APPLICATIONS OF NEURAL NETWORK IN MARKETING DECISION MAKING
Binshan Lin, Louisiana State University in Shreveport
Source: http://terabook.net/app...decision-making.html
ABSTRACT
Neural networks can be viewed as an enabling tool for marketing professionals to work smarter and achieve higher levels of effectiveness. This article reviews applications of neural networks in marketing decision making and some of the challenges of neural networks in marketing management
INTRODUCTION
One of the most exciting developments from the information technology community which has found application in business has been the development of neural networks. In recent years, neural network has been moving from research laboratories into the business world and are already at work in the world of banking and finance and elsewhere (Hawley, Johnson, and Raina 1990). Considered by some to be one of the most important technological advances of the last ten years, neural networks are particularly applicable to risk management and forecasting (Huntley 1991), where the ability to identify intricate patterns is crucial to making predictions (Smith 1992). Their significance is underlined by the fact that about eighty percent of the Fortune 500 companies have an investment in neural networks (Johnston 1991).
The business community's enthusiasm seems to be related to two perspectives. The first is the increased availability of the necessary computing power and user friendly software which allows the simplified development of neural networks by individuals with minimal knowledge of the complex processes involved. Rapidly analyzing zillions of past business transactions is actually key to many reengineering efforts in today's business environment. Businesses ranging from airlines to retailers are striving to reorganize based on a better understanding of the customers' buying patterns. Secondly, neural networks promise a breakthrough in areas where traditional computer systems have difficulty. Neural networks represent a radical attempt to break the logjam by building computers that mimic the way in which humans think.
A neural network from a marketing perspective represents a software decision tool which assists the decision makers in the selection of an appropriate response to a particular situation. In essence, neural network, like other information technologies, is shaking up traditional marketing methods. Because of the dynamic nature of marketing, it would appear that this discipline is well positioned to take advantage of neural networks through a variety of new applications. In the long run, results from applications of neural networks to the marketing domain will not only lead to a deeper understanding of fundamental marketing decision processes but also enable study of the normative aspects of marketing systems.
This paper explores the power of neural networks for marketing decision-making. The general objectives of this paper are: (1) to identify neural networks as an enabling tool for marketing professionals to work smarter and achieve higher levels of effectiveness; and (2) to delineate some of the challenges and reservations of neural networks in marketing management.
OVERVIEW OF NEURAL NETWORKS
The term "neural networks" has been in use for over 40 years, but neural networks have recently been given a formal definition: a neural network is a system of many simple processing elements that usually operate in parallel whose function is determined by network structure, connection strengths, and the processing performed by the computing elements or nodes (DARPA 1988). In this section, a very brief overview of neural networks as it is portrayed by its advocates is given. The section is designed as a primer to these ideas for the decision makers in marketing.
The concept of neural networks is based upon the way we understand the human brain is structured (Wasserman 1989). Neural networks are computer systems linking inputs with outputs in a network structure of nodes and arcs. The neural network approach is also loosely described as "connectionism" or "parallel distributed processing". They are inspired by replicating portions of what is known about the way the human brain functions. In the human brain, neurons are connected by a web of millions of neural connections in a complex network, with activity generated by impulses from one neuron to another.
The simplest form of neural networks consists solely of two layers of neurons, the input and output layers. Each input is potentially linked to each output. An optional number of intermediate layers can be inserted between the input and output layers. In formal terms a neural network model may be expressed in terms of the interconnection between its neurons. These interconnections can be regarded as weights. For a particular layer above the first input layer we therefore have each neuron functionally dependent on neurons in the layer immediately below it.
In practice, instead of being programmed with explicit instructions, a neural network is trained to perform a task by learning from real-world examples. The system "learns" by adjusting the weights of relative impact of inputs upon output, trying many combinations of weights until a good fit to the training cases is obtained. Thereafter the resulting network can be used to evaluate future cases in assisting in classification, function estimation, data compression, and similar tasks.
Learning method is the most important distinguishing factor in various neural networks. Learning depends very heavily on the correct selection of training examples. Learning takes place through a statistically based procedure of iteratively adjusting the weights. The types of learning could be either supervised or unsupervised. In supervised learning, the system developed tells the neural system what the correct answer is and determines weights in such a way that once given the input, it would generate the desired output. In unsupervised learning, the system receives only the input, and no information on the expected output. Most of the neural networks employ the supervised learning.
Neural networks have four learning goals: associating patterns, replicating patterns, classifying patterns, and recognizing patterns (Bayle 1988).
1. Associating patterns. When input-output pairs are repeatedly presented during training, the network will learn to output the elements of the pair when presented with the other.
2. Replicating patterns. A set of patterns is presented during training. The network will learn to complete any pattern that is later presented with some future missing.
3. Classification Patterns. The network is presented during training with a predetermined set of
classes to which each pattern belongs. When a similar pattern is encountered in the future, the network should correctly classify it.
4. Recognizing patterns. In this case there is no training. The system develops its own set of classes that best classifies the input patterns, This is very similar to statistical techniques such as cluster and discriminant analysis.
NEURAL NETWORKS SUPPORTING MARKETING DECISIONS
Marketing companies use their knowledge of consumer behavior to segment markets, to design marketing strategies, and to measure marketing performance (Schiffman and Kanuk 1991). Today effective marketing practice requires companies to adopt the marketing concept and effective marketing segmentation, which encompasses accurate assessment of the needs and preferences of the segment of the market to be reached by the given product, is an essential element thereof.
Such problems, especially those within the purview of marketing executives, are too complex for formulation as mathematical models. Input data may be fuzzy, trends may be unextrapolable from past experience, or crucial factors may be difficult or impossible to quantify.
In fact, neural networks have been successfully used to analyze bankruptcy prediction (Odom and Sharda 1990), bond rating (Surkan and Singleton 1990), and going-concern problem analysis (Hansen and Messier 1991). Several firms are exploring the commercial use of neural networks for predicting detecting credit card fraud (Rathbum 1993), and verifying signatures (Francett, 1989). Many firms are beginning to use neural networks to improve accuracy, reduce cost, or both. Most neural network applications address problems described by one of the following three categories: (1) pattern classification, (2) market forecasting, and (3) marketing analysis. Examples from each category follow:
1 . Pattern Classification
Classification has emerged as an important decision making tool, and has been applied to a variety of problems in marketing, including customer classification (Sharma 1994), credit scoring (Capon 1982). Many of these studies have applied the neural network approach to predict the classification of a certain case. Spiegel Inc., a leading direct-mail catalog operation, used a neural network to fine-tune its marketing decisions (Schwartz 1992). Software created by NeuralWare Inc. examined the list of people who had made just one catalog purchase, as well as whatever demographic information (such as age, income, home ownership, etc.) that Spiegel had about the customer in its database. It then compared them to customers who had purchased more than one.
The neural network identified a number of patterns that could be used to single out those customers who were most likely to be repeat purchasers, allowing Spiegel to focus their marketing efforts.
2. Market Forecasting
Time-series forecasting in demand describes the processes or phenomena by which a statistical
model generates a sequence of observations that can be extrapolates into the future, assuming that the processes are stable and the observations show a high degree of correlation.
On the other hand, neural networks can be used as an alternative to conventional statistical models since neural networks can be universal approximators. For example, the Airline Marketing Assistant/Tactician developed by BahavHeuristics Inc. uses neural networks to forecast passenger demand and allocating Nationair Canada and USAir (Hall 1992). Recently neural networks are shown to have a better forecasting power of customer loyalty in buyer- seller relationships than the conventional marketing analytic techniques (Wray and Bejou 1994).
3. Marketing Analysis
Several neural networks software packages are available in marketing analysis.
• The Target Marketing System (TMS) developed by Churchill Systems is currently in use by Veratex Corp. to search the optimal marketing strategy and cut marketing costs by removing unlikely future customers from a list of potential customers (Hall, 1992).
• MARKETING ADVISOR uses a neural network to establish weighted connections between inputs and outputs, and thus to determine the relationships between previously unencounted market profiles and the appropriate application of available marketing devices for typical grocery store items (Miller 1992).
• Customer Insight Co. has tailored HNC Software Inc.'s Database Mining Workstation software, based on neural networks, for marketing analysis (Verity 1994). The neural networks software can automatically build a model of customer behavior based on an anlysis of previous transactions.
Identifying appropriate application domains where neural networks offer advantages is an arduous task. However, neural networks offer a novel approach to the decision problems where there is no information available regarding assumption of data distributions or relationships in the categorization dilemma. An increasing number of firms are using PC-based neural network software to solve problems that used to be attacked by conventional statistical analysis. Statistical techniques and neural networks are both inductive methods. In other words, the relationship between input and output is constructed from a data set. This makes the comparison and synergy between the two almost unavoidable. For most marketing data which would be composed of both qualitative and quantitative attributes which would never exactly meet the assumptions of the conventional statistical analysis. Neural-based inductive classification approaches present an alternative. A sample of decisions along with a set of attributes on which the decisions were based are given to the modeling system which then generates an approximate model of the expert system based on the sample data.
NEURAL NETWORKS AND MARKETING INFORMATION SYSTEMS
Emergent works have identified the importance of information in marketing. Day and Wensley (1983) argue that marketing should be the area where information should reside. Hutt et al. (1988) emphasize that research on strategy formation should focus on how marketing managers utilize information in decision making. Anderson and Tushman (1990) note that a key to marketing's role in strategic planning involves gathering, processing, and communicating customer information to other functions.
Marketing information systems consist of people, equipment, and procedures to gather, sort, analyze, evaluate, and distribute needed, timely, and accurate information to marketing decision makers (Lodewyck and Deng 1993). Kotler and Armstrong (1994) have also presented a model for marketing information systems. The model comprises four major subsystems: an internal records system, a marketing intelligence system, a marketing research system, and a marketing decision support system. From the marketing management point of view, the marketing information system can be viewed in a variety of ways. Kotler and Armstrong (1994) recognized its decision-support capabilities, whereas King and Cleland (1974) viewed it as a way to engage in strategic planning. Brein and Stafford described how it could be used in developing marketing programs (1968). Higby and Farah (1991) regarded it as a tool for marketing research, planning, budgeting, analyzing different courses of action, and for reporting and control.
Neural networks change the way to use information in marketing. With such a new information technology, a company using a neural network, will eventually have affordable, near real-time access to all the raw numbers it wants. These data may be obtained from consumer credit card applications, point-of -purchase credit-card sales, and credit agency reports. The real difference among competitors will be the quality of analysis each performs and the capacity of decisions flowing from it. Neural networks help managers gather and process information, such as age, income, credit history, and products purchased.
Neural networks have been applied to a wide range of information-processing activities, such as associate memory, pattern classification and clustering, and function approximation and prediction (Pao, 1988). These applications are characterized by unstructured decision processes, multi-objectives and multiple stage decision activities. Such applications may not be efficiently supported using expert and decision support systems technologies.
A neural network can be developed to shed light on the way in which consumers respond to stimuli contained in advertising messages. Considerable research suggests that advertising executional cues can influence communications effectiveness. The levels of processing form advertisements influences outcomes typically associated with effective advertising. MacInnis, Moorman and Jaworski (1991) developed a framework that explicitly provides a linkage between executional cues to communication effectiveness through their impact on consumers' motivation, opportunity, and ability, and the levels of processing from advertisements is influenced by consumers' motivation, ability and opportunity to process brand information during or immediately after exposure to an advertisement. A simple perception-type model would postulate that consumers respond to certain characteristics of the advertising of a product with decision or intentions to purchases (Curry and Moutinho 1993).
CONCLUSIONS
The intent of this paper was to examine neural networks applications in marketing and address some perspectives for future research. This objective was sought to be accomplished by reviewing some impressive neural network applications in marketing. Problems and limitations confronting the neural networks applications to marketing were examined. In sum, it is clear that neural network is a main application area of information technology. In this regard, the article presented here hopefully sheds some light both on how neural networks can be applied in marketing decision-making and on the way that care is taken to limit the scope and ambition of models.
Because of the current availability of sophisticated computer power and almost operationally automatic software, it is quite easy to get caught up in the neural network fade. This, however, may result in blindly feeding data to a neural network simulator without considering some very crucial issues in network construction and applications. Moreover, to further the process of neural networks technology applications in marketing, the characteristics, amenable application areas, and potential benefits of neural networks must be understood.
REFERENCES