Discovering diamonds under coal piles: Revealing exclusive business intelligence about online consumers through the use of Web Data Mining techniques embedded in an analytical customer relationship management framework

Myriam Ertz Int. Journal of Engineering Research and Applications www.ijera.com
ISSN: 2248-9622, Vol. 6, Issue 2, (Part - 4) February 2016, pp.28-33
www.ijera.com 28|P a g e
Discovering diamonds under coal piles: Revealing exclusive
business intelligence about online consumers through the use of
Web Data Mining techniques embedded in an analytical customer
relationship management framework
Myriam Ertz, M.Sc*, Raoul Graf, Ph.D**
*(Department of Marketing, School of Management Sciences, University of Quebec in Montreal, Canada)
** (Department of Marketing, School of Management Sciences, University of Quebec in Montreal, Canada)
ABSTRACT
Web Mining has gained prominence over the last decade. This rise is concomitant with the upsurge of pure
players, the multiple challenges of data deluge, the trend toward automation and integration within organization,
as well as a desire for hyper segmentation. Confronted, partly or totally, with these multiple issues, companies
recourse increasingly to replicate the data mining toolbox on web data. Although much is known about the
technical aspect of WM, little is known about the extent to which WM actually fits within a customer
relationship management system, designed at attracting and retaining the maximum amount of customers. An
exploratory study involving twelve senior professionals and scholars indicated that WM is well-suited to achieve
most of the customer relationship management objective, with regards to the profiling of existing web customers.
The results of this study suggest that the engineering of WM processes into analytic customer relationship
management systems, may yield highly beneficial returns, provided that some guidelines are scrupulously
followed.
Keywords: Web mining, analytical customer relationship management, segmentation, data mining, profiling
I. INTRODUCTION
With the advent of big and large data originating
from the Internet (e.g. social media, geo-referencing,
credit card information), the World Wide Web is a
large and ever-growing database, which constitutes a
fertile area for analytical research, but does also
require increasing work on data carpentry and online
analytical systems engineering [1].
In addition to overwhelming data flows,
automated, flawless and integrated processes tend to
become the norm in the industry since businesses
tend to rely heavily on integrated processes (e.g.
Customer Relationship Management Systems, or
Enterprise Resource Planning Systems), to manage
every aspects of customer relationships [2]. Those
systems are also increasingly more fed with web data
(Web Houses) [2].
Increased and diversified data, as well as more
integrated analytical systems, enable companies to
craft hyper-segmentation strategies. The level of
targeting becomes as low as that of the individual,
with markets being fragmented into micro-segments
[3]. The level of targeting narrowed from mass
markets to unique individuals. Meanwhile, traditional
research (e.g. surveys) is time-consuming, costly and
bias prone, it is also increasingly more difficult to
administer to increasingly busy consumers [4, 5].
Given the multiple challenges of data deluge,
automation and integration trends, as well as hyper-
segmentation, that companies increasingly face, is
web mining a solution? More specifically, is it a
useful approach to sift through the large data about
existing online customers, and integrate that
knowledge into Customer Relationship Management
(hereafter, CRM) systems to carve-craft powerful
personalized strategies?
More specifically, this study seeks to answer the
following research questions: (1) to what extent do
WM methods applied to web data provide accurate
profiles of existing web customers? (2) To what
extent do WM methods applied to web data identify
strategically important existing web customers? (3)
To what extent do WM methods identify existing
web customers‟ loyalty or defection statuses?
In this study, we investigate the extent to which
Web-Mining (hereafter, WM) within an analytical
CRM framework, profiles well existing customers
who interact with a given E-commerce platform.
II. LITERATURE REVIEW
1.1. Web mining
WM refers to the automatic discovery and
extraction of information from web data [6,7]. WM
refines large, complete, integer, reliable and cheap
RESEARCH ARTICLE OPEN ACCESS

data into a time-effective manner [4]. It therefore
draws on the vast web data and overcomes traditional
market research drawbacks. WM enables one-to-one
relationships and thus mass customization [8], so that
it may contribute to achieve hyper-segmentation.
Besides, if appropriately disseminated throughout
organizational layers and divisions, WM is a valuable
part of analytical CRM (a CRM), acting as a
powerful platform for tactic and strategic decision-
making [9].
However, despite its multiple advantages, WM is
hard to implement from an operational viewpoint
[10]. Second, for an improved dissemination of WM
extracted knowledge across the organization, WM
needs to be integrated into the CRM applet of the
marketing function as well as into the broader
Knowledge Management (KM) or Business
Intelligence (BI) framework of the entire
organization, which requires typically tremendous
business process re-engineering. Third, web data are
generally very large and not always meaningful
leading to poor data quality issues if not
appropriately sifted through [11]. Fourth, WM should
not be a standalone technique randomly appended to
the already existing arsenal of data analytics
techniques. Rather, companies that adopt market-
oriented strategies to better compete on the global
marketplace, need to assess objectively and critically
the benefits of the WM methods and techniques and
their place as well as utility as inputs of the
organizational decision-making process. Eventually,
an important challenge for organizations, is to be able
to diffuse and disseminate efficiently the information
derived from WM by being holistically integrated to
the aCRM applet of CRM systems and to KM-BI
systems.
1.2. Analytical Customer Relationship Management
Systems
Xu and Walton [12] developed a typology of he
four main objectives of a CRM-enabled customer
knowledge acquisition framework: (1) profiling
existing customers; (2) Explaining the behavior of
existing customers; (3) profiling prospective
customers, and (4) Explaining the behavior of
prospective customers. The framework refers
however to an offline-based CRM process
encompassing Data-Mining, forecasting, and scoring
techniques [2]. Data from the web, and thus WM
techniques to process them, are not comprised in this
framework. This study focuses on evaluating the use
of WM methods and techniques to fulfill the first
objective of Xu and Walton‟s [12] typology of
profiling existing customers.
By drawing on Xu and Walton‟s [12]
framework, it is possible to develop a framework of a
CRM framework for web users‟ knowledge
acquisition. The overall assumption is that,
integrating the WM process into a CRM framework
is assumed to turn operational web data into
meaningful and relevant knowledge of current web
customers.
1.3. Web Mining techniques for profiling existing
web customers
In order to profile existing web customers, WM
offers a many useful methods and their respective
techniques:
 Clustering method: creating homogeneous
groups of customers (e.g. agglomerative or
hierarchical clustering, K-means, TwoStep
clustering, Kohonen network/Self-Organizing
Map, K-nearest neighbour, principal component
analysis, factor analysis) [13];
 Classification method: explain or predict the
qualitative characteristic of an individual based
on other qualitative or quantitative characteristics
of that individual (e.g. decision trees, artificial
neural networks, discriminant analysis, logistic
regression, decision rules, support vector
machines, Bayesian networks) [14,15];
 Prediction method: explain or predict the
quantitative characteristic of an individual based
on other quantitative characteristics of that
individual (e.g. decision trees, artificial neural
networks, ordinary least squares regression,
support vector machine, generalized linear
models) [14].
Clustering techniques provide clusters of customers,
which may be used to reorganize the website
according to discovered clusters, as well as models to
enrich the database by integrating the code of the
cluster in the customer database or by integrating the
model directly on the website [13]. Classification
techniques provide scores that enrich the database
since the scores are integrated in the customer
database; and models that are useful to develop
recommendation modules (e.g. intelligent agents,
recommendation systems, choice matrices) [1].
Therefore, the first research propositions read as
follows:
RP1.1: web data generated by existing web
customers are sufficiently detailed and accurate to
provide a strong basis for the creation of precise
profiles about existing web customers.
RP1.2: Clustering and classification techniques
applied to web data (e.g. web log data, search results,
and web pages) create homogeneous groups of
existing web customers.
Classification and prediction methods provide models
that enrich the database by integrating the score in the

customer database for future predictions [14]. One
important variable on which customers are generally
classified refers to the Recency, Frequency, and
Monetary (RFM) of a customer‟s purchases, which
assesses her individual value in the form of the
Customer Lifetime Value (CLV) score [16]. RFM-
based CLV and profit-cost ratios identify
strategically important customers in a web context.
According to past research, classification and
prediction techniques are well-suited to assess CLVs
based on RFM data [17]. Therefore, the next research
propositions posit that:
RP2.1: Web data generated by existing web
customers encompass enough information about the
profit-cost, the RFM of purchases made by existing
web customers, which contributes to identify
strategically important existing customers.
RP2.2: Classification and prediction methods applied
to web data predict the value of a given web
customer to identify strategically important existing
web customers.
Web Usage Mining (WUM) refers to a global
analysis that records the user‟s behavior and how she
interacts with an application from the instant she
accesses a site to the moment when she leaves the
site [14]. Besides, Web Content Mining (WCM)
refers to the analysis of the content of web pages,
whereas Web Structure Mining (WSM) refers to the
analysis of links and hyperlinks. Through these
different approaches, core classification and
clustering methods can be useful tools to determine
important customer metrics such as loyalty and
attrition statuses of one or several customers in an
attempt to categorize them. The third set of research
propositions reads therefore as follows:
RP3.1: Web data generated by existing web
customers indicate whether a web customer is loyal
to a given business or defects from that business.
RP3.2: Classification and clustering methods applied
to web data predict membership of an individual to
the loyal or defecting customer group.
III. METHODOLOGY
A questionnaire was developed based on the
research propositions to be explored. Since the design
of the study is highly exploratory, the questions were
exclusively written in an open-ended form. A
convenience sample was drawn from a pool of
potential participants. A total of twelve valid in-depth
semi-structured interviews were conducted. A
condition of eligibility was that participants had to
have a thorough knowledge of WM methods and
techniques and also a sound understanding of
business issues. About half of the respondents were
senior directors or c-level executives in public or
private organizations, while the other half consisted
mainly of scholars from IT, IS, marketing, statistics,
mathematics or engineering disciplines. The sample
is therefore heterogeneous enough to allow for a
diversity of opinion and responses. Respondents
input was tape-recorded, transcribed and analyzed
using a response matrix, in which each participant is
crossed with each research theme, corresponding to
the six research propositions.
IV. RESULTS
Both research propositions 1.1 and 1.2 appear
valid. WM methods applied to web data (e.g. web log
data) provide accurate profiles of existing web
customers. More specifically, in response to RP 1.1,
internal company data may be coupled with external
syndicated data. Both should be large, granular, of
good quality, ideally issued by logged in web users.
Regarding RP1.2, both online and offline data should
be triangulated to optimize the segmentation of
existing web customers. Besides, recent advances in
WM have made WM tools directly actionable on the
website. Visitors are segmented “on the spot” and
subsequent personalization allows for immediate and
dynamic customization. Fig. 1 summarizes he
findings related to RP1.1 and RP1.2 (answering the
first research question).
Figure 1. WM-enabled process of profiling existing
web customers
Research propositions 2.1 and 2.2 are validated.
WM techniques determine successfully web
customers‟ RFM, CLV, and thus, strategic
importance. Additional specificities need however to
be met regarding web data adequacy (cf. RP2.1)
since the RFM-based computation of CLV requires
transactional data that is easily accessible. Besides,

CLV computed with offline data can enrich the
companies‟ insight into consumers‟ specific behavior
in the future and supports the crafting of relevant E-
business strategies to maximize returns. Regarding
RP2.2, business rules can be derived by aggregating
huge datasets of CLV-based segmentations that were
obtained by means of WM techniques. Fig. 2
summarizes the key findings pertaining to the second
blog of research propositions (in answer to the
second research question).
Figure 2. WM-enabled identification of strategically
important existing web customers
Eventually, the findings lend support to RP3.1
and RP3.2 so that WM techniques enable companies
to identify existing web customers‟ loyalty as well as
defection statuses, which may constitute an additional
segmentation approach. With respect to RP3.1, data
preprocessing (i.e. filtering, selecting, cleansing,
formatting) is of utmost importance, especially for
this specific objective of loyalty and defection status
identification. Regarding RP3.2, automated methods
should be preferably used in order to compute, in
real-time, the loyalty or defection status (risk or
extent) of the customer. Such knowledge constitutes
then an opportunity to customize web page content or
structure to retain consumers who are identified as
being likely to defect; or an opportunity to display
cross-/up-/deep-selling content for consumers who
are likely to remain loyal. Multiple other forms of
surgical content engineering strategies may therefore
be implemented. Fig. 3 summarizes the findings
pertaining to the third research question.
Figure 3. WM-enabled identification of existing web
customers‟ loyalty or defection statuses
As a wrap-up, Fig. 4 summarizes the findings
related to the six research propositions altogether. At
the core of the figure, WM enables to classify
existing web customers into company-specific
categories, and therefore the consumer varies from a
business-specific low to high profile. The variables
which can be used to determine the extent to which a
consumer may exhibit more or less of a business-
specific profile, may be done in a number of ways. In
this study, we explored two approaches, which are
frequently used in business contexts: (1) the RFM-
based CLV approach, and (2) the loyalty vs.
defection probability. Others may exist though. Both
loyalty status (ranging from defection to loyalty
likelihood) and customer strategic importance
(ranging from unprofitable to profitable), may
therefore be complemented by many other attributes
and variables to form multi-attribute profiles. As with
most data analytic approaches, the variable used to
determine an existing web customers‟ business-
specific profile determines the WM-derived model,
typology or rule which then also predicts the
customers‟ future positioning on this variable.
Finally, it is also posited that the more general the
level of analysis, and thus the recourse to mass
marketing, the more likely web consumers will defect
from the company and be unprofitable. The usage of
WM enables to reach a personalization level through
granular marketing which is strongly associated with
more profitable and loyal web consumers.
Figure 4. WM-enabled process of existing web
customers profiling
V. DISCUSSION
Depending on the level of segmentation sought
by the company and, therefore, the strategic
marketing orientation, resulting from the subsequent
tactical (i.e. targeting) efforts deployed to implement
those strategies, WM has several benefits. First, it
identifies the profile of existing web customer on a
multitude of attributes. Two of such attributes were
investigated in the current study. The findings
Personalized level Granular marketing

revealed that, WM determines well the loyalty status
(defection vs. loyalty) of existing web customers; and
finally, WM evaluates equally well the strategic
importance of existing web customers (unprofitable
vs. profitable).
WM enables to fulfill the three first objectives of
a CRM, as delineated by Xu and Walton [12].
Knowledge about these three aspects is useful in
order to develop business-specific profiles ranging on
a continuum from low to high, or even in categorical
nature, given the segmentation type being sought by
the company.
VI. MANAGERIAL IMPLICATIONS
WM-powered CRM processes enable
successfully to predict existing web customers‟ future
behavior and to develop dynamic response
framework in real-time that are appropriate to the
current as well as future customer‟s profile.
Yet, managers should remain conscious of the
caveats that surround WM. First, practitioners should
be aware of the “garbage in/garbage out” issue since
all information is not necessarily good or useful to
process. An important job of comparison, judgement
and selection needs to be done prior to conduct any
WM project. Second, according to most respondents,
data produced by logged in customers appear the best
kind of data since profiles can be confidently
attached to a specific customer. Websites may
therefore benefit from implementing a technology of
user recognition which logs automatically the
customer in, without her having to do anything.
Third, companies‟ databased should also allow for
high volumes of data entries, since higher quantities
of data leverage better results. Fourth, in addition to
data quality, special care should be addressed to the
planning of the WM project, as well as to the
robustness of the WM analytical process. Adherence
to such principles may prove to yield the higher
returns on investments.
VII. CONCLUSION
This study positions itself into the growing
literature stream exploring WM engineering for
business purposes [4, 13, 14, 15, 17, 18, 19]. In that
respect, the study investigated the benefits of using
WM methods and techniques in order to reach the
specific aCRM objective of profiling customers in an
online context.
The authors considered WM methods from a
broad viewpoint, without going into the
particularities of each WM technique. The WM field
of research is also fast-evolving. Therefore, new
methods and techniques may have emerged.
Additional research could focus on investigating
broader arrays of WM methods and techniques and
also investigate some WM techniques in particular.
We only considered Xu and Walton‟s [5] aCRM
framework, but other aCRM models or typologies
may exist. It would therefore be of interest, for future
research, to investigate the extent to which WM fits
into other CRM frameworks, in general, and aCRM
ones, in particular.
Finally, a particularly promising avenue of
research is being paved by the recent technological
advances in sentiment analysis and opinion mining,
both being conflated with WM. Beyond the
transactional data-which were studied in this article-
the web should also be mined for feelings [17]. This
study remained essentially focused on facts. Yet,
future research could determine the extent to which
sentiment analysis may be usefully enacted in order
to create additional segmentation variables within the
already existing business specific classificatory
scheme. For example, conversations, opinions,
sentiments, and other content-related variables may
enable companies to enrich their existing fact-based
classification schemes with variables that are more
attitudinal and emotional in nature. WM could
contribute tremendously to further knowledge in that
regards.
VIII. Acknowledgements
The authors are thankful to the Canadian
Marketing Association (CMA) as well as to the
Marketing Research and Intelligence Association
(MRIA) for their kind support in our recruitment of
the participants for this study.
REFERENCES
[1] L. Van Wel, and L. Royakkers, Ethical
issues in web data-mining, Ethics and
Information Technology, 6, 2004, 129-140.
[2] R. Kimball, and M. Ross, The data
warehouse toolkit: The complete guide to
dimensional modeling (2nd
Ed.) (New York,
NY: Wiley and sons, 2002).
[3] J. Bousquet, Y. Lachance, S. Laferté, and F.
Marticotte, Marketing stratégique (Québec :
Chenelièere Éducation, 2007).
[4] I. Mihai, Web-Mining in E-commerce,
Annals of the University of Oradea,
Economic Science Series. 959-962, 2009.N.
K. Malhotra, Marketing research: An
applied orientation (6th
ed.) (Upper Saddle
River, NJ: Pearson Education, 2010).
[5] R. Cooley, B. Mobasher, and J.Srivastava,
Web mining: Information and pattern
discovery on the world wide web, Proc. 9th
IEEE International Conf. on Tools with
Artificial Intelligence, Newport Beach, CA,
1997.

[6] R. Kosala, H. Blockeel, Web mining
research: A survey, Proc. ACM SIGKDD
Explorations, 2(1), 2000, 1-15.
[7] B. Liu, B. Mobasher, and O. Nasraoui, Web
usage mining in B. Liu (Ed.), Web data
mining: Exploring hyperlinks, contents, and
usage data, (Berlin: Springer-Verlag, 2011)
527-603.
[8] J. Ranjan, V. Bhatnagar, Role of knowledge
management and analytical CRM in
business data mining, The Learning
Organization, 18(2), 2011, 131-148.
[9] Aggarwal, V. Mangat, Application areas of
web usage mining, Proc. 5th
International
Conference on Advanced Computing &
Communication Technologies (ACCT),
Haryana, India, 2015, 208-211.
[10] A. Reid, M. Catterall, Hidden data quality
problems in CRM implementation in H. E.
Spotts (Ed.), Marketing, technology and
customer commitment new economy,
(Springfield, IN: Springer International
Publishing, 2015) 184-189.
[11] M. Xu, and J. Walton, Gaining customer
knowledge through analytical CRM,
Industrial Management and Data Systems,
105(7), 2005, 955-971.
[12] M. Bazsalicza, and P. Naim, Data mining
pour le web: Profiling, filtrage
collaborative, personalisation client (Paris:
Eyrolles, 2001).
[13] F. Tufféry, Data-mining et statistiques
décisionnelles: L’intelligence des données
(Paris : Technip, 2011).
[14] K. J. Cios, R. W. Swiniarski, L. A. Kurgan,
Data Mining, (New York, NY: Springer,
2007).
[15] M. Jeffery, Data-driven marketing: The 15
metrics everyone in marketing should know
(Evanston, IL: Northwestern University
Press, 2010).
[16] A. lbadvi, and M. Shahbazi, Integrating
rating-based collaborative filtering with
customer lifetime value: New product
recommendation technique, Intelligent Data
Analysis, 14, 2010, 143-155.
[17] A. Wright, Mining the web for feelings, not
facts, The New-York Times, August 23, 2009.
[18] M. Ertz, and R. Graf, How do theybehave on
the web? An exploratory study of mining
web for analytical customer relationship
management, International Journal of
Electronic Commerce Studies, 6(2), 2015,
289-304.
[19] M. Ertz, R. Graf, Spotting the „elusive‟
prospect customer: Exploratory study of a
web-powered customer relationship
management framework, Journal of Applied
Business Research, 31(5), 2015, 1935-1850.

Discovering diamonds under coal piles: Revealing exclusive business intelligence about online consumers through the use of Web Data Mining techniques embedded in an analytical customer relationship management framework

More Related Content

What's hot (19)

Similar to Discovering diamonds under coal piles: Revealing exclusive business intelligence about online consumers through the use of Web Data Mining techniques embedded in an analytical customer relationship management framework (20)

Recently uploaded (20)

Discovering diamonds under coal piles: Revealing exclusive business intelligence about online consumers through the use of Web Data Mining techniques embedded in an analytical customer relationship management framework