A decision-making framework for precision marketing


A decision-making framework for precision marketing

Zhen You a, Yain-Whar Si b, Defu Zhang a,⇑, XiangXiang Zeng a, Stephen C.H. Leung c, Tao Li a,d
a Department of Computer Science, Xiamen University, Xiamen 361005, China
b Department of Computer and Information Science, University of Macau, Macau, China
c Department of Management Sciences, City University of Hong Kong, Hong Kong
d School of Computer Science, Florida International University, Miami, FL, USA

a r t i c l e i n f o

Article history:
Available online 20 December 2014

Keywords:
Data mining
Decision tree
Forecasting
Precision marketing
Decision-making

a b s t r a c t

Precision marketing offers personalized customer service and is used to help enterprises increase their
profits by means of high-efficiency marketing. This paper presents a novel decision-making framework
for precision marking using data-mining techniques. First, this study presents a trend model to accurately
predict monthly supply quantity; second, it uses a RFM (Recency, Frequency and Monetary) model to
select attributes to cluster customers into different groups; third, it uses CHAID decision trees and Pareto
values to identify important attribute values to distinguish different customer groups; and finally, it
creates different supply strategies targeting each customer group. The objective of the proposed
precision-making framework is to help managers identify the potential characteristics of different
customer categories and put forward appropriate precision marketing strategies, which can greatly
reduce inventory for every customer category. The real-world data from a company in China were
collected and used in a case study to illustrate how to implement the proposed framework. This case
study demonstrates that our proposed decision-making framework is efficient and capable of providing
a very good precision marketing strategy for enterprises.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Due to the accelerated pace of economic globalization and
increasing market competition, economic pressures and competi-
tion have led enterprise managers to face the problem of choosing
the right strategic decision-making policies for selling the right
products to the right customers at the right time, such that the
companies can increase their profits. Recently, it has been recog-
nized that precision marketing has become a key means of gener-
ating profit and is becoming increasingly important as customers
become better informed about the products and their rights as con-
sumers. The availability of customer data and transaction records
provides better understanding of customers’ consumption behav-
iors and preferences. In the increasingly competitive environment,
enterprises have to create a decision-making model for precision
marketing that provides appropriate strategies to manage the mar-
ket positioning system for fulfilling their customers’ needs. The
research motivation of this paper is stemmed from a real-world
project. This project considers a marketing problem where the

supplier or manufacturer provides different products for retail cus-
tomers, of which some may sell well in some customer segments
and some may not. Products that are not sold will be returned back
to the supplier. Therefore, the supplier needs to find a good mar-
keting strategy that minimizes goods in stock and satisfies the sup-
plier, retailers and consumers. In recent years, the decision-making
problems have received much attention due to a wide range of
real-world applications. Many decision-making techniques have
been proposed in literature. Chen and Wang (2009) presented
multi-criteria optimization and compromise solutions. Saen
(2010) developed a technique for order performance via similarity
to ideal solution. Hsu, Chiang, and Shu (2010) developed a nonlin-
ear programming for decision-making, and Lin, Chen, and Ting
(2011) presented a Linear programming for decision making.

When discussing the decision making in this context, it is
important to pay attention to the role of artificial intelligence
(AI). The decision support framework of AI is considered to be a
major tool for obtaining information related to historical data col-
lections using some artificial Intelligence technique such as genetic
algorithm (GA), artificial colony optimization (Ghasab, Khamis,
Mohammad, and Fariman, 2015), and other data-mining tech-
niques (Chai, Liu, & Ngai, 2013). Nowadays, data-mining, which
can extract useful customer information and discover the hidden
customer’s behaviors from big data, has a great influence on

http://dx.doi.org/10.1016/j.eswa.2014.12.022
0957-4174/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel.: +86 18959217108; fax: +86 0592 2580258.
E-mail addresses: uzhen@foxmail.com (Z. You), fstasp@umac.mo (Y.-W. Si),

dfzhang@xmu.edu.cn (D. Zhang), xzeng@xmu.edu.cn (X. Zeng), mssleung@cityu.
edu.hk (S.C.H. Leung), taoli@cs.fiu.edu (T. Li).

Expert Systems with Applications 42 (2015) 3357–3367

Contents lists available at ScienceDirect

Expert Systems with Applications

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / e s w a

http://crossmark.crossref.org/dialog/?doi=10.1016/j.eswa.2014.12.022&domain=pdf
http://dx.doi.org/10.1016/j.eswa.2014.12.022
mailto:uzhen@foxmail.com
mailto:fstasp@umac.mo
mailto:dfzhang@xmu.edu.cn
mailto:xzeng@xmu.edu.cn
mailto:mssleung@cityu.edu.hk
mailto:mssleung@cityu.edu.hk
mailto:taoli@cs.fiu.edu
http://dx.doi.org/10.1016/j.eswa.2014.12.022
http://www.sciencedirect.com/science/journal/09574174
http://www.elsevier.com/locate/eswa


guiding decision making and forecasting the effects of decisions.
Guo, Yuan, and Tian (2009) used a hierarchical potential support
vector machine (HPSVM) for supplier selection and improved the
accuracy. Chang and Hung (2010) presented a rough set theory
(RST) to analyze the rules of supplier selection and guide the deci-
sion making. Chai et al. (2013) conducted a systematic review of
literature about the application of decision-making techniques in
supplier selection. They classified these techniques into three cat-
egories: Multi-criteria decision making techniques (MCDM), math-
ematical programming techniques (MP), and artificial intelligence
techniques (AI). MCDM is a methodological framework that aims
to provide decision makers a knowledgeable recommendation
amid a finite set of alternatives (Chai et al., 2013). Wang, Wu,
Wang, Zhang, and Chen (2014) developed two kinds of prioritized
aggregation operators of IVHFLNs for MCDM.

However, in practice, it is not an easy task to choose a good
data mining tool since, each data-mining tool has its own advan-
tages and disadvantages. For example, artificial neural networks
(ANNs) involve too many hidden neurons and training parame-
ters (Zhang, Jiang, & Li, 2005). Its disadvantages include the
‘‘black box’’ nature, greater computational burden and proneness
to over-fitting. However, an advantage of ANN is that ANN has
the ability to implicitly detect complex nonlinear relationships
between dependent and independent variables. Decision trees
are simple to use and easy to understand and they offer many
advantages compared with other decision-making tools. How-
ever, the disadvantages of decision trees include their instability
and relatively low performance. Hence, there is no single best
model for all the cases (Akın, 2015). Recently, researchers
attempt to combine different models together by considering
their respective advantages. Guo, Wong, and Li (2013) presented
a multivariate intelligent decision-making mode which combined
three different model to complete every phase of the decision
making process. Tadić, Zečević, and Krstić (2014) developed a
novel hybrid MCDM model that combines fuzzy decision making
trial and evaluation laboratory model to provide support to deci-
sion makers. Furthermore, Yan and Ma (2015) proposed a novel
two-stage group decision-making approach to uncertain quality
function deployment.

To the best of our knowledge, the marketing problem consid-
ered in this paper is still a new problem. The objective of this paper
is to propose a decision-making framework for combining various
data-mining algorithms to achieve precision marketing of real
products. Real-world data that include historical monthly supply
(quantity) and information of every customer were collected from
a company in China. The goal is to find a model that can classify
targeted customers and predict supply quantity and then provide
a strategy for precision marketing, i.e., deciding the quantity of
products that every store needs. Depending on the different char-
acteristics and requirements of each phase of the marketing model,
the decision-making framework uses four data-mining models/
algorithms, which are K-means algorithm, decision tree, Pareto
ratio method, and RFM model, for decision-making. Overall, the
purpose of this study is to generate, using data-mining techniques,
a decision-making model for products’ precision marketing. The
proposed decision-making framework is more accurate due to its
integrated precision strategies which combine prediction models,
clustering, and classifying model. Moreover, the proposed frame-
work incorporates RFM model and Pareto ratio into the process
of customer segmentation so that the generated strategies are
more convincing.

The rest of this paper is organized as follows. Section 2
introduces related works on common data-mining models and
algorithms. Section 3 describes the methodology in our proposed
framework briefly. Section 4 presents a case study and results of
analyses. Finally, Section 5 concludes the paper.

2. Related works

Since a single data-mining model may only be suitable for a
specific problem, such as prediction or clustering or classification,
in the proposed framework, we have combined four data-mining
models or algorithms to derive a precision marketing strategy for
enterprises. The literature on these four models or algorithms
was reviewed below.

2.1. K-means algorithm

Clustering is the process of grouping a set of physical or abstract
items into classes of similar items where the groups are either
meaningful or useful, or both. A well-known clustering algorithm
is K-means, which was first proposed by MacQueen (1967). The
accuracy of this algorithm depends on the initialization and the
number of clusters (Mesforoush & Tarokh, 2013). The basic idea
of K-means is to discover k clusters, such that the records within
each cluster are similar to each other and distinct from the records
in other clusters. K-means is an iterative algorithm: an initial set of
clusters is defined and the clusters are repeatedly updated until no
further improvement is possible (or the number of iterations
exceeds a specified threshold). The K-means algorithm is widely
used to pre-process data or for clustering because of its simplicity
and efficiency (Mesforoush & Tarokh, 2013). K-means has been
widely used to effectively identify the valuable customers and
develop the related marketing strategies (Wei, Lee, Chen, & Wu,
2013; Mesforoush & Tarokh, 2013). In particular, Cheng and Chen
(2009) use the RFM model and K-means to perform customer rela-
tionship management, and the experimental results demonstrate
that their proposed model is an effective method in customer value
analysis.

2.2. Decision tree

A decision tree is an efficient data mining algorithm with a
strong explanatory capability (Zhang, Zhou, Leung, & Zheng,
2010). This study uses the Chi-squared Automatic Interaction
Detector (CHAID) decision tree, a decision tree algorithm that is a
type of database segmentation. The concept of CHAID was first
published in Kass (1980). CHAID is used for prediction and classi-
fication. Like other decision trees, CHAID’s advantages are that its
output is highly visual and easy to interpret. A recent study indi-
cates that CHAID is superior to judgment RFM for identification
of likely responders (McCarty & Hastak, 2007). CHAID is similar
to the RFM approach because it can identify the terminal nodes
that will break even with respect to expected profit and costs.
Therefore, CHAID was widely adopted used for segmenting cus-
tomers and extracting rules that show the associations between
the input and output variables (McCarty & Hastak, 2007; Chen &
Wang, 2009; Mistikoglu et al., 2015). In addition, Coussement,
Van den Bossche, and De Bock (2014) demonstrate that, when per-
forming customer segmentation at different levels of uncertainty,
CHAID outperforms RFM and logistic regression. Furthermore,
CHAID can also assess its robustness to data accuracy problems.
So in our study, we use CHAID to segment customers.

2.3. Pareto ratio

The concept of Pareto ratio is from the ‘‘Pareto Principle’’ (also
known as the 80–20 rule, the law of the vital few or the principle
of factor sparsity) which states that for many events, roughly
80% of the effects come from 20% of the causes. This is a common
rule of thumb in business. This distribution is relevant to entrepre-
neurs and business managers, stating, for example, that 80% of the

3358 Z. You et al. / Expert Systems with Applications 42 (2015) 3357–3367



https://isiarticles.com/article/41312