Advances in Engineering Research (AER), volume 116

International Conference on Communication and Electronic Information Engineering (CEIE 2016)

Data mining techniques for customer relationship management

Feng Guo
Information School of Beijing Wuzi University, Beijing
E-mail:[email protected]

Data mining has made customer relationship management (CRM) a new area where firms
can gain a competitive advantage, and it plays a key role in the firms’ management
decision. In this paper, we first analyze the value and application fields of data mining
techniques for CRM, and further explore how data mining being applied to Customer
churn analysis.

Keywords: Data mining; Customer Relationship Management (CRM); Customer Loyalty.

1. Introduction
A new business culture is developing today. The conventional production
centered and sales purposed market strategy is gradually shifting to customer
centered and service purposed. Customers’ value orientation is increasingly
affecting the firms’. And customer resource has become one of the most
important strategic resources. Therefore, understanding customers’ needs and
discriminating the most contributed customers has become the driving force of
most modern business.

2. Customer Relationship Management

Customer relationship management (CRM) is a customer centered business
strategy. Firms abidingly accumulate, analyze and exploit customer
comprehensive information to allocate resources rationally according to
customer oriented working mode, and meet customer demand through
meaningful communication, understanding and influence customer behavior.
Thereby increasing customer satisfaction and loyalty, and to obtain the
maximum profit. The general CRM covers customer resource management,
marketing management, sales management, customer care and service
management, and many other aspects. Its ultimate goal is to provide the basis for
business management and decision-making through proactive market research
and analysis and processing of customer feedback. Simple from the content of

Copyright © 2017, the Authors. Published by Atlantis Press. 9

This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).
 Advances in Engineering Research (AER), volume 116

customer relationship management, CRM is customer propaganda, tracking,

revisit and classification. Of course, with the development of modern
information technology, the pursuit of personalized service has become the new
target in the field of after-sales service.

3. Overview of Data Mining Techniques

Firms want to acquaint customer behavior as much as possible, but as the
number of customers accumulated in large quantities and customer information
has become increasingly complex, more advanced technologies and tools are
needed to manage and analyze customer information so as to find out the
valuable knowledge of management decision-making. Data mining technology
provides a good technical support for CRM to analyze large amounts of complex
customer data and explore customers’ value.

3.1. Definition
Data mining can extract potentially valuable knowledge, model and rules from
mass of data. It is an effective tool to exploit potential associations and patterns.
With the help of statistics and artificial intelligence technology, data mining can
make further analysis of data and find out knowledge, and build different models
in accordance with different business issues, thus provide a technical reference
for the firms’ decision-making. Data mining involves a considerable number of
subject areas, most of them are mature algorithms and techniques in artificial
intelligence such as machine learning, statistics, neural network and database
techniques. According to the purpose of data mining, it can be divided into
association analysis, time series analysis, classification, cluster analysis and data
summary.

3.2. Functionality
Through the prediction for future trends and behavior, data mining can help
firms make a forward-looking and knowledge-based decision-making. The goal
of data mining is to find implied and meaningful knowledge from database. Its
functionality is to designate type of pattern that data mining tasks are looking for,
and can be divided into two categories: description and prediction. Descriptive
mining tasks characterize the general characteristics of the data in the database,
while predictive ones extrapolate the existing data to predict. Specifically, firms
can discover most valuable customers, make the combined sales more efficient
and find fraud with a smaller cost with the help of data mining techniques.

10
 Advances in Engineering Research (AER), volume 116

3.3. Methods
As an advanced information processing technology, the essential difference
between data mining and traditional data analysis is that it is a process of
exploration of data relationships, and in most cases with no assumptions and
premises. There are numbers of different ways in data mining for users to
embark on a comprehensive analysis of data from different perspectives.
(1) Correlation and regression analysis
Correlation analysis mainly analyzes the closeness of variables, and
regression analysis mainly establishes dependencies between variables based on
observational data. Both of them can reflect valuable associations and contacts
among variables. So they can be collectively referred to as association analysis.
(2) Time series analysis
Similar to association analysis, time series analysis is aimed at tap the
intrinsic link between data. The difference is that time series analysis focuses on
chronological causal relationship of data, while association analysis focuses on
parallel relationship.
(3) Classification and prediction analysis
Classification and prediction analysis are used to extract models describing
important data. The model is then applied to determine and classify new
observations, or predict future data trends.
(4) Cluster analysis
Cluster analysis categorizes data into more than one classes or clusters
according to certain characteristics. Objects must have a high degree of
similarity if they are from a same cluster, and the difference will be much larger
if not. We could see that the cluster analysis is the reverse process of
classification and prediction to a certain extent.

4. Value of Data Mining for CRM

CRM collect and accumulate a large number of market and customer
information for the firms to build a huge data warehouse. The key to data mining
is to distinguish truly valuable information and find the association between
them from massive data.
Every firm wants to quickly and accurately extract the most important and
valuable customer information through the use of certain tools and instruments.
After a deep analysis for a huge data warehouse, data mining techniques can
find hidden information and knowledge conducive to business operations, thus
improve the competitiveness of the firm. Data mining can help firms manage
customer life cycle stages, including acquiring new customers, making existing

11
 Advances in Engineering Research (AER), volume 116

customers create more profits, maintaining valuable customer, etc. It can do in-
depth analysis of customer needs, to meet firms’ needs of customer relationship
in individual market segments. So the firms are able to determine customer
characteristics, provide them with targeted services and increase business
opportunities.

5. Application Fields Of Data Mining For CRM

Data mining can filter customer information, and mining implicit, unknown and
potentially valuable knowledge on business decisions and rules from a huge
number of customer data. To help firms identify new opportunities, predict
success in business strategies and make quick decisions. The following aspects
of its application are particularly prominent.

5.1. Customer character analysis

In addition to customers’ address, gender, age, occupation, income, education
and other basic information, access, such as hobbies, marriage, spouse, health,
home environment and other features can help firms get a more detailed
understanding of customers, observe their behavioral regularities and thus better
develop customer strategies to improve their campaign response speed. The
main idea is to use classification and clustering techniques, divide customers
into different group with different demand and transaction practices according to
their age, gender, income, trading behavior, etc. And finally arrive at customers’
concern, targeted to the development of personalized marketing strategy.

5.2. Customer loyalty analysis

Customer loyalty analysis is to analyze and categorize upscale customers, stable
customers, valuable customers, more consumption demand customers, etc. To
help firms make customers persistent and stability analysis and respond quickly
to customer needs, so that each customer will get a highly personalized service.
Statistically, firms must pay much higher cost to get new customers than to keep
old customers. The gap is recognized to be more than six to eight times, no
matter what business they focus on. Meanwhile, according to 80/20 principles of
marketing, that 20% of customers contributed 80% of sales, more strategies
should be adopted to develop high consumption customers’ loyalty.

12
 Advances in Engineering Research (AER), volume 116

5.3. Cross marketing analysis

The previously customer information may contain the key or even critical
factors to determine customer behavior. Using data mining techniques, firms can
get the key factors affecting customer purchasing behavior from the customer
information especially former purchase behavior, and build prediction model to
predict customer future purchase behavior. Association analysis is often used to
help companies find some implicit, subtle and great commercial value of
relationship. This mining process is generally divided into two steps: First, find
all the frequent item sets whose frequency should be at least equal to the
minimum support frequency, thereby identify all related merchandises that may
be purchased together; Then, generate strong association rules according to the
frequent item sets. These rules must meet the minimum threshold of confidence;
thereby identify all related merchandises that are very likely to be purchased
together. To sum up, firms can discover hidden relationship between seemingly
independent events; find factors that affect customer behavior through data
mining technology. And then target to expand marketing strategy and promote
other products in a timely manner.

5.4. Customer acquisition and churn analysis

The growth and expansion of the firm need continuously maintain old customers
and draw new customers. Through classification, clustering, decision tree and
other techniques, it can extract mass customer information, identify potential
customer, and determine what kind of customer are most likely to loss and what
features they have. Thus build customer churn prediction model to more
accurately identify easily lost customers. This would help firms making plans to
take appropriate marketing measures in advance to hold old customers.

5.5. Tracking evaluation

Through clean and centralize, customers’ feedbacks are automatically store in
data warehouse. So the firm can track customer behavior, analyze customer
satisfaction, credit rating and so on, in order to evaluate and optimize the
existing strategy, such as take different credit terms on different credit rating
customers. This could maintain customer loyalty and avoid unnecessary risks at
the same time. By tracking the evaluation to ensure that the firm customer
relationship management to achieve the goals, also establish a good customer
relationship.

13
 Advances in Engineering Research (AER), volume 116

6. Customer Churn Analysis

Statistically, firms must pay much higher cost to get new customers than to keep
old customers. The gap is recognized to be more than six to eight times, no
matter what business they focus on. So firms should take strategy to develop
high consumption and high loyalty customers’ loyalty, and use data mining
techniques to identify customers that are likely to loss and take measures to
prevent the loss. Commercial enterprises usually record sales information and
customer information through transaction records, sweepstakes, membership and
other activities. And then dig out customer preferences merchandise mixes,
customer habits and characteristics of lost customers from this information. The
following will take customer churn analysis as an example to illustrate the
process of data mining in CRM.

6.1. Data selection

The target variables are customers’ loss states. In another word, by analyzing a
specific problem, select a combination of one or a plurality of known amount
variables as target variable. What specific consumption characteristics lead to
customer loss? Is that income change? Or preference change? Or the demand of
shopping environment change?
The input arguments include static and dynamic data. The static data is the
data that changes infrequently, such as the customer basic information (gender,
marital status, occupation, residential areas, etc.); the dynamic data is the data
that changes frequently or regularly, such as the consumption amount in a month,
consumption characteristics, etc. There are intrinsic links between the input
arguments and the target variables, but cannot be quantified. This provides data
mining a play space. If we are unable to determine whether some sort of data is
associated with customer churn probability, we could temporarily select it in the
model, and examine each variable distribution and correlation in the subsequent
step to determine whether to keep it.
Select the modeling data. There two kinds of customer churn: subside and
transfer to competitors. The formal is always caused by firms’ relocation, or
customer service upgrade such as ordinary customers upgrade to the VIP. But
the latter is what firms really care about. We must choose the latter loss of
customer data to carry out the modeling.

14
 Advances in Engineering Research (AER), volume 116

6.2. Data reduction

Take data reduction before building the model, on the one hand, can ensure the
correctness and validity of the modeling data, and on the other hand, can make
the data more in line with the modeling needs through the adjustment of data
format and content. The main tasks include data conversion, integration,
sampling, randomization, missing value handling, etc. Such as proportionally
extract the information of existing customers and lost customers and merge these
two types of data as data source for modeling. The model need large amount of
data to test, so the data sample is divided into two part, one for modeling and the
other for model checking and correction.

6.3. Model building

Before modeling, we should use correlation analysis to identify the correlation
between each input argument and the probability of loss of customers, delete the
less correlation variable, thereby reducing the degree of complexity of the model,
thus shorten the modeling cycle. There exist many data mining methods, such as
neural networks, regression, decision tree, association analysis, clustering, etc.
We can use a variety of methods pre-build some models then compare the pros
and cons of these models and select the most suitable for customer churn
analysis. There are also a variety of approximation algorithms to simplify the
model optimization process. The core part of data mining process is adjustment,
which is usually complete by data analysis experts. Different business issues and
data distribution properties will affect the modeling and adjust its strategy;
therefore business experts are required in order to avoid inappropriate
optimization resulting in loss of business information.

6.4. Model evaluation and interpretation

To obtain accurate result, we can use the data not involved in the modeling for
model evaluation. The evaluate matric is to use the mode to predict the already
known customer state, and compare the predicted result and the actual customer
state. The model with the highest prediction accuracy is the best. Through
making a reasonable business explanation to the model, firms can identify some
potential law to guide businesses behavior. In turn, the explanation can also
prove the correctness and effectiveness of the mathematical model.

15
 Advances in Engineering Research (AER), volume 116

6.5. Model application

To make the model practical, we can first select a pilot to implement the model,
and pay attention to the earnings of the model application during the pilot. If an
exception occurs, stop the application and correct the model. After the end of the
pilot, if the model is proved to be good, then it can consider a wide range of
promotion. The promotion should pay attention to regional differences, and
cannot just copy the model. The model bias may increase with the growth of
time and significant changes in economic, and then we should consider re-
establish a more applicable model.
The churn model can monitor the loss of customer, if the possibility of loss
is too high, firms can take measures to increase customer loyalty, and prevent
the occurrence of loss of customers, which completely change the situation that
firms cannot monitor the loss of customers and cannot effectively achieve
customer care even they successfully get many customers.

7. Conclusion
The CRM system using data mining technology has become the key for firms to
winning in the increasingly competitive knowledge economy and e-commerce
economy mode. Data mining as the premise and foundation of CRM is no longer
just confined to customer contact level customer relationship management. On
the contrary, it has been deep into customer behavior and customer preferences,
and can anticipate their needs and meet their expectations. Thus firms can
insight into customer, understand customer value from a more in-depth and
comprehensive perspective. Effective use of data mining in CRM, can
continually promote a single customer value enhancement and expansion of the
customer base, guide high-level decision makers to work out best marketing
strategy, reduce operating costs, increase profits, strengthen customer
relationships, improve customer satisfaction, improve, improve commodity
promotions, enhance information exchange of goods, improve customer loyalty,
accelerate the development of firms, and enable firms to gain long-term benefits.

References
1. Xianyan Lin. A Research on Telecom Customer Loss Based on Data
Mining [J]. Journal of Mobile Communication, 34(8). P71-75.
2. Guoen Xia. Customer Churn Prediction Based on Simple Support Vector
Machine [J]. Research on Application of Computer, 2010, 27(3).

16
 Advances in Engineering Research (AER), volume 116

3. Liuyin Guo, Zhi Fang. Application of Data Mining on Retaining Customers

in Telecom Industry [J]. China New Telecommunications, 2005(11).
4. Hongxia Wang, Gencai Chen. Data Mining Techniques Applied in
Retailing’s CRM [J]. Computer and Digital Engineering, 2006(1).
5. Alex Berson, Stephen Smith, Kurt Theading. Building Data Mining
Applications for CRM [M]. Beijing: People’s Posts and
Telecommunications Press, 2001.
6. MeiZhen Zhu. To Promote Customer Relationship Management Data
Mining [J]. Technology and Innovation Management, 2006(4).
7. Sheng Tian. The Application of Data Warehouse and Data Mining in CRM
[J]. Journal of Information, 2005, 24(9). P18-20.
8. Ning Dong. Data Mining Techniques for CRM [J]. Computer Engineering
and Design, 2007, 28(6).

17