloguing in Publication Data
A Cataloguing in Publication record for this book is available from the British Library.
All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the
authors, but not necessarily of the publisher.1
Chapter 1
Customer Relationship
Management and Data Mining:
A Classification Decision Tree
to Predict Customer Purchasing
Behavior in Global Market
Niccolò Gordini
University of Milan – Bicocca, Italy
Valerio Veglio
University of Milan – Bicocca, Italy
ABSTRACT
In the global market of today, Customer Relationship Management (CRM) plays a fundamental role
in market-oriented companies to understand customer behaviors, achieve and maintain a long-term
relationship with them, and maximize the customer value. Moreover, the digital revolution has made
information easy and fairly inexpensive to capture. Thus, companies have stored a large amount of data
about their current and potential customers. However, this data is often raw and meaningless. Within
the CRM framework, Data Mining (DM) is a very popular tool for extracting useful information from
this data and for predicting customer behaviors in order to make profitable marketing decisions. This
research aims to demonstrate the classification decision tree as one of the main computational data
mining models able to forecast accurate marketing performance within global organizations. Particular
attention is paid to the identification of the best marketing activities to which firms should concentrate
their future marketing investments. The criteria is based on the loss functions that confirm the accuracy
of this model.
DOI: 10.4018/978-1-4666-4450-2.ch001
Copyright © 2014, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.Customer Relationship Management and Data Mining
INTRODUCTION dissemination of intelligence horizontally and
vertically within the organization, and organization
Globalization draws new competitive boundaries, wide action or responsiveness to it”. Day (1994)
modifying the traditional concepts of time and simply states that market orientation represents
space (Brondoni, 2008a, 2008b, 2008c). In the superior skills in understanding and satisfying
global market, companies have crossed spatial customers. In a later study, Slater and Narver
and temporal borders (both regional and cultural), (1994) highlight that a company is market-oriented
selling their products to customers located every- when “its culture is systematically and entirely
where and have been aided greatly by the digital committed to the continuous creation of superior
revolution that has permitted them to interact with customer value”. Deshpandé et al. (1993) define
a large amount of data. customer value “as the set of beliefs that puts the
In this new, global, and highly competitive customer’s interest first, while not excluding those
(D’Aveni, 1994) arena, a company has to think in of other stakeholders such as owners, managers,
terms of market-orientation and not only in terms and employees, in order to develop a long-term
of product orientation and marketing orientation profitable enterprise.” Specifically, “this entails
in order to achieve competitive advantage. If com- collecting and coordinating information on cus-
petitive advantage was once based on structural tomers, competitors, and other significant market
characteristics such as market power, today the influencers to use in building that value” (Slater
emphasis has shifted to capabilities that enable a & Narver, 1994). Thus, the heart of market ori-
company to consistently deliver superior value to entation is its customer focus. In addition, today’s
its customers (Slater & Narver, 1994). As a result, customers have such varied needs and preferences
the concept of market-orientation (Kohli & Jawor- that it is not possible to group them into large
ski, 1990; Narver & Slater, 1990) was developed, homogenous populations to develop marketing
which emphasized the establishment of effective strategies (Shaw et al., 2001). In fact, each customer
information processes and capabilities within the wants to be served according to his individual and
firm to understand the expressed and latent needs unique needs. Hence, marketing decisions based
of customers, thus making firms more efficient on traditional segmentation approaches and not
and effective in managing customer relationships on a market orientation, result in a poor response
(Boulding, 2004). A bulk of research (Day, 1994; rate and increased costs. Slater and Narver (1994)
Gordini, 2010, 2012; Kohli & Jaworsky, 1990; suggest that market oriented companies have to
Jaworski & Kohli, 1993; Narver & Slater, 1990; understand cost and revenue dynamics of not
Slater & Narver, 1994, 1998, 1999) has high- only current customers but also of future target
lighted that a market-orientation provides a strong buyers. Therefore market-oriented companies are
foundation to reach superior customer value and committed to understanding both the expressed
competitive advantage. Among the most com- and latent needs of their customers (Day, 1994;
monly cited, Narver and Slater (1990) stated that Gordini, 2010, 2012; Kohli & Jaworsky, 1990;
“market orientation consists of three behavioral Slater & Narver, 1994, 1998, 1999) as reacting
components (customer orientation, competitor only to customers’ expressed needs is usually in-
orientation, and inter-functional coordination) adequate for the creation of competitive advantage
and two decision criteria (long-term focus and in a global market. Certainly, market-oriented
profitability).” Jaworski and Kohli (1993) sug- companies do not ignore the expressed needs
gest that market-orientation is “the organization of their customers, but they have to realize that,
wide generation of market intelligence pertain- since competitive advantage is often temporary,
ing to current and future needs of the customers, the firm must understand how customers’ needs
2Customer Relationship Management and Data Mining
are evolving and develop innovative solutions through filtering, integrating, extracting or format-
to such needs (D’Aveni, 1994). Therefore, the ting customer data. Transforming customer data
greater the market orientation of the firm, the into customer information, companies use various
greater the proportion of its activities oriented information systems”. One of the most important
towards understanding latent needs. Developing a is data mining. Within the CRM framework, data
long-term relationship with customers is the best mining techniques are a very popular tool for
way to know the customers’ expressed and latent extracting useful information from enormous
needs and to enable them to become loyal (Dowl- customer databases. In fact, data mining tools can
ing, 2002). Consequently, customer relationship help companies to uncover the hidden informa-
management becomes a fundamental concept in tion about the expressed and latent needs of their
market-oriented companies in order to analyze customers and, hence, to understand the customer
and understand customer behaviors, to acquire and better, channeling this information into effective
retain customers, to achieve and maintain a strong, marketing strategies. Thus, the application of data
long-term relationship with current and potential mining in CRM is justified in market-oriented
customers, to maximize customer value and, companies. Despite its relevance, data mining and
consequently, to obtain a competitive advantage. its application to CRM have not been, however,
Simultaneously at this shift of paradigm and at the thoroughly studied.
statement of the market orientation, there was an The aim of this study is to examine the con-
explosion of available customer data. In fact, the cepts of customer relationship management and
advent of information technology and the digital data mining, their relationship, and to test the
revolution has made data easy to capture and fairly prediction power of the data-mining technique in
inexpensive to store, transforming the way market- estimating the probability of customer conversion
ing is done and how companies manage data. As in global competitive companies. In the follow-
a result, companies, especially global companies, ing paragraph we make a brief overview of the
have realized that these huge amounts of data concepts of CRM and data mining, illustrating
based on their current and potential customers, their relationship and analyzing the advantages
suppliers, business partners and competitors, are and disadvantages along with the challenges
key factors to support important marketing deci- deriving from that relationship. The next three
sions, and they have started to collect and store sections (data and methodology, findings, future
them in their own databases. The main problem research directions) show the empirical analysis
for the companies is how to understand which conducted on the effectiveness of data mining in
of this data could contain competitive informa- estimating the probability of customer conversion.
tion and, hence, relevant knowledge in creating A concluding section discusses and summarizes
strategic marketing decision-making. In fact, this the results of our research.
data remains often raw and meaningless, and is
stored in databases without transforming it into
meaningful information to create customer value BACKGROUND
and profit for the company. According to Belbaly
et al. (2007) there are strong reasons to convert Customer Relationship Management
raw data into information and to spread this in-
formation throughout the companies in order to The globalization of markets, characterized by the
develop knowledge useful in the decision-making lack of spatial-temporal boundaries, a fast and high
process. In order to do that, Buchnowska (2011) technological development, and a even increas-
suggests: “customer information is obtained ing number of competitors, obliged companies to
3Customer Relationship Management and Data Mining
renew their strategies in order to adapt themselves defines CRM as an integrated series of customer-
to the new competitive arena. In this even more oriented IT solutions (Stone & Woodcock, 2001).
competitive, complex and saturated market, the Finally, the third perspective defines CRM in a
digital revolution has made data easy to capture more strategic and holistic way that highlights
and fairly inexpensive to store. the management of the customer relationship to
The large amount of data available and the new create value for both the firm and the customers.
database technologies have enabled companies Within this perspective, Swift (2001) defines CRM
to gain the knowledge on who are the customers, as an “enterprise approach to understanding and
what and when they bought (expressed needs), and influencing customer behavior through meaning-
even predictions on what they would like to buy in ful communications in order to improve customer
the future (latent needs). Thus, many companies acquisition, customer retention, customer loyalty,
have collected and stored a large amount of data and customer profitability”. Parvatiyar and Sheth
about their current and potential customers in their (2001) consider CRM as “a comprehensive
database and customer relationship management strategy and process of acquiring, retaining, and
has become a fundamental tool to compete in partnering with selective customers to create su-
today’s global and highly competitive markets. perior value for the company and the customer.
According to Rigby and Bilodeau (2009), CRM It involves the integration of marketing, sales,
was the fourth most used tool in 2008. CRM is a customer service, and the supply chain functions
business model that dynamically integrates sales, of the organization to achieve greater efficiencies
marketing and customer expressed and latent and effectiveness in delivering customer value”.
needs in order to help firms to build long term, Scott (2001) suggests that CRM is “a set of business
profitable relationships with current and poten- processes and overall policies designed to capture,
tial customers (Ling & Yen, 2001), to manage retain and provide service to customers”. Chang
this relationship in an organized way (Xu et al., et al. (2002) analyzed the importance of CRM in
2002), and to add value for both the company and enhancing the ability of a firm to obtain and retain
its customers. The CRM concept emerged first in key customer, while Hansotia (2002) suggests that
the practitioner community (especially vendor “at the heart of CRM is the organization’s ability
community) in the mid-1990s and later in the to leverage customer data creatively, effectively
academic community. It is not a new concept in and efficiently to design and implement customer-
the academic literature, being a development of focused strategies”. Kincaid (2003) views CRM
another well-known marketing concept, relational as “the strategic use of information, processes,
marketing, defined by Reinartz and Kumar (2003) technology and people to manage the customer’s
as “the establishment and maintenance of long- relationship with your company (Marketing,
term buyer–seller relationships”. Although CRM Sales, Services, and Support) across the whole
has become widely recognized as an important customer life cycle”. According to Injazz and
marketing tool, there is not a unique definition Karen (2004), CRM is “a coherent and complete
of this concept in the literature (Khanna, 2001; set of processes and technologies for managing
Kim, 2006; Ling & Yen, 2001; Ngai, 2005; Ngai relationships with current and potential customers
et al., 2009; Payene & Frow, 2005; Parvatiyar & and associates of the company, using the market-
Sheth, 2001; Stone & Woodcock, 2001; Swift, ing, sales and service departments, regardless of
2001). According to Payne and Frow (2005), the channel of communication”. Payne and Frow
CRM can be defined from three perspectives. (2005) state that “CRM is a strategic approach that
The first perspective defines CRM in a narrow is concerned with creating improved shareholder
way, as a specific technology solution project value through the development of appropriate
(Khanna, 2001). The second perspective, instead, relationships with key customers and customer
4Customer Relationship Management and Data Mining
segments. CRM unites the potential of relation- Thus, it is possible to quibble about the specific
ship marketing strategies and IT to create profit- wording used in all the prior definitions, but we
able, long-term relationships with customers and have to identify the basic element of this concept.
other key stakeholders. CRM provides enhanced Therefore, in this paper, according to Boulding et
opportunities to use data and information to both al. (2005), and to Payne and Frow (2005), CRM
understand customers’ needs and create value with relates to strategy, the management of the dual
them. This requires a cross-functional integrating creation of value (for both firm and customers), the
of processes, people, operations, and marketing intelligent use of data and technology, the acquisi-
capabilities, enabled through information, tech- tion of customer knowledge and the diffusion of
nology, and applications”. Panagiotis et al. (2007) this knowledge to the appropriate stakeholders, the
suggest that “CRM optimizes values as profitabil- development of profitable long-term relationships
ity, revenue and customer satisfaction (what and with specific customers and/or customer groups,
why) by organizing around customer segments, and the integration of processes across the many
fostering customer satisfying behaviours and areas of the firm and across the network of firms
implementing customer-centric business models that collaborate to generate customer value. What
(how)”. Therefore, CRM should enable greater emerges from all these definitions is that CRM re-
customer insight, increased customer access and quires a “cross-functional integration of processes,
more effective customer interactions (outcomes)”. people, operations, and marketing capabilities that
Ngai et al. (2009) state that to analyze and under- is enabled through information, technology, and
stand “customer behaviors characteristics is the applications” (Payne & Frow, 2005). Thus, CRM
foundation of a competitive CRM strategy, so goes beyond a customer focus. Not only does CRM
as to acquire and retain potential customers and build relationships and use systems to collect and
maximize customer value”. Finally, Feng (2012) analyze data, but it also includes the integration
suggests that the “goals of CRM are, on the one of all these activities across the firm, linking
hand, to attract and retain more customers by these activities to both firm and customer value,
providing more quickly good quality service, on extending this integration along the value chain,
the other hand, to reduce costs through the overall and developing the capability of integrating these
management of business process”. activities across the network of firms in order to
As seen, many definitions of CRM exist. collaborate and generate customer value, while
These plethora of definitions, ranging from CRM creating shareholder value for the firm.
being the implementation of specific technology From an architecture point of view, the CRM
solutions to the implementation of an integrated can be classified into operational and analytical
series of customer-oriented technology solution, (Berson et al., 2000, He et al., 2004; Teo et al.,
to a holistic approach of managing customer 2006). According to He et al. (2004) operational
relationships simultaneously creating value for CRM includes all activities concerning the direct
both the customer and the firm, have caused some customer contact, such as campaigns, hotlines or
confusion. Parvatiyar and Sheth (2001) have noted customer clubs. Every CRM activity is generally
that a prerequisite for a new concept to merge into implemented in one of the following three activi-
an established field is to institute an acceptable ties: marketing, sales or service, since these are the
definition that captures all the major aspects of activities concerned with direct customer contact
the concept itself. In fact, the way a concept is (ECCS, 1999). According to He et al. (2004)
defined is not merely a semantic question directly analytical CRM refers to the analysis of customer
affecting the way a company accepts and uses the characteristics and behaviors so as to support the
concept itself. organizations customer management strategies and
5Customer Relationship Management and Data Mining
to accomplish operational CRM activities, with who are being lost to the competition and
respect to the customers’ needs and expectations how they can be won back (Kracklauer et
Skinner (1990). Thus, the idealistic goal is to pro- al., 2004). This phase includes two main
vide all information necessary to create a tailored elements: target customer analysis and
cross-channel dialogue with each single customer customer segmentation. Target customer
on the basis of his or her actual reactions (Arndt analysis involves seeking the profitable
& Gersten, 2001). As such, analytical CRM could segments of customers through analysis
help an organization to better discriminate and of customers’underlying characteristics,
more effectively allocate resources to the most whilst customer segmentation involves
profitable group of customers. the subdivision of an entire customer base
Many organizations have collected and stored into smaller customer groups or segments,
a wealth of data about their current customers, consisting of customers who are relatively
potential customers, suppliers and business similar within each specific segment (Woo
partners. However, the inability to discover valu- et al., 2005).
able information hidden in the data prevents the • Customer Attraction: This phase follows
organizations from transforming these data into the customer identification. In fact, after
valuable and useful knowledge (Berson et al., identifying the segments of potential cus-
2000). Data mining tools are a popular means tomers, companies can direct effort and
of analyzing customer data within the analytical resources into attracting these customer
CRM framework. The application of data mining segments. Elements of customer attraction
tools in CRM is an emerging trend in the global are developing new product, advertising,
economy. Analyzing and understanding customer direct marketing (Cheung et al., 2003; He
behaviors and characteristics is the foundation of et al., 2004; Liao and Chen, 2004; Prinzie
the development of a competitive CRM strategy, and Poel, 2005).
so as to acquire and retain potential customers • Customer Retention: This is the core con-
and maximize customer value. Appropriate data cern for CRM. In fact, it is not always true
mining tools, which are good at extracting and that building many relationships is always
identifying useful information and knowledge better, instead, building the right and long-
from enormous customer databases, are one of term relationship is critical. Thus, custom-
the best supporting tools for making different er satisfaction, which refers to the compari-
CRM decisions (Berson et al., 2000). As such, the son of customers’expectations with his or
application of data mining techniques in CRM is her perception of being satisfied, becomes
worth pursuing in a customer-centric economy. an essential condition for retaining custom-
According to Swift (2001), Parvatiyar and ers (Kracklauer et al., 2004). Elements of
Sheth (2001), Kracklauer et al. (2004) and Ngai customer retention are, for example, one-
et al. (2009), CRM consists of four dimensions: to-one marketing and loyalty programs
(Chen et al., 2005; Jiang & Tuzhilin, 2006;
• Customer Identification: Customer iden- Kim & Moon, 2006).
tification or, in other words, customer ac- • Customer Development: This phase aims
quisition is the first phase of CRM. This at increasing the number of transaction
phase involves targeting the population intensity, transaction value and individual
who are most likely to become customers or customer profitability. Metrics of custom-
most profitable to the company and, at the er development include customer lifetime
same time, it involves analysing customers value analysis, up/cross selling and market
6Customer Relationship Management and Data Mining
basket analysis. Customer lifetime value companies overspend on marginal customers”. In
analysis is defined as the prediction of the a CRM paradigm, a core object is to define dif-
total net income a company can expect ferent resource allocations for different tiers of
from a customer (Drew et al., 2001; Etzion customers, where the customer’s tier membership
et al., 2005; Rosset et al., 2003). Up/Cross depends on the economic value of that customer
selling refers to promotion activities, which or segment to the firm (Zeithaml et al., 2001).
aim at increasing the number of associated In addition to the theoretical evolution, a pre-
or closely related services that a customer requisite so as a concept is really applicable and
uses within a firm (Prinzie & Poel, 2006). useful in any marketing activity is that it should
Market basket analysis aims at maximiz- demonstrably enhance firm performance. (Bould-
ing the customer transaction intensity and ing et al., 2004; Lehmann, 2004; Rust et al., 2004).
value by revealing regularities in the pur- Several studies analyze this topic using various
chase behaviour of customers (Aggarval research methods and different measures of firm
& Yu, 2002; Brijs et al., 2004; Carrier & performance. Firstly, these studies demonstrate
Povel, 2003; Chen et al., 2005; Giudici & that CRM leads to a process of dual-creation of
Passerone, 2002; Kubat et al., 2003). value for both the customers and the firm (Boulding
et al., 2005; Levitt, 1960, 1969; Payne & Frow,
Finally, analyzing the efficiency and the 2005; Rogers, 2005; Vargo & Lusch, 2004). The
effectiveness of these four phases, companies dual creation of customer and firm value is a core
should recognize that the relationships evolve concept in CRM because the emphasis is not only
with distinct phases (Dwyer et al., 1987; Reinartz on how to sell a product, but rather on how to
et al., 2004), firms interact with customers (Re- create value for the customers and, consequently,
inartz et al., 2004; Srivastava et al., 1998), and for the firm. In fact, according to Levitt (1960,
the distribution of relationship value to the firm 1969), one of the main idea in marketing is that, if
is not homogeneous (Mulhern, 1999; Niraj et al., a firm want to grow up, it should focus on fulfill-
2001; Reinartz et al., 2004). ing expressed and latent needs of the customers.
Firstly, according to Dwyer et al. (1987), CRM According to Boulding et al. (2005) it is essential
process should recognize that relationships evolve that the firm “develops measures that are directly
with distinct phases and, consequently, relation- connected with this value dual-creation process,
ships cannot be viewed as multiple independent enabling the firm to understand the drivers of
transactions; rather, the interdependency of the value and thus to ensure long-term success”.
transactions creates its own dynamic over time. Therefore, through the CRM, the firms normally
In other words, CRM processes are longitudinal obtain not only final measures such as profit or
phenomena. shareholder value, but also intermediate measures
Secondly, firms should interact with custom- such as customer-life time value and acquisition
ers and manage relationships differently at each and retention costs, which relate to the value dual-
stage (Srivastava et al., 1998). In fact, the main creation process. As Boulding et al. (2005) state,
goal of CRM is to manage the various stages of “good CRM process measures provide the firm
the relationship systematically and proactively. with the opportunity to gain deeper insights into
Finally, the value of each relationship to the how these intermediate process measure link to
firm is not homogeneous (Mulhern 1999; Niraj downstream firm performance” such as profit or
et al., 2001). According to Reinartz et al. (2004) shareholders value. Secondly, CRM helps firms to
“a common finding is that best customers do not begin to treat marketing costs as firm investments
receive their fair share of attention and that some (Rust et al. 2004). Furthermore, this implies that
7Customer Relationship Management and Data Mining
marketing could gain a central role in managing (2005) elaborates a process that identifies dynamic
a key asset of the firm as the customer asset. Fi- customer behavior, thus helping firms to create
nally, these studies also showed that CRM has a a pricing scheme that increase long-term profits.
positive impact on performance of firm operating Thomas and Sullivan (2005) using a corporate
in different industry sectors. This is an important database develop a CRM model that allows the
result because the success and the positive impact firm to modify its communication tool depending
of CRM on firm performance is not contingent to on where customers live and how they shop and,
a particular industry, but it is usefully applicable to consequently, to increase its profit. The authors
all industry sectors. Analyzing some of these stud- also examine the dual creation of value from the
ies, Ryals (2005), using a case study approach and firm perspective proposing a process that enables
measuring the costs and the revenues associated firms to migrate customers into more profitable
with CRM to assess overall profit, shows that one channels.
of the business units analyzed was able to achieve
a 270% increase in business unit profit above target Data Mining
by implementing several CRM measures. In addi-
tion, she shows that firms reduce their attention to The origin of data mining goes back to the first
customers after they determine that they are not storage of data on computers, increases with im-
able to garner enough value from these customers. provements in data access, until today technology
Thus, for certain customers, value is taken away allows users to navigate through data in real time.
so that firms can increase the value they receive. The digital revolution has made a huge amount
Mithas et al. (2005) using a multiform database of data easy to capture. Nowadays, information
demonstrate that the use of CRM is associated with and knowledge are strategic and indispensable
increased customer knowledge and, consequently, prerogatives in search for competitive advantage
with greater customer satisfaction. Srinivasan and and decision-making with time getting shorter
Moorman (2005) using a cross-sectional database and shorter. Therefore, so as CRM can really use
show that firms that invest more in CRM have this enormous amount of data it has a need for the
greater customer satisfaction than other firms. appropriate tools. In fact, most organizations have
Jayachandran et al. (2005) suggest that firms that built up massive databases about their customers
use CRM obtain performance in term of retention and their purchase transactions. But, this data is
and customer satisfaction greater that firms that often raw data, and raw data is often meaningless
not implement CRM. Cao and Gruca (2005), using and rarely of direct benefit. According to Berson
data collected within a single firm over time and et al. (2000), the inability to discover valuable
its customers, focus their attention on acquiring information hidden in these sets of data prevents
the right customer in order to develop a specific the companies from transforming this data into a
CRM model to increase the firm performance. valuable and useful knowledge base, thus a wealth
The authors provide a framework whereby the of customer information is permanently hidden
firm can better limit its target market to customers and unutilized in such databases. The true value
who both want to hear about the firm’s particular of this data is predicated on the ability to extract
offer and qualify for that offer. As a result, the information useful for decision support or explo-
firm does not send messages to customers who ration, and on understanding the phenomenon
are unlikely to respond, thus minimizing the governing the data source. Therefore, companies
disturbance to these customers. This leads to an have a need for adequate statistical techniques to
obvious win-win situation for the firm and its analyze and transform this data into useful knowl-
customers (i.e., the dual creation of value). Lewis edge. Traditional statistical techniques and data
8Customer Relationship Management and Data Mining
management tools are no longer adequate for this knowledge from enormous customer databases.
purpose. All this has prompted a requirement for In fact, analyzing and understanding customer
using soft computing data analysis methodologies, behaviors and characteristics are the foundation of
which could discover useful knowledge from data. the development of a competitive CRM strategy,
Under these conditions, several companies have so as to acquire and retain potential customers
adopted data mining tools to monitor funding, and maximize customer value. Within the CRM,
client consumption, prevent fraud and foreseeing data mining can be seen as a business driven
customer behaviors (Galvão & Marin, 2009). In process aimed at the discovery and consistent
fact, data mining tools can help uncover hidden use of profitable knowledge from organizational
knowledge in large datasets and understand the data (Ling & Yen, 2001). It can be used to guide
customer better. As a consequence, the support of decision-making and forecast the effects of de-
corporate decision making through data mining cisions. For instance, data mining can increase
has received increasing interest and importance in the response rates of the marketing campaign by
operational research. Progress in technology and segmenting customers into groups with different
storage capacity has enabled the accumulation characteristics and needs; it can predict how likely
of customer data, inducing large, rich datasets an existing customer is to take his/her business to
of heterogeneous scales. On the one hand, the a competitor (Carrier & Povel, 2003).
enhanced data has created particular challenges Data mining has formed a branch of soft com-
in transforming attributes of different scales into a puting techniques (Liao, 2012; Mitra et al., 2002)
mathematically feasible and computationally suit- and assists companies to discover and extract
able format. On the other hand, this has advanced the hidden knowledge in large volumes of data
the application of data mining methods like deci- (Ahmed et al., 2004; Berson et al., 2000; Lejeune,
sion trees (DT), artificial neural networks, genetic 2001). In fact, data mining is an interdisciplinary
algorithms (GAs) and support vector machines field with a general goal of predicting outcomes
(SVM), capable of mining large datasets. and uncovering relationships in data. It uses
According to relevant literature, Turban et al. automated tools employing sophisticated algo-
(2007) define data mining as “the process that uses rithms to discover hidden patterns, associations,
statistical, mathematical, artificial intelligence anomalies and/or structures from large amounts
and machine-learning techniques to extract and of data stored in data warehouses or other infor-
identify useful information and subsequently gain mation repositories. Thus, data mining is a tool
knowledge from large databases”. Berson et al. to give meaning to the data and it is actually part
(2000), Lejeune (2001), Ahmed (2004) and Berry of a larger process called “knowledge discovery”
and Linoff (2004) define data mining in a similar which describes the steps that must be taken to
way as the process of extracting or detecting hid- ensure meaningful results. Although data min-
den patterns or information from large databases. ing techniques are used in several areas such as
In addition, scholars (among others, Langley & fraud detection, bankruptcy prediction, medical
Simon, 1995; Lau et al., 2003; Su et al., 2002) diagnosis, and scientific discoveries, their use for
agree that with comprehensive customer data, marketing decision support highlights unique and
data mining can provide business intelligence to interesting issues such as real-time interactive mar-
generate new opportunities. According to Berson keting, customer profiling, cross-organizational
et al. (2000) appropriate data mining tools, appear management of knowledge, and CRM. It should
to be one of the best supporting tools for making be clear from the discussion previously that CRM
different CRM decisions as they are efficient at is a broad topic with many layers, one of which
extracting and identifying useful information and is data mining, and that data mining is a method
9Customer Relationship Management and Data Mining
or tool that can aid companies in their quest to 4. Forecasting: It estimates the future value
better understand and manage raw data (Rygielski based on a record patterns. It deals with
et al., 2002). continuously valued outcomes (Ahmed,
The DM development process involves various 2004; Berry & Linoff, 2004). It relates to
models and, within each model, different statistical modelling and the logical relationships of the
techniques in order to make possible the extrac- model at some time in the future. Demand
tion of new knowledge. According to various forecast is a typical example of a forecasting
researchers (Ahmed, 2004; Carrier & Povel, 2003; model. Common tools for forecasting include
Mitra et al., 2002; Shaw et al., 2001; Turban et al., neural networks and survival analysis.
2007), CRM can be supported by different data 5. Regression: It is a kind of statistical estima-
mining models that generally include the follow- tion technique used to map a data item to
ing seven: (1) Association; (2) Classification; (3) a real-valued prediction value (Carrier &
Clustering; (4) Forecasting; (5) Regression; (6) Povel, 2003; Mitra et al., 2002). It includes
Sequence discovery; (7) Visualization. curve fitting, prediction, modeling of causal
relationships, and testing scientific hypoth-
1. Association: It aims to establishing relation- eses about relationships between variables.
ships between items, which exist together in a Common tools for regression include linear
given record (Ahmed, 2004, Jiao et al., 2006; regression and logistic regression.
Mitra et al., 2002). Market basket analysis 6. Sequence Discovery: It is the identification
and cross selling programs are typical ex- of associations or patterns over time, like
amples for which association modelling is time-series analysis (Berson et al., 2000;
usually adopted. Carrier & Povel, 2003; Mitra et al., 2002).
2. Classification: It is one of the most common According to Mitra et al. (2002) “the goal is
learning models in data mining (Ahmed, to model the states of the process generating
2004, Berry & Linoff, 2004; Carrier & Povel, the sequence or to extract and report devia-
2003). It aims at classifying a data records tion and trends over time”. Common tools
into one of several predefined classes based for sequence discovery are statistics and set
on certain criteria and, consequently, at theory.
building a model to predict future customer 7. Visualization: It refers to the presentation of
behaviours (Ahmed, 2004; Berson et al., data so that users can view complex patterns
2000; Chen et al., 2003; Mitra et al., 2002). (Shaw et al., 2001). It is used in conjunction
Common tools used for classification are with other data mining models to provide a
neural networks, decision trees and if-then- clearer understanding of the discovered pat-
else rules. terns or relationships (Turban et al., 2007).
3. Clustering: It aims at segmenting a het- Examples of visualization model are 3D
erogeneous population into a number graphs, “Hygraphs” and “SeeNet” (Shaw
of more homogenous clusters based on et al., 2001).
similarity metrics or probability density
models. (Ahmed, 2004, Berry & Linoff, Each of this type of data mining models uses
2004; Carrier & Povel, 2003; Mitra et al., various statistical techniques. The most used in
2002). It is different to classification in that CRM are association rule, neural networks, genetic
clusters are unknown, are no predefined at algorithms, fuzzy logic, and decision tree.
the time the algorithm starts. Common tools Association rules regard the discovery of
for clustering include neural networks and association relationships, which are above an in-
discrimination analysis. teresting threshold, hidden in databases (Berry &
10Customer Relationship Management and Data Mining
Linoff, 2004, Brijs et al., 2004; Ngai et al., 2009; genetics and evolutionary principle (Davis, 1991;
Wang et al., 2005). The threshold tells how strong Etemadi et al., 2009; Holland, 1975; Goldberg,
the pattern is and how likely the rule is to occur 1989; Shin & Lee, 2002; Varetto, 1998). GAs
again (Berson et al., 2000). This tecnique can be are based on three fundamental steps: selection
used to build a model for predicting the value of of better individuals, crossover and mutation.
a future customer (Wang et al., 2005). The selection usually starts from a population
Fuzzy logic (Elamvazuthi et al., 2010; 2012; of randomly generated individuals and happens
Feng & Yuan, 2011; Madronero et al., 2010; in generations. In each generation, the fitness of
Vasant, 2006; Vasant et al., 2010; Vasant et al., every individual in the population is evaluated,
2011) is a mathematic theory that imitates the multiple individuals are stochastically selected
human ability of making decisions in environ- from the current population based on their fit-
ments of uncertainties and inaccuracy. Through ness, and modified, recombined and possibly
Fuzzy logic, intelligent systems of control and randomly mutated to form a new population. The
decision support can be built (Han & Kamber, new population is then used in the next iteration
2006). According to (Galvão & Marin, 2009) of the algorithm (Gordini, 2013). With the evo-
the fuzzy logic can be used mainly in two forms. lution of the algorithm, only the solutions with
Firstly, fuzzy logic could represent the classic higher prevision power survive, until they reach
logic extension for a more flexible one, aiming an ideal solution.
at making formal inaccurate concepts. Secondly, Decision tree (DT) is a technique can be
fuzzy sets are applied to several theories and used to extract models describing sequences of
technologies to process inaccurate information, interrelated decisions or predicting future data
such as in decision-making processes. trends (Berry & Linoff, 2004; Chen et al., 2003;
Artificial neural network (ANN) is an artificial Kim et al., 2005). It classifies specific entities
intelligence technique introduced by McCulloch into particular classes based upon the features of
and Pitts in 1943. It mimics the biological neural the entities: a root is followed by internal nodes,
network of the human nervous system. Thus, the each node is labeled with a question, and an arc
basic idea of ANNs is that it learns from examples associated with each node covers all possible
using several constructs and algorithms just like responses (Buckinx et al., 2004; Chen et al.,
a human being learns new things. Unlike Fuzzy 2003). Graphically, decision tree is represented
sets, It is effective in function approximation, by branches, similar to a tree (Han & Kamber,
forecasting, classification, clustering and opti- 2006). Each branch of the tree represents a deci-
mization tasks depending on the neural network sion about a variable that determines how the data
architecture (Berry & Linoff, 2004; Mitra et al., present division to a series of branches. Thus, it
2002; Turban et al., 2007). describes an association between the attribute and
Genetic algorithms (GAs), developed by the target variable or, in other words, the associa-
Holland in 1975, mimic Darwinian principles of tion of each branch with other branches (Galvão
natural selection and evolution to solve nonlinear, & Marin, 2009). The aim of the induction of a
non-convex global optimization problem. (Armin Decision Tree is to produce an accurate prediction
& Babak, 2011; Gaby et al., 2010; Ganesan et al., model or discover the predictive structure of the
2011, 2012; Leng et al., 2012; Pinkey et al., 2011; problem. Decision tree has several advantages: it
Provas Kumar & Dharmadas 2012; Svancara et is simple to understand and interpret, it has value
al., 2012; Vasant, 2012, 2013). GAs are stochastic even with little hard data, possible scenarios can
search techniques that are able to seek out large be added, worst, best and expected values can be
and complicated spaces on the ideas from natural determined for different scenarios, unlike ANNs
11Customer Relationship Management and Data Mining
it uses a white box model and it can be combined product, positioning) and “external” (competition,
with other decision techniques. A variety of dif- demographic) factors, which help to determine
ferent DT paradigms have been developed such the customer consumptions, customer satisfaction
as ID3, C4.5, CART or CHAID. and corporate profits. Third, it provides a link
between individual transactions and analytical
Customer Relationship systems. Thus, the relationship between CRM
Management and Data Mining and DM enables the company to manage a vast
amount of data and to discover hidden affiliations
The application of data mining tools in CRM is an with high business value.
emerging trend in the global economy that have Data mining has also some disadvantages.
many advantages, but also some disadvantages and It is clear that the role of technology is relevant
some interesting challenges for both researchers throughout the CRM process but it cannot alone
and practitioners (especially for marketers). be sufficient for building a profitable and lasting
The most important advantages are the fol- relationship. Past experiences showed that these
lowing. First, DM is a fundamental tool for the misunderstanding were often penalizing for the
firms to manage and organize data because it companies. According to Veglio (2013), some of
helps companies to begin a cleaning process that the main disadvantages can be summarized as:
eliminates errors and ensures consistency. In fact,
companies have to organize raw customer data • Drawing the Customer-Centric Marketing
that must be transformed into information. The Strategies after the implementation of the
data preparation is a critical, vital process for the CRM process. Decision-makers believe
success of CRM because the data comes from that implementing CRM software is equiv-
various sources. In fact, companies are able to alent to creating marketing strategies, as
obtain data about their customers both from inside a first stage company must formulate this
sources (massive databases that contain market- strategy and clarify the purpose for this.
ing, human resources and financial data) and • Assuming that more CRM technology is
from outside sources (i.e. they can purchase data better. Many executives also mistakenly
from external consultant companies). According believe that CRM is a technology-inten-
to Bergeron (2001), data mining allows firms to sive product and are apt to put emphasis
obtain a large amount of detailed, complete, homo- on new functions of CRM software. When
geneous information about customers organized a company begins to use a CRM software
in a database, to know how to segment customers, package, it is very important for them to
differentiating profitable customers from those narrow down the specifications of the soft-
who are not, and to establish appropriate business ware in order to minimize the burden on its
plans for each case, to reach a deeper knowledge users and to suppress bugs. If a company
of expressed and latent needs and perceptions of concentrates excessively on new functions,
the customers in real time, and consequently to this will cause false integration of the CRM
better discriminate and more effectively allocate software and existing system.
resources to the most profitable group of cus- • Focus on customers. Organizations must
tomers and reach lower costs, to reach a greater establish contact only with individuals who
customer satisfaction and improve and extend have a real interest in their company and
customer relationships, generating new business product. When they approach the wrong
opportunities. Second, in order to create high people, they can be perceived as stalkers
value for companies, DM focuses the attention on and lose potential customers.
consumers in respect to both the “internal” (price,
12Customer Relationship Management and Data Mining
• Cost of the Tools. The price of the data multiple memberships. A marketer may also want
mining software is really high because of to use multiple memberships to gain important
their complexity in terms of specific algo- knowledge about customers, instead of simplify-
rithms and models that are implemented. ing the classes and losing valuable information.
For this reason, firms sometimes cannot But, according to Spangler (1999), current data
develop data mining analysis but just sta- mining techniques have been shown to be limited
tistical analysis. in handling memberships for multiple classes.
• The link between data mining software and Therefore, marketers need reliable classification
campaign management. In the past this tools. A third challenge is Web mining. Accord-
link was mostly non-automatic. It required ing to Shaw et al. (2001), in past years the Web
that physical copies of the scoring from the has become an important and convenient tool for
data model to be created and transferred purchasing goods and, consequently, is a source
to the database. This separation of data of customer data, which is very usefulness for the
mining and marketing management soft- marketers. However, the multiple data formats
ware introduces considerable inefficiency and distributed nature of knowledge on the Web
and was prone to human error. Today the makes it a challenge to collect, discover, organize
trend is to integrate the two components and manage, in a manner that is useful for mar-
in order to gain a competitive advantage. keting decision support. Therefore, web mining
Businesses can gain a competitive ad- needs to be addressed as an important marketing
vantage by ensuring that their data min- knowledge management issue.
ing software and campaign management Finally, the choice of data mining model and
software share the same definition of the of the more accurate techniques depends on the
customer segments in order to model the data available and the business requirements. In
entire database (Thearling, 1998). fact, the choice and the development of an effective
and efficient data mining model that aids in the
The relationship between CRM and data min- measurement of customer value and in the inter-
ing also presents interesting challenges for both action with heterogeneous expressed and latent
researchers and practitioners (especially for mar- needs of the customers increases the prediction
keters) alike. A first challenge regards the need to accuracy rate of potential customers and, at the
manage data that crosses organizational boundar- same time, decreases the possibility to commit a
ies and is distributed across supply chain partners. Type I and Type II error. According to Reinartz et
Customer knowledge is typically distributed across al. (2004), companies want to avoid the mistake of
supply chain partners, and marketing is an impor- not identifying a good customer and subsequently
tant beneficiary of this knowledge. But, manag- not rewarding the customer accordingly (Type I
ing the cross-organizational knowledge requires error) and, at the same time, companies also want
organizational and industry level efforts. Further to prevent wrongful classification of low-value
research should analyze and develop appropriate customers as high-value customers and subsequent
inter-organizational CR models, protection of over-spending of resources (Type II error).
knowledge rights, and distribution of knowledge In this paper we use a decision tree classi-
benefits amongst partners. A second challenge fication model to predict customers purchasing
is when customers can belong to more than one behavior in global market and reduce both Type
category. The even more complex customer pref- I and Type II error. Classification is one of the
erences make this issue particularly relevant for most common learning models in data mining
marketers, as they may encounter customers with (Ahmed, 2004; Berry & Linoff, 2004; Carrier &
13Customer Relationship Management and Data Mining
Povel, 2003). It aims at building a model to predict channels provided the database. Currently, this
future customer behaviors through classifying firm operates in many different countries around
database records into a number of predefined the world such as Europe, South America, Asia,
classes based on certain criteria (Ahmed, 2004; Africa and Australia. We cannot give other details
Berson et al., 2000; Chen et al., 2003; Mitra et al., about the company for privacy reasons.
2002). Common techniques used for classifica- A data mining explorative analysis is used to
tion are neural networks, if-then-else rules, and accomplish our research goal. The entire database
decision trees (Chen et al., 2003). According to contained more than 1,463,199 potential custom-
Gehrke et al. (1999), Quinlan (1987), Ravi et al. ers and an initial set of 42 variables related to their
(2008), Ravi Kumar et al. (2007) and Rud (2000) purchase behavior. Table 1 provides a description
decision trees form a part of machine learning, of the variables.
an important area of artificial intelligence. A vast The database, showed in Table 1, is not ag-
number of algorithms are used for building deci- gregated by potential customer. It could be useful
sion tree including CART, Chi squared automatic to aggregate the dataset by “user id” (unique code
interaction detection (CHAID), Quest and C5.0. for each potential customer). In other words, at
For the purpose of our research, a classification de- each “user id” will correspond a different poten-
cision tree model based on the “CART” algorithm tial customer. Therefore, in Table 2, we proceed
and on Gini Impurity was the best data mining to aggregate variables and we show the aggrega-
model. The main reason of this choice is that the tion criteria used.
“CART” algorithm is one of the most popular Finally, in order to select only those variables
criteria used to manage business problems within with both the greatest prediction capacity and the
global organizations (Breiman et al., 1984) and it lowest correlation level, we decided to eliminate
is relatively simple to interpret by decision makers, not significant variables in the dataset. To do so,
while C4.5 and C5.0 algorithms are inadequate in we adopted a two-stage variables selection process.
solving business problems because of widely used In the first stage, we decided to eliminate a vari-
in engineering research (Giudici, 2010). able relying on the opinion of the Executive Vice
President (EVP) of the company. Table 3 shows
the results of this first stage.
DATA AND METHODOLOGY In the second stage, in order to select only
those variables with the lowest correlation level,
The main purpose of this research is to demon- we carried out a multicollinearity analysis. Mul-
strate the strategic predictive power of the data ticollinearity analysis is a good method for dis-
mining models in forecasting punctual marketing covering redundant variables. Literature identifies
performance in global competitive companies. different causes of multicollinearity including,
Special focus is paid to the identification of the among others, the improper use of dummy vari-
best marketing drivers, which lead potential cus- ables, the use of a variable that is computed from
tomers in a customer state achieving to increase other variables in the equation, or the use of the
the probability of customer conversion. The data same or variable twice. Or, simply, it may just be
analyzed concerns the launching of a quarterly that variables actually are highly correlated (Lat-
online marketing campaign. The first part of the tin et al. 2003). In this study we use the Variance
campaign was proposed in December 2010, while Inflation Factor (VIF) method (Montgomery &
the second part was in January and February 2011. Peck, 1992) as indicator of multicollinearity.
A global marketing consultant company that offers Although in literature there is not a general rule
digital data driven solutions across all interactive about the interpretation of the VIF value, high
14Customer Relationship Management and Data Mining
Table 1. Description of the variables in the dataset
Variables
Variables Description
Measures
Activity Timestamp Timestamp for the activity on the advertiser’s website Nominal
Activity tag name associated to the action that the potential customer performed on
Activity Tag name Nominal
the client (company) site
Advertisement Name Advertisement name associated with the exposure Nominal
Type of Banner Type of banner proposed to the potential customer Nominal
Name of the Advertiser Name of the advertiser Nominal
Amount of Model Conver- Amount of conversion attributed to a specific potential customer in a journey based
Scale
sion on the results of the model
Quantity associated with the activity. Quantity can typically be 1 representing the
Activity Quantity Scale
activity but in some cases this will be > 1
Revenue Activity Revenue associated with the activity Scale
Cost per Click. It is an internet advertising model used to direct traffic to websites,
Cost per Click Scale
where advertisers pay the publisher when the advertisement is clicked
Click Through Rate. It is a way of measuring the success of an online campaign
for a particular product or service. The click through rate advertisement is defined
Click Through Rate Scale
as the number of clicks on an advertisements divided by the number of times the
advertisement is shown
Average Position The average position of a search term in the search engine Scale
Brand Search The search term includes any of the branded terms Scale
Name of the Campaign Name of the campaign Nominal
Creative Height Creative Height Scale
Creative Type Creative Type Scale
Creative Width Creative Width Scale
Creative Name Creative for display advertisement Scale
Head Flag Search flag used to mark high volume keywords Nominal
Timestamp for impression
Timestamp for impressions and clicks Nominal
and clicks
Impression or Click ‘Imp’ if the event is an impression, ‘Click’ if the event is a click Nominal
Keywords Advertising
Group of Advertising Keyword. Nominal
Group
Keywords Campaign Keywords Campaign Nominal
Keywords Category Keywords Category Nominal
Keywords Name Keywords on which search advertisements appear in the web Nominal
Match Type Type of key word in form users Nominal
Max Search Click If 1 then includes search if 0 then no search Scale
Price Paid Price paid by the consumer for the purchase Scale
Quantity Sold Number of items sold Scale
Purchases Number of items purchased by potential customers Scale
Min Search Click If 1 is only search, if 0 then includes display Scale
Single Conversion Activity Flags journeys where the potential customer only has a single conversion/activity Scale
continued on following page
15Customer Relationship Management and Data Mining
Table 1. Continued
Variables
Variables Description
Measures
Site Placement Placement on the Site (homepage) Scale
Rank0 Variable no identified from the company Scale
Auto incrementing value representing all touches points of a potential customer
Rank1 Scale
journey. Does not reset on a conversion/activity
Auto incrementing value representing all touches points of a potential customer
Rank2 Scale
journey. Does reset on a conversion/activity
Auto incrementing value representing the conversion number for the user. The same
Rank3 value will repeat across all exposures leading to an activity and then will increment Scale
and repeat for the next set of exposures leading to a conversion/activity
Record Number Record number Scale
Potential customers search on the search engine information related to the marketing
Search Engine Name Nominal
campaign
Search Click Represents an exposure that is a search click Nominal
Client specific field for this report. For instance shows a segment applied to users
Segment Nominal
based on location/energy consumption
Site Name Site (for display advertisement) Nominal
User Id ID of the potential customers Scale
Table 2. Variables aggregation criteria
New
Variable
Variables Aggregation Criteria Variable
Measures
Measures
Activity Timestamp Categorization into groups and creation of dummy variables Nominal Scale
Advertisement Name Categorization into groups and sum Nominal Scale
Type of Banner Categorization into groups and sum Nominal Scale
Name of the Advertiser Categorization into groups and sum Nominal Scale
Cost per Click Average Value Scale Scale
Click Through Rate Average Value Scale Scale
Average Position Average Value Scale Scale
Brand Search Categorization into groups and sum Scale Scale
Name of the Campaign Categorization into groups and sum Nominal Scale
Creative Height Categorization into groups Scale Scale
Creative Type Categorization into groups Scale Scale
Creative Width Categorization into groups Scale Scale
Head Flag Categorization into groups and sum Nominal Scale
Timestamp for impression and
Categorization into groups and creation of dummy variables Nominal Scale
click
Impression or Click Categorization into groups and sum Nominal Scale
Keywords Advertising Group Categorization into groups and sum Nominal Scale
continued on following page
16Customer Relationship Management and Data Mining
Table 2. Continued
New
Variable
Variables Aggregation Criteria Variable
Measures
Measures
Keywords Campaign Categorization into groups and sum Nominal Scale
Keywords Category Categorization into groups and sum Nominal Scale
Keywords Name Categorization into groups and sum Nominal Scale
Match Type Categorization into groups and sum Scale Scale
Max Search Click Sum Scale Scale
Quantity Sold Sum Scale Scale
Purchases Sum Scale Scale
Min Search Click Categorization into groups and sum Scale Scale
Rank 1 Maximum Value Scale Scale
Rank 2 Maximum Value Scale Scale
Rank 3 Maximum Value Scale Scale
Search Engine Name Categorization into groups and sum Nominal Scale
Search Click Categorization into groups and sum Nominal Scale
Segment Categorization into groups and sum Nominal Scale
Site Name Categorization into groups and sum Nominal Scale
User Id Parameter of aggregation Scale Scale
Table 3. Non significant variables according to the executive vice president of the company
Variables
Variables Description Notes
Measures
Activity tag name associated to the action that
Unclear
Activity Tag name the potential customer performed on the client Nominal
Variable
(company) site
Amount of conversion attributed to a specific
Unclear
Amount of Model Conversion potential customer in a journey based on the Scale
Variable
results of the model
Quantity associated with the activity Quantity
Redundant
Activity Quantity can typically be 1 representing the activity but in Scale
Variable
some cases this will be > 1
Redundant
Revenue Activity Revenue associated with the activity Scale
Variable
Creative for display advertisement, belong to an Low Predictive
Creative Name Scale
advertiser Value
Many Missing
Price Paid Price paid by the consumer for the purchase Scale
Data
Flags journeys where the potential customer only
Single Conversion Activity Scale Redundant
has a single conversion/activity.
Low Predictive
Site Placement Placement on the Site (for instance: homepage) Scale
Value
Unclear
Rank 0 Variable no identified from the company Scale
Variables
17Customer Relationship Management and Data Mining
values of the VIF means that the variables with- Table 5 describes the variable used in the
in the model are highly correlated (Caramanis & study. In addition, it provides an encoding of the
Spathis, 2006; Judge et al., 1987; Studenmund, variables analysed in order to better understand
2006). A VIF greater of 10 could indicate a mul- the classification tree output. Finally, particular
ticollinearity problem (Neter et al., 1996), while attention must be paid to the statistical measure
VIF values less than 2 mean that the variables are of the variables.
almost independent (Fernandez, 2007; Judge et
al., 1987; Leow & Mues, 2012). Table 4 shows
the variables with a VIF value less than 10.
Table 5. Description of the selected variables
The final dataset contains 1,463,199 potential
Variables
customers and 18 quantitative variables (1 target
Variables Description
Measures
variables and 17 independent variables) relating
Number of times that a
to their purchase behaviour. The target variable
Dynamic Click potential customer click on Scale
is dichotomous and it assumes two values: 0 when
Banner Moving
the potential customer does not purchase the
Click Through Rate. It is a
way of measuring the success
service (bad customer) and 1 when the potential
of an online campaign for
customer buys the service offered by the com-
a particular product or
Click Through service. The click through
pany (good customer).
Scale
Rate rate advertisement is defined
as the number of clicks on
an advertisements divided
by the number of times the
advertisement is shown
Table 4. Multicollinearity analysis
Number of times that the site
Average
name appears in the Top Best
Position Scale
COLLINEARITY
Five after the research in the
Best Five
STATISTICS search engine
Variables
TOLERANCE VIF Number of times that potential
customer digits one of the
Mean Click Trough Rate ,906 1,103 Brand Search Scale
brand name company in the
search engine
Average Position ,804 1,243
Number of times that a
Dynamic Click ,721 1,388
Match Type: potential customer digits one
Scale
Average Position Best Five ,132 7,599
Broad of the keyword on the search
engine
Brand Search ,176 5,692
Number of times that a
Impression or Click at 1pm ,126 7,928
Match Type: potential customer digits an
Scale
Impression or Click at 2pm ,110 9,065
Exact exact keyword on the search
engine
Impression or Click at 3pm ,123 8,124
Target Variable. If the potential
Impression or Click at 4pm ,175 5,719
customers purchases a service
Purchases Scale
Match Type: Broad ,484 2,066
online the variable is marks
with 1 otherwise 0
Match Type: Exact ,135 7,430
Number of times that a
Search Engine on Google ,125 7,997 Impression or
potential customer performs an Scale
Click at 1pm
Site Name: ConionMCUK ,924 1,082 impression or click at 1pm
Site Name: Adjug6 ,855 1,170 Number of times that a
Impression or
potential customer performs an Scale
Click at 2pm
Site Name: Affiliate Window ,721 1,387
impression or click at 2pm
Site Name: Drive PM ,904 1,106
Site Name: MCUK Quidco ,700 1,429
continued on following page
18Customer Relationship Management and Data Mining
Table 5. Continued
argmaxΔR(s,t)
ΔR,t =
( )
(1)
s∈S
Variables
Variables Description
Measures
where ΔR s,t =R t −R t −R t . is an
( ) ( ) ( ) ( )
L R
Number of times that a
Impression or
improvement in the re-substitution estimate for
potential customer performs an Scale
Click at 3pm
impression or click at 3pm
split s of t. The re-substitution estimate R(t) is
Number of times that a
defined as follows:
Impression or
potential customer performs an Scale
Click at 4pm
impression or click at 4pm
2
1
Average Average position of a search
Scale
R t = y −y t (2)
( ) ( ( ))

n
Position term in the search engine
N
x ∈t
n
Number of times that potential
Search Engine
customers insert a company Scale
on Google
The variables t and t are left and right value
keyword on Google
L R
for split t. The variable y(t) is defined as follows:
Site Name: Number of times that potential
Conion customers are exposed to a Scale
MCUK specific banner
1
y t = y (3)
Number of times that potential
( )

n
Site Name:
N t
customers are exposed to a Scale ( ) x ∈t
n
Adjug6
specific banner
Site Name: Number of times that potential
where N(t) is the total number of cases in t. The
Affiliate customers are exposed to a Scale
tree continues to grow until a node is reached such
Window specific banner
that no significant decrease in the re-substitution
Number of times that potential
Site Name:
customers are exposed to a Scale estimate is possible. This node is the terminal node.
Drive PM
specific banner
According to Giudici (2010), the Gini Impurity
Number of times that potential
is equal to:
Site Name:
customers are exposed to a Scale
Yahoo Q2006
specific banner
k m
( )
2
(4)
I m =1− π
( )

G i
i=1
Given the nature of the target variable a clas-
sification decision tree model based on the
where π are the fitted probabilities of the levels
i
“CART” Algorithm (Equations 1-3) and on the
present at node m, which are the most k(m).
Gini Impurity (Equations 4 and 5) has been de-
The fitted success probability is given by:
veloped in this research.
n
According to Pendharkar et al. (2005), CART
m
y

i=1 lm
π =
constructs a binary decision tree by splitting a (5)
i
n
m
database in such a way that the data in the de-
scendant subsets are more pure than the data in
where the observations y can take the value 0
lm
the parent set.
or 1, and the fitted probability corresponds to
For example, let (x , y ) represent nth example,
n n
the observation proportion of success in group
where x is the nth example vector on independent
n
M (Figini & Giudici, 2009; Giudici, 2010). The
variables and y is the value of the target vari-
n
output of the analysis is represented through a
able. If there are a total N examples, then CART
tree. This implies that the partition performed at a
calculates a best split s* so that the following is
certain level is influenced by the previous choices.
maximized over all possible splits S:
For a classification tree, a discriminant rule can be
19Customer Relationship Management and Data Mining
derived at each leaf of the tree. Each leaf points out ing a non-buyer as a buyer equally. It is easier to
a clear allocation rule of the observations, which incorporate the issue of unequal misclassification
is read by going through the path that connects costs into the PCC criterion rather than into AUC.
the initial node to each of them (Berry & Linoff, For instance, the probability of a misclassification
2011). In particular, the final output includes is multiplied by the cost of misclassification (for
the following information at each terminal node. both buyers and non-buyers). Both performance
Firstly, it shows a probability, which represents criteria will be calculated on a test or holdout
the membership degree of a terminal node of the sample, which only consists of observations not
class. Secondly, it provides the class assigned to used during model estimation, and which is half
the terminal node related to its probability and to the size of the total sample (Van den Poel, 2003).
pre-specified misclassification costs (Colombet
et al., 2000).
Finally, in order to evaluate the correctness of DISCUSSIONS
the predict model proposed, the criteria based on
the loss functions such as Percentage Correctly Before explaining the results relating to the clas-
Classified (PPC) and Area Under the receiver sification decision tree an explorative analysis of
operating Characteristic curve (AUC) have been the target variable is given next.
implemented in this research. Both measures are Table 6 shows the purchase frequency of po-
commonly used as performance criteria (Mozer tential customers in a given period. Just 1.74% of
et al., 2000). The PPC compares the “posterior” potential customers purchased the service at least
probability of defection with the true status of the once. In other words, 25,433 potential customers
customer. The resulting confusion matrix is used are “good” because of they purchased the service
to calculate the accuracy of the models. It contains offered through a marketing campaign, while
the number of elements that have been correctly 1,437,766 are “bad” in as much as they did not
or incorrectly classified for each class. The main purchase the service online.
diagonal shows the number of observations that Table 7 provides some position and dispersion
have been correctly classified for each class; the indexes related to the target variable. On average,
off-diagonal elements indicate the number of ob- just 1.70% of potential customers purchase the
servations that have been incorrectly classified. A service online. The minimum and maximum
disadvantage of this measure is that it is not very value confirms that no outliers are present in the
robust concerning the chosen cut off value in the distribution of the variable. Finally, the high
“a posterior” probabilities (Baesens et al., 2002). value of the coefficient of variation highlights
The AUC measure takes into account all possible that the arithmetic mean is an inaccurate indicator
cut off levels. For all these points, it considers the in explaining the distribution of the variable.
sensitivity (the number of true positive versus the
total number of events) and the specificity (the
number of true negatives versus the total number
of non-events) of the confusion matrix in a two- Table 6. Frequency of the variable “purchases”
dimensional graph, resulting in a ROC curve. The
Absolute Relative
area under this curve can be used to evaluate the
Purchases
Frequency Frequency
predictive accuracy of the classification model
0 1,437,766 98,26
(Hanley & McNeil, 1982). Both PPC and AUC
1 25,433 1,74
weigh the opportunity cost of misclassifying a
Total 1,463,199 100
buyer as a non-buyer and the cost of misclassify-
20Customer Relationship Management and Data Mining
Table 7. Descriptive values of the variable “purchases”
Minimum Maximum Arithmetic Standard Coefficient of
Variance
Value Value Mean Deviation Variation
Purchases ,00 1,00 ,017 ,017 ,130 7,64
Owing to the huge volume of the data, in order without identifying cause and effect (Ahlgren et
to discover the main variables to enter into the al., 2003).
classification tree we have estimated the value of From a statistical point of view only the
the Pearson Correlation Index among the target variables with p-value < 0.005 are significantly
variable (Purchases) and each independent vari- correlated to the target variables (Baum, 2006).
able collected in the database. Table 8 shows the The value of the p-value represents a decreasing
results of the Pearson Correlation Analysis. index of the reliability of a result (Moody, 2009).
The Pearson Correlation Coefficient (PCC) In addition, the PCC provides information about
measures the strength and direction (decreasing the collinearity of the variables. In our case, the
or increasing, depending on the coefficient sign) correlation values suggest that the variables are
of a linear relationship between two variables not redundant between them.
According to the EVP) Data Platforms of the
Company, the variables related to a specific banner,
Table 8. Pearson correlation analysis
which exposed to the potential customer, will be
included in the classification tree due to the fact
Pearson
Variables Correlation that they could contain important strategic value. In
Value
addition, the EVP suggests continuous monitoring
Purchases 1,00
of the variable “Dynamic Click” because it will
Dynamic Click ,62**
represent the number of “Fraud Clicks” generated
Average Click Through Rate ,29**
by potential customers. According to Asdemir et al.
Average Position Best Five ,38** (2008) click fraud occurs when a web users click
on a sponsored link with the malicious intent of
Brand Search ,39**
hurting a competitor or gaining undue monetary
Match Type: Broad ,16**
benefits. Competitors could generate a Dynamic
Match Type: Exact ,38**
Click creating a direct connection with the payment
Impression or Click at 1pm ,10**
page. Despite its strategic value, this variable must
Impression or Click at 2pm ,10**
be removed from the classification tree because
Impression or Click at 3pm ,10**
only 18,515 out of 1,463,199 potential customers
Impression or Click at 4pm ,10**
generated a dynamic click into the web marketing
Average Position ,40**
campaign. Finally, variables such as “Average
Search Engine on Google ,39**
Position Best Five” and “Brand Search” seem to
Site Name: Aconiom MCUK ,09**
be redundant even though they contain different
Site Name: Adjug6 ,05**
business information.
Site Name: Affiliate Window ,41**
A first look at the exploratory analysis shows
Site Name: Drive PM ,05**
some very interesting outcomes but a predictive
Site Name: Yahoo Q2006 ,09** model is needed to better understand the strength
of these relationships. A classification decision
** Correlation is significant at the 0.01 level
* Correlation is significant at the 0.05 level
21Customer Relationship Management and Data Mining
tree has been developed in this research in order important knowledge structure used for the clas-
to identify the best marketing drivers that could sification of future events to support the decision
increase the probability of customer conversion. making process as it is very simple to interpret.
The Classification Decision Tree is one of the main Figure 1 provides a representation of the
computational data mining models able to assist classification decision tree highlighting its effi-
marketers in the detection of the main marketing ciency, effectiveness, and validity in identifying
drivers in which organizations should make the the potential customer purchase behaviour (target
most of their future marketing investments, in order variable) using our explanatory variables.
to maximize the probability of customer conver- Table 9 shows a more in depth and detailed
sion. In addition, tree models aim to identify the interpretation of the decision tree, identifying how
best marketing activities performed by potential the explanatory variables (in the terminal node)
customers before purchasing the service proposed influence the purchase probability and the number
by the marketing campaign. Finally, it could be an of potential customers.
Figure 1. Classification decision tree
22Customer Relationship Management and Data Mining
From the classification decision it emerges node 15 (MCUKYahoo2006) distinguishes a
that “Affiliate Websites” are a key driver of cus- customer category that is very unlikely to buy the
tomer conversion. 91% out of the 7,000 potential service proposed by the company. More pre-
customers who have visited an affiliate website cisely, potential customers that neither visited
purchased the company’s service online. Also, affiliate websites nor searched a specific keyword
Table 9. Interpretation of the classification tree
Purchase N. of potential
Node Customer Category
probability customers
The potential customer has been exposed to a banner on an affiliate website at
2 91% 7,362
least one time.
The potential customer has never been exposed to a banner on an affiliate website
7 AND has been exposed to banners whose mean click through rate is greater than 14% 1,726
0,5% AND has never digit the company’s brand name on a search engine.
The potential customer has never been exposed to a banner on an affiliate website
AND has been exposed to banners whose mean click through rate is lower than
10 31% 3,904
0,5% AND has never digits a campaign keyword on a search engine AND visited
the ‘Conion MCUK’ website more than 6 times.
The potential customer has never been exposed to a banner on an affiliate website
AND has been exposed to banners whose mean click through rate is lower than
11 35,40% 915
0,5% AND has searched on Google at least one time AND has never digits a
specific campaign keyword on a search engine.
The potential customer has never been exposed to a banner on an affiliate website
AND has been exposed to banners whose mean click through rate is lower than
12 75% 1,583
0,5% AND has searched on Google at least one time AND has digits a specific
campaign keyword on a search engine at least once.
The potential customer has never been exposed to a banner on an affiliate website
AND has been exposed to banners whose mean click through rate is greater than
13 0,5% AND has digits the company’s brand name on a search engine at least one 61% 7,001
time AND has digits a specific campaign keyword on a search engine either one or
zero times.
The potential customer has never been exposed to a banner on an affiliate
website AND has been exposed to banners whose mean click through rate is
15 lower than 0,5% AND has never digit a campaign keyword on Google AND 1% 1,438,292
has visited Conion MCUK website less than seven times AND has visited
MCUKYahooQ2006 website less than 21 times
The potential customer has never been exposed to a banner on an affiliate website
AND has been exposed to banners whose mean click through rate is lower than
16 0,5% AND has never digit a campaign keyword on Google AND has visited 41% 680
Conion MCUK website less than seven times AND has visited MCUK Yahoo
Q2006 website at least 22 times.
The potential customer has never been exposed to a banner on an affiliate website
AND has been exposed to banners whose mean click through rate is greater than
17 0,5% AND has digit the company’s brand name on a search engine at least one 80% 890
time AND has digits exactly a campaign keyword on a search engine at least two
times AND visited websites that have an average search position lower than 1,014.
The potential customer has never been exposed to a banner on an affiliate website
AND has been exposed to banners whose mean click through rate is greater than
0,5% AND has digit the company’s brand name on a search engine at least one
18 92% 846
time AND has digits exactly a campaign keyword on a search engine at least two
times AND visited websites that have an average search position greater than
1,014.
23Customer Relationship Management and Data Mining
campaign had a probability of customer conver- Due to the weakness of the Confusion Matrix,
sion lower than 1%. Instead, node 10 (Conion Figure 2 provides a robust measure to evaluate
MCUK) identifies a group of potential customers the accuracy of the classification decision tree
with a low probability (31%) of purchase. This model.
group is mainly characterized by “Conion MCUK” The predictive accuracy of the classification
visitors who do not visit “Affiliate Websites” and decision tree is moderate because it is equal to
do not search for campaign keywords on the in- 80.20% (Swets, 1988).
ternet. On the contrary, node 13 (Match Type:
Exact) demonstrates that activities such as to
digit the company’s name on a search engine and FUTURE RESEARCH DIRECTIONS
being exposed to banners with a higher click
through rate increases the conversion probability. This research focused attention on the effective-
This category is composed by 7,001 potential ness of the computational data mining predictive
customers and has a probability of purchase equal models in estimating the probability of customer
to 60%. conversion. Particular attention has been paid to
Finally, the Criteria Based on the Loss Func- the classification decision tree in order to test its
tions confirms that the classification decision tree predictive power in forecasting accurate market-
based on the “CART” algorithm is an accurate ing performance within global organizations in
computational data mining predictive tool to today’s competitive landscape.
forecast the main marketing activities helpful to Future research could test new computational
increase the probability of customer conversion. predictive data mining models such as hierarchi-
Table 10 notes that the classification tree cor- cal logistic regression based on Enter Methods,
rectly classifies 13,627 out of 17,692 potential classification decision tree based on CHAID
customers belonging to the group of potential algorithm, parametric models based on the Ex-
customers. On the other hand, the classification ponential Family of distributions and graphical
tree correctly identifies 1,433,711 out of 1,445,517 models known as expert systems or Bayesian
“bad” potential customers. Therefore, the clas- networks in order to detect the best models able
sification tree accurately predicts the 99,7% of to maximize the level of the marketing campaign
potential customers, 53,6% of them as customer. profitability. Futures research could also concern
In general, based on cut-off of 0,05, the accuracy an analysis of a hybrid model (i.e. neural network
of the classification tree is really high because of and genetic algorithms) in order to draw an ac-
equal to 98,9% (Swets, 1988). curate churn prediction model able to discover the
Table 10. Confusion matrix
Figure 2. Roc curve purchases
PREDICTED VALUES
OBSERVED Purchase Variable
VALUES
Percentage
0 1
Correct
0 1,433,711 4,055 99,7
1 11,806 13,627 53,6
Total Overall
98,9*
Percentage
*The cut-off value is 0.05
24Customer Relationship Management and Data Mining
main marketing drivers useful to predict customer accomplished. In fact, traditional statistical models
churn. Also, a framework to identify a loyalty index based on a double moving average and exponential
and an analysis of the causes of churn should be smoothing are definitely inadequate in forecasting
provided. A text mining analysis joined with an punctual marketing performance within global
opinion mining could support the design of this organizations. First, the double moving average
new framework. Moreover, with new data, the models the predictive value of the data and will be
model proposed can be enhanced to predict future less sensitive to the actual changes, and the moving
sales using current marketing drivers. It could be average is not always an effective trend indica-
interesting to develop a cluster analysis for identi- tor. Second, the predicted value always remains
fying potential customer partition based on their at the level of the past and cannot be expected to
behavior and a survival analysis for estimating predict a higher or lower volatility of the future.
how many times customers will so remain in the Instead, the exponential smoothing model requires
company. Finally, we could test the effectiveness a more complete historical data before starting the
of the data mining techniques on a sample of small prediction and if season factors greatly influence
and medium-sized enterprises characterized by a business sales, times decomposition is more ap-
completely different strategic management. plicable than exponential smoothing. In our case, it
would have been impossible to implement double
moving average or exponential smoothing models
CONCLUSION due to the large amount of variables collected in
the dataset analyzed.
Data mining needs to become an essential busi- In the current literature a few scientific articles
ness process, incorporated into to other process argue about this business problem. The exponential
including marketing, sales, customer support, smoothing model is the main approach used in
product design, finance, engineering and inventory many organizations in order to predict marketing
control. This virtuous cycle places this methodol- performance rather than their performance, which
ogy in the larger contest of business, shifting the is totally inaccurate and without sense (Geng
focus away from the discovery mechanism to the & Du, 2010; Hyndman et al., 2008). Against it
actions based on the discovery. In addition, data emerges the strategic importance of our argued
mining methodology could be a strategic tool research question. In other words, many compa-
for global and international entrepreneurship in nies develop inaccurate marketing forecasts due
order to discover new business opportunities and to the implementation of inadequate predictive
maximize the level of company profitability. techniques unable to manage huge amounts of
This research focused the attention on the data. It is obvious that predictive data mining
effectiveness of the computational data mining models compared to traditional statistical mod-
predictive models in estimating the probability els are the key to solve business problems in the
of customer conversion. Particular attention has presence of enormous amounts of data (Geng &
been paid to the classification decision tree in Du, 2010). More precisely, the main strengths of
order to test its predictive power in forecasting the classification decision tree are the following.
accurate marketing performance within global First, classification trees are predictive rather than
organizations in today’s competitive landscape. descriptive. Second, they perform a classification
Our research question - are innovative compu- on the observations on the basis of the observa-
tational data mining models more efficient than tions of all independent variables and supervised
traditional statistical models in forecasting market- by the presence of the target variable. Third, in
ing performance with global organizations? - is classification decision trees the segmentation is
25Customer Relationship Management and Data Mining
typically carried out using only the maximally over vulnerable people or discriminate against a
independent variables. In contrast, the main weak- certain group of people. In addition, the price of
ness is that compared with traditional statistical the data mining software is very high because of
model classification, decision tree models produce their complexity in terms of specific algorithms
rules that are less explicit analytically but easier and models implemented. For this reason, firms
to understand graphically. Subsequently, tree sometimes cannot develop data mining analysis
models require a higher demand of computational but merely statistical analysis.
resources due to the fact that they do not require In summing up, our research suggests that
assumptions about their probability distribution decision tree is an accurate prediction model
of the target variable. Finally, the structure of able to forecast marketing performance in today’s
these predictive models can change at any time competitive landscape. In particular, the classifi-
thus it is impossible generalize their structure in cation decision provided a punctual description
another context. From this discussion emerges of the main marketing activities that could grow
the use of sophisticated and computationally the level of sales. “Affiliate Web Site” is the best
intensive analytical methods which are expected key driver of customer conversion. In addition,
to become even more common place with recent this model accurately discriminates a potential
research breakthroughs in computational methods customer category that is very unlikely to buy the
and their commercialization by leading vendors service proposed by the company. The ROC curve
in global business today (Grossman et al., 2001). confirms that the classification decision tree is an
As a consequence, many authors argued about the effective predictive to forecast punctual market-
main differences between data mining and statis- ing performance within global organizations. In
tics. Sato (2000) observes that the data mining conclusion, success in using data will transform
analysis differs from the statistical data analysis. organizations from reactive to proactive.
For instance, statisticians use sample observations
to study the population parameters by estimation,
testing and predicting, whilst data mining analysis NOMENCLATURE
is governed by the need to uncover, in a timely
manner, emerging trends, whereas statistical data • ANN: Artificial Neural Network
analysis is related to historical facts and is based • AUC: Area Under the receiver operating
on observed data. On the other hand, in adopting a Characteristic Curve
data mining methodology decision-makers should • CART: Classification and Regression Tree
consider some limits such as: • CHAID: CHI-Squared Automatic
Interaction Detection
• Privacy and Security Issues • CRM: Customer Relationship Management
• Misuse of Information • DM: Data Mininig
• Cost of the Tools • DT: Decision Tree
• EVP: Executive Vice President
People are often afraid that somebody may • GAs: Genetich Algorithms
have access to their personal information and • GSP: Generalized Sequential Pattern
then use that information in an unethical way. • ID3: Interactive Dichotomizer 3
Although companies have much information about • PCC: Pearson Correlation Coefficient
us available online, they do not have sufficient • ROC Curve: Receiver Operating
security systems in place to protect the informa- Characteristic Curve
tion. Unethical businesses may use the informa- • SVM: Support Vector Machine
tion obtained from data mining to take advantage
26Customer Relationship Management and Data Mining
REFERENCES Baum, C. F. (2006). An introduction to modern
econometric using Stata. Stata Press.
Abbott, P. A., & Lee, S. M. (2006). Data mining
Belbay, N., Benbya, H., & Meissonier, R. (2007).
and knowledge discovery. In Essentials of nursing
An empirical investigation of the customer Knowl-
informatics (4th ed.). New York, NY: McGraw-
edge creation impact on NPD Performance. In
Hill Medical Pub. Division.
Proceedings of the 40th Hawaii International
Aggarval, C. C., & Yu, P. S. (2002). Finding lo-
Conference on System Science. IEEE Computer
calized associations in market basket data. IEEE
Society Press.
Transactions on Knowledge and Data Engineer-
Bergeron, B. (2001). Essentials of CRM: Customer
ing, 14, 51–62. doi:10.1109/69.979972.
relationship management for executives. New
Ahlgren, P., Jarneving, B., & Rousseau, R. (2003).
York: John Wiley & Sons.
Requirement for a cocitation similarity measure,
Berry, M. J. A., & Linoff, G. (2004). Data mining
with special reference to Pearson’s correlation
techniques second edition–for marketing, sales,
coefficient. Journal of the American Society
and customer relationship management. New
for Information Science and Technology, 54(6),
York: John Wiley & Sons.
550–560. doi:10.1002/asi.10242.
Berry, M. J. A., & Linoff, G. S. (2011). Data min-
Ahmed, S. R. (2004). Applications of data min-
ing techniques for marketing, sales, and customer
ing in retail business. Information Technology:
relationship management. New York: John Wiley
Coding and Computing, 2, 455–459.
& Sons, Inc..
Armin, E. M., & Babak, M. (2011). Genetic al-
Berson, A., Smith, S., & Thearling, K. (2000).
gorithm based on optimal load frequency control
Building data mining applications for CRM. New
in two-area interconnected power system. Global
York: McGraw-Hill.
Journal of Technology & Optimization, 2(1), 6–10.
Bloemer, J. M. M., Brijs, T., Vanhoof, K., &
Arndt, D., & Gersten, W. (2001). Data manage-
Swinnen, G. (2003). Comparing complete and
ment in analytical customer relationship man-
partial classification for identifying customers
agement. In Proceedings of the Data Mining for
at risk. International Journal of Research in
Marketing Applications Workshop at ECML/
Marketing, 20(2), 117–131. doi:10.1016/S0167-
PKDD. EMCL/PKDD.
8116(03)00014-4.
Asdemir, K., Yurtseven, O., & Yahya, M. A.
Boulding, W., Staelin, R., Ehret, M., & Johnson,
(2008). An economic model of click fraud in
W. J. (2005). A customer relationship management
publisher networks. International Journal of
roadmap: What is known, potential pitfalls, and
Electronic Commerce, 13, 61–90. doi:10.2753/
where to go. Journal of Marketing, 69, 155–166.
JEC1086-4415130203.
doi:10.1509/jmkg.2005.69.4.155.
Baesens, B., Viaene, S., Van den Poel, D., Ven-
Breiman, L., Friedman, J. H., Olshen, R. A., &
thienen, J., & Dedene, G. (2002). Bayesian neural
Stone, C. J. (1984). Classification and regression
network learning for repeat purchase modelling
trees. Palo Alto, CA: Wadsworth.
in direct marketing. European Journal of Opera-
tional Research, 138(1), 191–211. doi:10.1016/
S0377-2217(01)00129-1.
27Customer Relationship Management and Data Mining
Brijs, T., Swinnen, G., Vanhoof, K., & Wets, G. Chang, J., Yen, D., Young, D., & Ku, C. (2002).
(2004). Building an association rules framework Critical issues in CRM adoption and implementa-
to improve product assortment decisions. Data tion. International Journal of Service Technology
Mining and Knowledge Discovery, 8, 7–23. and Measurement, 3(3), 311–324. doi:10.1504/
doi:10.1023/B:DAMI.0000005256.79013.69. IJSTM.2002.001633.
Brondoni, S. M. (2008a). Overture the market- Chen, Y. L., Hsu, C. L., & Chou, S. C. (2003).
driven management and global markets-1. Sym- Constructing a multi-valued and multilabeled de-
phonya: Emerging Issues in Management, 1, 1–13. cision tree. Expert Systems with Applications, 25,
199–209. doi:10.1016/S0957-4174(03)00047-2.
Brondoni, S. M. (2008b). Market-driven manage-
ment, competitive space and global networks. Chen, Y. L., Tang, K., Shen, R. J., & Hu, Y. H.
Symphonya: Emerging Issues in Management, (2005). Market basket analysis in a multiple store
1, 14–27. environment. Decision Support Systems, 40,
339–354. doi:10.1016/j.dss.2004.04.009.
Brondoni, S. M. (2008c). Overture the market-
driven management and global markets-2. Cheung, K. W., Kwok, J. T., Law, M. H., & Tsui,
Symphonya: Emeriging Issues in Management, K. C. (2003). Mining customer product ratings
2, 1–12. for personalized marketing. Decision Support
Systems, 35, 231–243. doi:10.1016/S0167-
Buchnowska, D. (2011). Customer knowledge
9236(02)00108-2.
management models: Assessment and proposal.
Springer-Verlag Berlin Heidelberg, 93, 25–38. Colombet, I., Ruelland, A., Chatellier, G., Gulyf-
fier, F., Degoulet, P., & Jaulent, M. C. (2000).
Buckinx, W., Moons, E., Poel, D. V. D., &
Models to predict cardiovascular risk: compari-
Wets, G. (2004). Customer-adapted coupon
son of CART, multilayer perceptron and logistic
targeting using feature selection. Expert Systems
regression. Symposium Proceedings Archive,
with Applications, 26, 509–518. doi:10.1016/j.
156-160.
eswa.2003.10.009.
Corniani, M. (2002). Demand Bubble Manage-
Cao, Y., & Gruca, T. S. (2005). Reducing adverse
ment. Symphonya: Emeriging Issue in Manage-
selection through customer relationship man-
ment, 1, 87–99.
agement. Journal of Marketing, 69, 219–229.
doi:10.1509/jmkg.2005.69.4.219. Crone, S. V., Lessmann, S., & Stahlbock, R. (2006).
The impact of preprocessing on data mining: An
Caramanis, C., & Spathis, C. (2006). Auditee and
evaluation of classifier sensitivity in direct market-
audit firm characteristics as determinants of audit
ing. European Journal of Operational Research,
qualifications: Evidence from the Athens stock
173, 781–800. doi:10.1016/j.ejor.2005.07.023.
exchange. Managerial Auditing Journal, 21(9),
905–920. doi:10.1108/02686900610705000. D’Aveni, R. (1994). Hypercompetition: Manag-
ing the dynamic of strategic maneuvering. New
Carrier, C. G., & Povel, O. (2003). Characterising
York: Free Press.
data mining software. Intelligent Data Analysis,
7, 181–192. Davis, L. (1991). Handbook of genetic algorithms.
New York: Van Nostrand Reinhold.
Cavique, L. (2004). Graph-based structures for
the market baskets analysis. Investing Operations,
24(2), 233–246.
28Customer Relationship Management and Data Mining
Day, G. S. (1994). The capabilities of market- Etemadi, H., Rostamy, A. A. A., & Dehkordi, H.
driven organization. Journal of Marketing, 58(4), F. (2009). A genetic programming model for bank-
37–52. doi:10.2307/1251915. ruptcy prediction: Empirical evidence from Iran.
Expert Systems with Applications, 36, 3199–3207.
Deshpandè, R., Farley, J. U., & Webster, F. Jr.
doi:10.1016/j.eswa.2008.01.012.
(1993). Corporate culture, customer orientation,
and innovativeness in Japanese firms: a quadrad Etzion, O., Fisher, A., & Wasserkrug, S. (2005).
analysis. Journal of Marketing, 57, 23–73. E-CLV: A modelling approach for customer life-
doi:10.2307/1252055. time evaluation in e-commerce domains, with
an application and case study for online auction.
Dowling, G. (2002). Customer relationship man-
Information Systems Frontiers, 7, 421–434.
agement: in B2C markets, often less is more.
doi:10.1007/s10796-005-4812-6.
California Management Review, 44(3), 87–104.
doi:10.2307/41166134. Feng, F. (2012). Research of customer relation-
ship management solutions based on data mining.
Drew, J. H., Mani, D. R., Betz, A. L., & Datta, P.
Information Engineering and Applications, 154,
(2001). Targeting customers with statistical and da-
47–52. doi:10.1007/978-1-4471-2386-6_87.
ta-mining techniques. Journal of Service Research,
3, 205–220. doi:10.1177/109467050133002. Feng, X., & Yuan, G. (2011). Optimizing two-
stage fuzzy multi-product multi-period production
Dwyer, R. R., Schur, P. H., & Oh, S. (1987).
planning problem. Information, 14(6), 1879–1893.
Developing buyer-seller relations. Journal of
Marketing, 51, 11–28. doi:10.2307/1251126. Fernandez, G. C. J. (2007). Effects of multicol-
linearity in all possible mixed model selection.
ECCS. (1999). CRM defining customer relation-
In Proceedings of the PharamaSUG Conference,
ship marketing and management. Retrieved from
Statistics and Pharmacokinetics. Denver, DO:
http://www.eccs.uk.com/crm/crmdefinitions/
PharamaSUG.
define.asp
Figini, S., & Giudici, P. (2009). Applied data
Elamvazuthi, I., Vasant, P., & Ganesan, T. (2010).
mining for business and industry. New York, NY:
Fuzzy linear programming using modified logistic
John Wiley & Sons, Inc..
membership function. International Review of
Automatic Control, 3(4), 370–377. Gaby, P., Uriel, I., Inessa, A., & Gad, R. (2010). GA
for the resource sharing and scheduling problem.
Elamvazuthi, I., Vasant, P., & Ganesan, T. (2012).
Global Journal on Technology & Optimization,
Integration of fuzzy logic techniques into DSS
1, 84–87.
for profitability quantification in a manufacturing
environment. In M. Khan, & A. Ansari (Eds.), Galvão, N. D., & Marin, H. (2009). Data mining:
Handbook of research on industrial informatics A literature review. Acta Paul Enferm, 22(5),
and manufacturing intelligence: Innovations and 686–690.
solutions (pp. 171-192). doi: doi:10.4018/978-1-
Ganesan, T., Vasant, P., & Elamvazuthi, I. (2011).
4666-0294-6. ch007.
Hybrid neuro-genetic programming approach
for optimizing operations in 3-D land seismic
surveys. Mathematical and Computer Modelling,
54, 2913–2922. doi:10.1016/j.mcm.2011.07.012.
29Customer Relationship Management and Data Mining
Ganesan, T., Vasant, P., & Elamvazuthi, I. (2012). Hadden, J., Tiwari, A., Roy, R., & Ruta, D. (2005).
Hybrid PSO approach for solving nonconvex opti- Computer assisted customer churn management:
mization problems. Archives of Control Sciences, State-of-the-art and future trends. Comput-
22(1), 5–23. doi:10.2478/v10170-011-0014-2. ers & Operations Research, 34, 2902–2917.
doi:10.1016/j.cor.2005.11.007.
Gehrke, J., Ramakrishnan, R., & Loh, W. Y.
(1999). BOAT-optimistic decision tree construc- Han, J., & Kamber, M. (2006). Data mining:
tion. In Proceedings ACM SIGMOD International Concepts and techniques (2nd ed.). Amsterdam:
Conference Management of Data, (pp. 169-180). Morgan Kaufmann.
Philadelphia, PA: ACM.
Hand, D. (1997). Construction and assessment
Geng, Y., & Du, X. (2010). The research of data of classification rules. New York: John Wiley &
mining based sales forecast. Washington, DC: Sons, Ltd..
IEEE. doi:10.1109/ICMULT.2010.5629850.
Hand, D. J., Mannila, H., & Smyth, P. (2001).
Giudici, P. (2010). Data mining. Milano, Italy: Principles of data mining. Cambridge, MA: MIT
McGraw-Hill. Press.
Giudici, P., & Passerone, G. (2002). Data mining Hanley, J. A., & McNeil, B. J. (1982). The mean-
of association structures to model consumer behav- ing and use of the area under a receiver operat-
iour. Computational Statistics & Data Analysis, 38, ing characteristic (ROC) curve. Radiology, 143,
533–541. doi:10.1016/S0167-9473(01)00077-9. 29–36. PMID:7063747.
Goldberg, D. (1989). Genetic algorithms in search, Hansotia, B. (2002). Gearing up for CRM: An-
optimization, and machine learning. Reading, tecedents to successful implementation. Jour-
MA: Addison-Wesley. nal of Database Marketing, 10(2), 121–132.
doi:10.1057/palgrave.jdm.3240103.
Gordini, N. (2010). Market-driven management:
A critical literature review. Symphonya: Emerging He, Z., Xu, X., Huang, J. Z., & Deng, S. (2004).
Issues In Management, 2, 95–107. Mining class outliers: Concepts, algorithms,
and applications in CRM. Expert Systems with
Gordini, N. (2012). Market-driven management:
Applications, 27, 681–697. doi:10.1016/j.
From the Japanese experience to the European
eswa.2004.07.002.
schools. In Market-Driven Management and
Corporate Growth (pp. 95-115). Giappichelli, Holland, J. H. (1975). Adaptation in natural and
Torino: Academic Press. artificial system. Ann Arbor, MI: University of
Michigan Press.
Gordini, N. (2013). Genetic algorithms for small
enterprises default prediction: Empirical evidence Hyndman, R. J., Koehler, A. B., Ord, J. K., &
from Italy. In Handbook of Research on Novel Soft Snyder, D. S. (2008). Forecasting with exponential
Computing Intelligent Algorithms: Theory and smoothing. Springer. doi:10.1007/978-3-540-
Practical Applications. Hershey, PA: IGI-Global. 71918-2.
Grossman, R. L., Kamath, C., Kegelmeyear, P., Injazz, D., & Karen, P. (2004). Understanding cus-
Kumar, V., & Novnburu, R. (2001). Data min- tomer relationship management (CRM). People,
ing for scientific and engineering applications. Process, and Technology. Retrieved from http://
London: Springer-Verlag. doi:10.1007/978-1- www.emeraldinsight.com/1463-7154.htm
4615-1733-7.
30Customer Relationship Management and Data Mining
Jaworski, B. J., & Kohli, A. K. (1993). Market ori- Kincaid, J. W. (2003). Customer relationship
entation: Antecedents and consequences. Journal management: Getting it right! Upper Saddle River,
of Marketing, 57, 53–70. doi:10.2307/1251854. NJ: Prentice-Hall.
Jayachandran, S., Sharma, S., Kaufman, P., & Kohli, A. K., & Jaworski, B. (1990). Market ori-
Raman, P. (2005). The role of relational informa- entation: The construct, research propositions, and
tion processes and technology use in customer managerial implications. Journal of Marketing,
relationship management. Journal of Marketing, 54, 1–18. doi:10.2307/1251866.
69, 177–192. doi:10.1509/jmkg.2005.69.4.177.
Kracklauer, A. H., Mills, D. Q., & Seifert, D.
Jiang, T., & Tuzhilin, A. (2006). Segmenting (2004). Customer management as the origin of
customers from population to individuals: Does collaborative customer relationship manage-
1-to-1 keep your customers forever. IEEE Trans- ment. Collaborative Customer Relationship
actions on Knowledge and Data Engineering, 18, Management-Taking CRM to the Next Level, 3–6.
1297–1311. doi:10.1109/TKDE.2006.164.
Kubat, M., Hafez, A., Raghavan, V. V., Lekkala,
Jiao, J. R., Zhang, Y., & Helander, M. (2006). J. R., & Chen, W. K. (2003). Item set trees for
A Kansei mining system for affective design. targeted association querying. IEEE Transac-
Expert Systems with Applications, 30, 658–673. tions on Knowledge and Data Engineering, 15,
doi:10.1016/j.eswa.2005.07.020. 1522–1534. doi:10.1109/TKDE.2003.1245290.
Judge, G. G., Hill, R. C., Griffiths, W. E., Lutke- Langley, P., & Simon, H. A. (1995). Applica-
pohl, H., & Lee, T. (1987). Theory and Practice tions of machine learning and rule induction.
of Econometrics (2nd ed.). New York: Wiley. Communications of the ACM, 38, 55–64.
doi:10.1145/219717.219768.
Kass, G. V. (1980). An exploratory technique
for investigating large quantities of categori- Lattin, J., Douglas, C., & Green, P. (2003). Ana-
cal data. Applied Statistics, 29(2), 119–127. lyzing multivariate data. Stamford, CT: Cengage
doi:10.2307/2986296. Learning.
Khanna, S. (2001). Measuring the CRM ROI: Lau, H. C. W., Wong, C. W. Y., Hui, I. K., &
Show them benefits. Retrieved from http://www. Pun, K. F. (2003). Design and implementation
crm-forum.com of an integrated knowledge system. Knowledge-
Based Systems, 16, 69–76. doi:10.1016/S0950-
Kim, J. K., Song, H. S., Kim, T. S., & Kim, H. K.
7051(02)00052-7.
(2005). Detecting the change of customer behavior
based on decision tree analysis. Expert Systems Lehmann, D. R. (2004). Metrics for making mar-
with Applications, 22, 193–205. keting matter. Journal of Marketing, 68, 73–75.
doi:10.1509/jmkg.68.4.73.42727.
Kim, Y. H., & Moon, B. R. (2006). Multicam-
paign assignment problem. IEEE Transactions on Lejeune, M. A. P. M. (2001). Measuring the
Knowledge and Data Engineering, 18, 405–414. impact of data mining on churn manage-
doi:10.1109/TKDE.2006.49. ment. Internet Research: Electronic Network-
ing Applications and Policy, 11, 375–387.
Kim, Y. S. (2006). Toward a successful CRM:
doi:10.1108/10662240110410183.
Variable selection, sampling, and ensemble. Deci-
sion Support Systems, 41, 542–553. doi:10.1016/j.
dss.2004.09.008.
31Customer Relationship Management and Data Mining
Leng, K., & Chen, X. (2012). A genetic algorithm McCulloch, W. S., & Pitts, W. A. (1943). Logical
approach for TOC-based supply chain coordina- calculus of ideas immanent in nervous activity. The
tion. Applied Mathematics and Information Sci- Bulletin of Mathematical Biophysics, 5, 115–133.
ences, 6(3), 767–774. doi:10.1007/BF02478259.
Leow, M., & Mues, C. (2012). Predicting loss Mithas, S., Krishnan, M. S., & Fornell, C.
given default (LGD) for residential mortgage (2005). Why do customer relationship manage-
loans: A two-stage model and empirical evi- ment applications affect customer satisfaction?
dence for UK bank data. International Journal Journal of Marketing, 69, 201–209. doi:10.1509/
of Forecasting, 28, 183–195. doi:10.1016/j. jmkg.2005.69.4.201.
ijforecast.2011.01.010.
Mitra, S., Pal, S. K., & Mitra, P. (2002). Data
Levitt, T. (1960). Marketing myopia. Harvard mining in soft computing framework: A survey.
Business Review, 38, 45–60. PMID:15252891. IEEE Transactions on Neural Networks, 13, 3–14.
doi:10.1109/72.977258 PMID:18244404.
Levitt, T. (1969). The marketing mode: Pathways
to corporate growth. New York: McGraw-Hill. Montgomery, D. C., & Peck, E. A. (1992). Intro-
duction to linear regression analysis. New York:
Lewis, M. (2005). Incorporating strategic con-
John Wiley & Sons.
sumer behavior into customer valuation. Jour-
nal of Marketing, 69, 230–238. doi:10.1509/ Moody, C. (2009). Basic econometric with
jmkg.2005.69.4.230. stata. Retrieved from http://www.scribd.com/
doc/54549213/Stata-Manual-2009
Liao, S. H., & Chen, Y. J. (2004). Mining cus-
tomer knowledge for electronic catalog marketing. Mozer, M. C., Wolneiewicz, R., Grimes, D. B.,
Expert Systems with Applications, 27, 521–532. Johnson, E., & Kaushansky, H. (2000). Predicting
doi:10.1016/j.eswa.2004.05.013. subscriber dissatisfaction and improving retention
in the wireless telecommunications industry. IEEE
Liao, S. H., Chu, P. H., & Hsiao, P.-Y. (2012). Data
Transactions on Neural Networks, 11, 690–696.
mining techniques and applications: A decade
doi:10.1109/72.846740 PMID:18249796.
review from 2000 to 2011. Expert Systems with
Applications, 39, 11303–11311. doi:10.1016/j. Mulhern, F. (1999). Customer profitability
eswa.2012.02.063. analysis: Measurement, concentration, and re-
search directions. Journal of Interactive Mar-
Ling, R., & Yen, D. C. (2001). Customer relation-
keting, 13, 25–40. doi:10.1002/(SICI)1520-
ship management: An analysis framework and
6653(199924)13:1<25::AID-DIR3>3.0.CO;2-L.
implementation strategies. Journal of Computer
Information Systems, 41, 82–97. Narver, J. C., & Slater, S. F. (1990). The effect of a
market orientation on business profitability. Jour-
Madronero, M. D., Peidro, D., & Vasant, P. (2010).
nal of Marketing, 54, 20–35. doi:10.2307/1251757.
Vendor selection problem by using an interactive
fuzzy multi-objective approach with modified s- Neter, J., Kutner, M. H., Nachtsheim, C. J., &
curve membership functions. Computers & Math- Wasserman, W. (1996). Applied linear regression
ematics with Applications (Oxford, England), 60, models. Chicago: McGraw Hill Education.
1038–1048. doi:10.1016/j.camwa.2010.03.060.
32Customer Relationship Management and Data Mining
Ngai, E. W. T. (2005). Customer relationship Prinzie, A., & Poel, D. V. D. (2005). Constrained
management research (1992–2002): An aca- optimization of data-mining problems to improve
demic literature review and classification. Mar- model performance: A direct-marketing appli-
keting Intelligence & Planning, 23, 582–605. cation. Expert Systems with Applications, 29,
doi:10.1108/02634500510624147. 630–640. doi:10.1016/j.eswa.2005.04.022.
Ngai, E. W. T., Xiu, L., & Chau, D. C. K. (2009). Prinzie, A., & Poel, D. V. D. (2006). Investigat-
Application of data mining techniques in customer ing purchasing-sequence patterns for financial
relationship management: A literature review and services using Markov, MTD and MTDg models.
classification. Expert Systems with Applications, European Journal of Operational Research, 170,
36, 2592–2602. doi:10.1016/j.eswa.2008.02.021. 710–734. doi:10.1016/j.ejor.2004.05.004.
Niraj, R., Gupta, M., & Narasimhan, C. (2001). Provas Kumar, R., & Dharmadas, M. (2012).
Customer profitability in a supply chain. Optimal reactive power dispatch using quasi-
Journal of Marketing, 65, 1–16. doi:10.1509/ oppositional biogeography-based optimization.
jmkg.65.3.1.18332. International Journal of Energy Optimization
and Engineering, 1(4), 38–55. doi:10.4018/
Panagiotis, I., Soukakos, Nikolaos, B., Georgop-
ijeoe.2012100103.
oulos, & Victoria, P. E. (2007). Two interrelated
frameworks proposed for mapping and perfor- Quinlan, J. R. (1987). Decision trees as probabi-
mance management of customer relationship listic classifiers. In Proceedings of 4th Interna-
management strategies. International Journal of tional Workshop of Machine Learning. Irvine,
Knowledge and Learning, 3(2-3), 299–315. CA: ICML Press.
Parvatiyar, A., & Sheth, J. N. (2001). Customer Quinlan, J. R. (1993). C4.5: Programs for machine
relationship management: Emerging practice, learning. San Mateo, CA: Morgan Kaufmann.
process, and discipline. Journal of Economic &
Rajshekhar (Raj), G. J., Martin, C. L., & Young,
Social Research, 3, 1–34.
R. B. (2006). Marketing research, market orienta-
Payne, A., & Frow, P. (2005). A strategic frame- tion, and customer relationship management: A
work for customer relationship management. framework and implications for service providers.
Journal of Marketing, 69, 167–179. doi:10.1509/ Journal of Services Marketing, 20(1), 12–23.
jmkg.2005.69.4.167. doi:10.1108/08876040610646545.
Pendharkar, P. C., Subramanian, G. H., & Roger, Ravi, V., Kurniawan, H., Thai, P. N. K., & Ravi
J. A. (2005). A probabilistic model for predict- Kumar, P. (2008). Soft computing system for bank
ing software development effort. IEEE Transac- performance prediction. Applied Soft Computing,
tions on Software Engineering, 31(7), 615–624. 8, 305–315. doi:10.1016/j.asoc.2007.02.001.
doi:10.1109/TSE.2005.75.
Ravi Kumar, P., & Ravi, V. (2007). Bankruptcy
Pinkey, C., Deep, K., & Pant, M. (2011). Optimiz- prediction in banks and firms via statistical and in-
ing CNC turning process using real coded genetic telligent techniques–A review. European Journal
algorithm and differential evolution. Global Jour- of Operational Research, 180, 1–28. doi:10.1016/j.
nal Technology and Optimization, 2(2). ejor.2006.08.043.
33Customer Relationship Management and Data Mining
Reinartz, W., Krafft, M., & Hoyer, W. D. (2004). Sato, Y. (2000). Perspective on data mining from
The customer relationship management process: statistical viewpoints: Knowledge discovery and
Its measurement and impact on performance. JMR, data mining. In Proceedings in Current Issues and
th
Journal of Marketing Research, 41, 293–305. New Applications, 4 Pacific-Asia Conference.
doi:10.1509/jmkr.41.3.293.35991. Kyoto, Japan: Springer.
Reinartz, W. J., & Kumar, V. (2003). The impact Scott, D. (2001). Understanding organizational
of customer relationship characteristics on profit- evolution: Its impact on management and perfor-
able lifetime duration. Journal of Marketing, 67, mance. Quorum Books.
77–99. doi:10.1509/jmkg.67.1.77.18589.
Shaw, M. J., Subramaniam, C., Tan, G. W., &
Rigby, D., & Bilodeau, B. (2009). Management Welge, M. E. (2001). Knowledge management
tools and trends 2009, research report. Bain and data mining for marketing. Decision Sup-
& Company. Retrieved from http://www.bain. port Systems, 31, 127–137. doi:10.1016/S0167-
com/bainweb/PDFs/cms/Public/Management 9236(00)00123-8.
Tools_2009.pdf
Shin, K. S., & Lee, Y. J. (2002). Genetic algorithm
Rogers, M. (2005). Customer strategy: Observa- application in bankruptcy prediction modeling.
tions from the trenches. Journal of Marketing, Expert Systems with Applications, 1–8.
69, 262–263. doi:10.1509/jmkg.2005.69.4.262.
Skinner, S. J. (1990). Marketing. Boston: Hough-
Rosset, S., Neumann, E., Eick, U., & Vatnik, N. ton Mifflin.
(2003). Customer lifetime value models for deci-
Slater, S., & Narver, J. C. (1994). Market
sion support. Data Mining and Knowledge Discov-
orientation, customer value, and superior per-
ery, 7, 321–339. doi:10.1023/A:1024036305874.
formance. Business Horizons, 37(2), 22–28.
Rud, O. P. (2000). Data mining cook book. New doi:10.1016/0007-6813(94)90029-9.
York: John Wiley and Sons.
Slater, S. F., & Narver, J. C. (1998). Cus-
Rust, R. T., Ambler, T., Carpenter, G. S., Kumar, tomer-led and market-oriented: Let’s not
V., & Srivastava, R. K. (2004). Measuring market- confuse the two. Strategic Management
ing productivity: Current knowledge and future Journal, 19(10), 1001–1006. doi:10.1002/
directions. Journal of Marketing, 68, 76–89. (SICI)1097-0266(199810)19:10<1001::AID-
doi:10.1509/jmkg.68.4.76.42721. SMJ996>3.0.CO;2-4.
Ryals, L. (2005). Making customer relationship Slater, S. F., & Narver, J. C. (1999). Market-oriented
management work: The measurement and prof- is more that being customer-led. Strategic Manage-
itable management of customer relationships. ment Journal, 20(12), 1165–1168. doi:10.1002/
Journal of Marketing, 69, 252–261. doi:10.1509/ (SICI)1097-0266(199912)20:12<1165::AID-
jmkg.2005.69.4.252. SMJ73>3.0.CO;2-#.
Rygielski, C., Wang, J. C., & Yen, D. C. (2002). Song, H. S., Kim, J. K., Cho, Y. B., & Kim, S. H.
Data mining techniques for customer relationship (2004). A personalized defection detection and
management. Technology in Society, 24, 483–502. prevention procedure based on the self-organizing
doi:10.1016/S0160-791X(02)00038-6. map and association rule mining. Applied to online
game site. Artificial Intelligence Review, 21, 161–
184. doi:10.1023/B:AIRE.0000021067.66616.b0.
34Customer Relationship Management and Data Mining
Spangler, W. E., May, J. H., & Vargas, L. G. (1999). Swift, R. S. (2001). Accelarating customer rela-
Choosing data-mining methods for multiple clas- tionships: Using CRM and relationship technolo-
sification: Representational and performance gies. Upper Saddle River, NJ: Prentice Hall PTR.
measurement implications for decision support.
Teo, T. S. H., Devadoss, P., & Pan, S. L. (2006).
Journal of Management Information Systems,
Towards a holistic perspective of customer rela-
16(1), 37–62.
tionship management implementation: A case
Srinivasan, R., & Moorman, C. (2005). Strategic study of the housing and development board, Sin-
firm commitments and rewards for customer gapore. Decision Support Systems, 42, 1613–1627.
relationship management in online retailing. doi:10.1016/j.dss.2006.01.007.
Journal of Marketing, 69, 193–200. doi:10.1509/
Thearling, K. (1998). Data mining and privacy:
jmkg.2005.69.4.193.
A conflict in the making? Retrieved from www.
Srivastava, R., Shervani, T., & Fahey, L. (1998). thearling.com
Marketing-Based assets and shareholder value:
Thomas, J. S., & Sullivan, U. Y. (2005). Manag-
A framework for analysis. Journal of Marketing,
ing marketing communications with multichannel
62, 2–18. doi:10.2307/1251799.
customers. Journal of Marketing, 69, 239–251.
Srivastava, R. K., Shervani, T., & Fahey, L. (1998). doi:10.1509/jmkg.2005.69.4.239.
Market-based assets and shareholder value: A
Turban, E., Aronson, J. E., Liang, T. P., & Sharda,
framework for analysis. Journal of Marketing,
R. (2007). Decision support and business intel-
62, 2–18. doi:10.2307/1251799.
ligence systems (8th ed.). Upper Saddle River, NJ:
Stone, M., & Woodcock, N. (2001). Defining Pearson Education.
CRM and assessing its quality. In Successful
Van den Poel, D. (2003). Predicting mail-order
Customer Relationship Marketing. London, UK:
repeat buying. Which variables matter? Review of
Kogan Page.
Business and Economics, 48(3), 371–404.
Studenmund, A. H. (2006). Using econometrics:
Varetto, F. (1998). Genetic algorithms applica-
A practical guide (5th ed.). New York: Addison-
tions in the analysis of insolvency risk. Journal of
Wesley.
Banking & Finance, 22, 1421–1439. doi:10.1016/
Su, C. T., Hsu, H. H., & Tsai, C. H. (2002). Knowl- S0378-4266(98)00059-4.
edge mining from trained neural networks. Journal
Vargo, S. L., & Lusch, R. R. (2004). Evolving to a
of Computer Information Systems, 42, 61–70.
new dominant logic for marketing. Journal of Mar-
Svancara, J., Kralova, Z., & Blaho, M. (2012). keting, 68, 1–17. doi:10.1509/jmkg.68.1.1.24036.
Optimization of HMLV manufacturing systems
Vasant, P. (2006). Fuzzy decision making of
using genetic algorithm and simulation. Inter-
profit function in production planning using S-
national Review on Modelling and Simulations,
curve membership function. Computers & Indus-
5(1), 482–488.
trial Engineering, 51(4), 715–725. doi:10.1016/j.
Sweet, J. A. (1988). Measuring the accuracy of cie.2006.08.017.
diagnostic system. Science, 240, 1285–1293.
doi:10.1126/science.3287615 PMID:3287615.
35Customer Relationship Management and Data Mining
Vasant, P. (2012). Solving fuzzy optimization Zeithaml, V. A., Rust, R. T., & Lemon, K. N.
problems of uncertain technological coefficients (2001). The customer pyramid: Creating and serv-
with genetic algorithms and hybrid genetic al- ing profitable customers. California Management
gorithms pattern search approaches. Innovation Review, 43(4), 118–142. doi:10.2307/41166104.
in Power, Control, and Optimization: Emerging
Zhu, D., Porter, A., Cunningham, S., Carlisie, J.,
Energy Technologies. doi:10.4018/978-1-61350-
& Nayak, A. A. (1999). Process for mining science
138-2.
& technology documents databases, illustrated for
Vasant, P. (2013). Meta-heuristics optimization the case of “knowledge discovery and data min-
algorithms in engineering, business, economics, ing”. Journal of Information Science, 28(1), 7–14.
and finance. Hershey, PA: IGI-Global.
Vasant, P., Elamvazuthi, I., & Webb, J. F. (2010).
Fuzzy technique for optimization of objective func- ADDITIONAL READING
tion with uncertain resource variables and tech-
Ahmed, S. R. (2004). Applications of data min-
nological coefficients. International Journal of
ing in retail business. Information Technology:
Modeling, Simulation, and Scientific Computing,
Coding and Computing, 2, 455–459.
1(3), 349–367. doi:10.1142/S1793962310000225.
Ahn, J. H., Han, S. P., & Lee, Y. S. (2006). Cus-
Vasant, P., Ganesan, T., Elamvazuthi, I., & Webb,
tomer churn analysis: Churn determinants and
J. F. (2011). Fuzzy linear programming for the
mediation effects of partial defection in the Ko-
production planning: The case of textile firm.
rean mobile telecommunications service industry.
International Review on Modelling and Simula-
Telecommunications Policy, 30(10-11), 552–568.
tions, 4(2), 961–970.
doi:10.1016/j.telpol.2006.09.006.
Veglio, V. (2013). The strategic importance of
Ahn, W. H., Kim, W. J., & Par, D. (2004). Content-
data mining analysis for customer-centric mar-
aware cooperative caching for cluster-based web
keting strategies. In Customer Centric Marketing
servers. Journal of Systems and Software, 69(1-
Strategies: Tools for Building Organizational
2), 75–86. doi:10.1016/S0164-1212(03)00068-2.
Performance. Hershey, PA: IGI Global.
Berger, P. D., Bolton, R. N., Bowman, D., Briggs,
Wang, K., Zhou, S., Yang, Q., & Yeung, J. M. S.
E., Kumar, V., Parasuraman, A., & Creed, T.
(2005). Mining customer value: From associa-
(2002). Marketing actions and the value of cus-
tion rules to direct marketing. Data Mining and
tomer assets. Journal of Service Research, 5,
Knowledge Discovery, 11, 57–79. doi:10.1007/
39–54. doi:10.1177/1094670502005001005.
s10618-005-1355-x.
Bhattacharya, A., Abraham, A., Vasant, P., &
Woo, J. Y., Bae, S. M., & Park, S. C. (2005). Vi-
Grosan, C. (2007). Evolutionary artificial neural
sualization method for customer targeting using
network for selecting flexible manufacturing
customer map. Expert Systems with Applications,
systems under disparate level-of-satisfaction of
28, 763–772. doi:10.1016/j.eswa.2004.12.041.
decision maker. International Journal of Inno-
Xu, Y., Yen, D. C., Lin, B., & Chou, D.
vative Computing, Information, & Control, 3(1),
C. (2002). Adopting customer relationship
131–140.
management technology. Industrial Manage-
ment & Data Systems, 102(8/9), 442–452.
doi:10.1108/02635570210445871.
36Customer Relationship Management and Data Mining
Bhattacharya, A., & Vasant, P. (2007). Soft-sens- Elamvazuthi, I., Ganesan, T., & Vasant, P. (2011).
ing of level of satisfaction in TOC product-mix A comparative study of HNN and Hybrid HNN-
decision heuristic using robust fuzzy-LP. Euro- PSO techniques in the optimization of distributed
pean Journal of Operational Research, 177(1), generation power systems. In Proceedings of
55–70. doi:10.1016/j.ejor.2005.11.017. the 2011 International Conference on Advanced
Computer Science and Information Systems, (pp.
Cebi, S., Kahraman, C., & Kaya, I. (2012).
195-199). Jakarta, Indonesia: IEEE Press.
Soft computing and computational intelligent
techniques in the evaluation of emerging energy Elamvazuthi, I., Ganesan, T., Vasant, P., & Webb,
technologies. Innovation in power, control, and J. F. (2009). Application of a fuzzy programming
optimization: Emerging energy technologies. technique to production planning in the textile
Hershey, PA: IGI Global. industry. International Journal of Computer Sci-
ence and Information Security, 6(3), 238–243.
Chang, C. T. (2010). An approximation approach
for representing S-shaped membership functions. Feinberg, F. M., Krishna, A., & Zhang, Z. J.
IEEE Transactions on Fuzzy Systems, 18(2), (2002). Do we care what others get? A behaviorist
412–424. approach to targeted promotions. JMR, Journal of
Marketing Research, 39, 277–292. doi:10.1509/
Chen, G. Q., Wei, Q., Liu, D., & Wets, G. (2002).
jmkr.39.3.277.19108.
Simple association rules (SAR) and the SAR-
based rule discovery. Computers & Industrial Ganesan, T., Vasant, P., & Elamvazuthi, I. (2011).
Engineering, 43(4), 721–733. doi:10.1016/S0360- Solving engineering optimization problems with
8352(02)00135-3. KKT Hopfield neural networks. International Re-
view of Mechanical Engineering, 7(7), 1333–1339.
Chen, W. C., Hsu, C. C., & Hsu, J. N. (2011).
Optimal selection of potential customer range Ganesan, T., Vasant, P., & Elamvazuthi, I. (2012).
through the union sequential pattern by using a Hybrid neuro-swarm optimization approach for
response model. Expert Systems with Applications, design of distributed generation power system.
38, 7451–7461. doi:10.1016/j.eswa.2010.12.078. Neural Computing & Applications. doi:10.1007/
s00521-012-0976-4.
Coviello, N. E., Brodie, R. J., Danaher, P. J., &
Johnston, W. J. (2002). How firms relate to their Grönroos, C. (1994). Qua vadis, marketing? To-
markets: An empirical examination of contempo- ward a relationship marketing paradigm. Journal
rary marketing practices. Journal of Marketing, of Marketing Management, 10(5), 347–360. doi:
66, 33–47. doi:10.1509/jmkg.66.3.33.18500. 10.1080/0267257X.1994.9964283.
Dieu, V. N., & Ongsakul, W. (2012). Hopfield Hasuike, T., & Ishii, H. (2009). Product mix prob-
lagrange network for economic load dispatch. In lems considering several probabilistic conditions
Innovation in power, control, and optimization: and flexibility of constraints. Computers & Indus-
Emerging energy technologies. Hershey, PA: IGI trial Engineering, 56(3), 918–936. doi:10.1016/j.
Global. cie.2008.09.006.
Dostál, P. (2013). The use of soft computing for Johnson, M. D., & Selnes, F. (2004). Customer
optimization in business, economics, and finance. portfolio management: Toward a dynamic theory
In Meta-heuristics optimization algorithms in of exchange relationships. Journal of Marketing,
engineering, business, economics, and finance. 68, 1–17. doi:10.1509/jmkg.68.2.1.27786.
Hershey, PA: IGI Global.
37Customer Relationship Management and Data Mining
Khan, M. A., & Ansari, A. Q. (Eds.). (2012). Sadrnia, A., Nezamabadi-Pour, H., Nikbakht,
Handbook of research on industrial informatics M., & Ismail, N. (2013). A gravitational search
and manufacturing intelligence: Innovations and algorithm approach for optimizing closed-loop
solutions. Hershey, PA: IGI Global. logistics network. Meta-heuristics optimization
algorithms in engineering, business, economics,
Leng, K., & Wang, Y. (2012). Research on capacity
and finance. Hershey, PA: IGI Global.
allocation in a supply chain system based on TOC.
Lecture Notes in Electrical Engineering, 140, Senvar, O., Turanoglu, E., & Kahraman, C.
517–524. doi:10.1007/978-3-642-27296-7_79. (2013). Usage of metaheuristics in engineering: A
literature review. In Meta-heuristics optimization
Liu, S. (2011). T. Solution of fuzzy integrated
algorithms in engineering, business, economics,
production and marketing planning based on
and finance. Hershey, PA: IGI Global.
extension principle. Computers & Industrial
Engineering, 63(4), 1201–1208. doi:10.1016/j. Subramaniam, L. V., Faruquie, T. A., Ikbal, S.,
cie.2012.07.010. Godbole, S., & Mohania, M. K. (2009). Business
intelligence from voice of customer. In Proceed-
Loveman, G. W. (1998). Employee satisfac-
ings of IEEE International Conference on Data
tion, customer loyalty and financial perfor-
Engineering. IEEE Press.
mance. Journal of Service Research, 1, 18–31.
doi:10.1177/109467059800100103. Svancara, J., Kralova, Z., & Blaho, M. (2012).
Optimization of HMLV manufacturing systems
Peacock, P. R. (1998). Data mining in marketing:
using genetic algorithm and simulation. Inter-
Part 1. Marketing Management, 7, 9–18.
national Review on Modelling and Simulations,
Peidro, D., & Vasant, P. (2011). Transportation 5(1), 482–488.
planning with modified s-curve membership func-
Van den Poel, D., & Larivière, B. (2004). Cus-
tions using an interactive fuzzy multi-objective ap-
tomer attrition analysis for financial services using
proach. Applied Soft Computing, 11, 2656–2663.
proportional hazard models. European Journal
doi:10.1016/j.asoc.2010.10.014.
of Operational Research, 157(1), 196–217.
doi:10.1016/S0377-2217(03)00069-9.
Pettit, R. (2002). The state of CRM: Address-
ing efficiencies and the Achilles’ heel of CRM.
Varadarajan, R., & Jayachandran, S. (1999).
Business Intelligence Advisory Service Executive
Marketing strategy: An assessment of the state
Report, 2(3).
of the field and outlook. Journal of the Acad-
emy of Marketing Science, 27(2), 120–143.
Purnomo, H. D., & Wee, H. (2013). Soccer
doi:10.1177/0092070399272002.
game optimization: An innovative integration of
evolutionary algorithm and swarm intelligence
Vasant, P. (2010). Hybrid simulated annealing
algorithm. In Meta-heuristics optimization algo-
and genetic algorithms for industrial production
rithms in engineering, business, economics, and
management problems. International Journal
finance. Hershey, PA: IGI Global.
of Computational Methods, 7(2), 279–297.
Rubin, M. (1997). Creating customer-oriented doi:10.1142/S0219876210002209.
companies. Prism, 4(4), 5–27.
38Customer Relationship Management and Data Mining
Vasant, P. (2012). A novel hybrid genetic algo- Verhoef, P. C. (2003). Understandingthe effect of
rithms and pattern search techniques for industrial customer relationship management efforts on cus-
production planning. International Journal of tomer retention and customer share development.
Modeling, Simulation, and Scientific Computing. Journal of Marketing, 67, 30–45. doi:10.1509/
DOI: 10.1142/S1793962312500201 jmkg.67.4.30.18685.
Vasant, P., & Barsoum, N. (2009). Hybrid genetic Viaene, S., Baesens, B., Van Gestel, T., Suykens,
algorithms and line search method for indus- J. A. K., Van den Poel, D., & Vanthienen, J. et al.
trial production planning with non-linear fitness (2001). Knowledge discovery in a direct marketing
function. Engineering Applications of Artificial case using least square support vector machines.
Intelligence, 22(4-5), 767–777. doi:10.1016/j. International Journal of Intelligent Systems, 16(9),
engappai.2009.03.010. 1023–1036. doi:10.1002/int.1047.
Vasant, P., & Barsoum, N. (2010). Hybrid pattern Vo, D. N., & Schegner, P. (2013). An improved
search and simulated annealing for fuzzy produc- particle swarm optimization for optimal power
tion planning problems. Computers & Mathemat- flow. In Meta-heuristics optimization algorithms
ics with Applications (Oxford, England), 60(4), in engineering, business, economics, and finance
1058–1067. doi:10.1016/j.camwa.2010.03.063. (pp. 1–40). Hershey, PA: IGI Global.
Vasant, P., Bhattacharya, A., Sarkar, B., & Whiting, R. (2001). CRM’s realities don’t match
Mukherjee, S. K. (2007). Detection of level of hype. Information Week, 79-80.
satisfaction and fuzziness patterns for MCDM
Wright, P. (1986). Schemer schema: Consum-
model with modified flexible S-curve MF. Ap-
ers’ intuitive theories about marketers’ influence
plied Soft Computing Journal, 7(3), 1044–1054.
tactics. In Proceedings of Advances in Consumer
doi:10.1016/j.asoc.2006.10.005.
Research‘13. Provo, UT: Association for Con-
Vasant, P., Ganesan, T., & Elamvazuthi, I. (2012). sumer Research.
Improved tabu search recursive fuzzy method for
Zanjani, M. S., Rouzbehani, R., & Dabbagh, H.
crude oil industry. International Journal of Model-
(2008). Proposing conceptual model of customer
ing, Simulation, and Scientific Computing, 3(1).
knowledge management: A study of CKM tools
doi:10.1142/S1793962311500024.
in British dotcoms. Word Academy of Science.
Vasant, P., Nagarajan, R., & Yaacob, S. (2004). Engineering and Technology, 38, 303–307.
Decision making in industrial production planning
Zhang, G., Hu, M. Y., Patuwo, B. E., & Indro,
using fuzzy linear programming. IMA Journal
D. C. (1999). Artificial neural networks in
of Management Mathematics, 15(1), 53–65.
bankruptcy prediction: General framework and
doi:10.1093/imaman/15.1.53.
cross validation analysis. European Journal of
Operational Research, 116, 16–32. doi:10.1016/
S0377-2217(98)00051-4.
39Customer Relationship Management and Data Mining
KEY TERMS AND DEFINITIONS Decision Trees: As a hierarchical collection
of rules that describe how to divide a large col-
Customer Purchasing Behavior: Is the set
lection of records into successively smaller group
of factors and beliefs that lead customer to make
of records. With each successive division, the
a purchase.
member of the resulting segment became more
Customer Relationship Management: Is a
and more similar to one another with respect to
coherent and complete set of methodologies for
the dependent variable.
managing relationships with current and potential
Globalization: Is a process that, from the 80s,
customers.
draws new competitive boundaries and rules.
Customer Value: Is a process that puts the
Market-Oriented Companies: Are compa-
customer’s interest first, while not excluding
nies committed to understanding the expressed
those of other stakeholders in order to create a
and latent needs of their customers and of the
competitive advantage for the firm.
others players in the market better and before
Data Mining: Is the process that uses soft
than competitors.
computing techniques to extract and identify use-
ful information and subsequently gain knowledge
from large databases.
40