Target markets and ANN

Once we, marketers, have determined our business products or services, we must accurately identify our target markets. The target market is the actual customer group, or audience, in which our business will attempt to sell its products and services. Target marketing tailors a marketing mix for one or more segments identified by market segmentation and usually contrasts with mass marketing, depending on the industry, the competition, the marketing budget of the business, NGO vs non-NGO, etcetera. Many techniques have been used in order to select the target markets, such as statistical regression (Bult&Wansbeek, 1995), neural computing (Zahavi&Levin, 1997) and fuzzy clustering (Setnes&Kaymak, 2001). Direct marketing is, off course, a well-known type of marketing, used by businesses, in order to target, call to action and increase profits. A well-executed direct advertising campaign can have a satisfying Return On Investment, since many potential customers responded to a clear call-to-action. The growing interest in direct marketing techniques has made it an important application field for marketers and data mining professionals.

CHAID vs ANN

Artificial neural networks (ANN) are "computational models inspired by animals' central nervous systems that are capable of machine learning and pattern recognition". They are presented as systems of interconnected neurons that can compute values from inputs by feeding information through the network. 

Commonly, a class of statistical models may be called "neural" if they

1. consist of sets of adaptive weights, i.e. numerical parameters that are tuned by a learning algorithm, and
2. are capable of approximating non-linear functions of their inputs.

The adaptive weights are conceptually connection strengths between neurons, which are activated during training and prediction.

As aforementioned, marketers should divide the market into segments. A product offer would then be made to those people who are considered to be in the market segment for which the product was meant. Automation segmentation techniques have been used to divide the customers into a number of segments, like CHi-squared Automatic Interaction Detection (CHAID). CHAID is a type of decision tree technique, that is used for prediction as well as classification and detection of interaction between marketing variables. In practice, CHAID is often used in the context of direct marketing to select groups of consumers and predict how their responses to some variables affect other variables.

When a segmentation method such as CHAID is used to determine target segments which are more likely to respond to marketing actions, the data set is split into disjoint groups and the groups that are considered to be most likely to respond are targeted through a marketing action, e.g. mailed preferentially. Neural networks, however, do not segment data into disjoint groups. That means, a different approach is needed to develop target selection models by using neural networks. Since target selection models try to differentiate responders from non-responders, marketers could consider formulating the problem as a classification problem and develop a network that classifies each input pattern as a responder or non-responder. However, the classes are not well separated in target selection problems.  

Furthermore, the miss-classification costs  are asymmetric. Miss-classifying a possible responder as a non-responder and not targeting that individual is more costly to the direct marketing company than miss-classifying a non-responder as a responder. Therefore, another method than pure classification should be used. One solution could be to define a measure for the "likelihood" of an individual to respond (call-to-action) via the marketing action, e.g. mailing. The individuals are then ordered according to this measure, and only the ones that score above a given threshold θj are targeted (target scoring). 

Artificial neural network models for target selection fall into this category. The output of the neural network is then a measure of likelihood of response (Potharst et al., 2001). Now, the goal of determining the neural network would be to determine a correct set of network parameters (weights) such that a good indication of the likelihood of response of the supporters is obtained, given the inputs to the network. Naturally, these inputs must convey information about the characteristics of the supporters, including demographic, geographic, or even psychographic and behavioral variables.

ANN configuration and analysis

In target selection problems we form the model of the likelihood of response at a single point in time, given various supporter characteristics and/or their past behavior. In real life, feed-forward neural networks are sufficient for target selection problems. The nonlinear activation function of the neurons also has an influence on the final neural network model for a given network structure and the number of neurons. The logistic sigmoid function:

is often used since the neural network output is expected to lie in the range [0,1] (likelihood of response). Thus, the neural network model can be analyzed as a generalized nonlinear regression model with the likelihood [0,1] of response to our marketing action as an output .

We may visualize our target selection model through a hit probability chart. 

Potharst et al. 2001

A hit probability chart exposes what percentage of a selected population may respond to our marketing action. The x-axis shows a percentage of the total group selected for targeting and the y-axis shows what percentage of that group is a responder to our call-to-action. The hit probability chart for a successful model starts with high values of hit probability. As the size of the selected group increases, the percentage of responders within the selected group will decrease until the hit percentage is equal to the percentage of responders within the total marketing action eg. mailing considered in the campaign.

In real life, after data preparation and data set selection, marketers may use IBM SPSS Neural Networks as a modern modelling tool.  

With SPSS Neural Networks, marketers may select either the Multilayer Perceptron (MLP) or Radial Basis Function (RBF) procedure. Both of these are map relationship techniques implied by the data. Both use feedforward architectures, meaning that data moves in only one direction, from the input nodes through the hidden layer of nodes to the output nodes. Prior research has indicated that neural network models may perform better than the CHAID model, while they perform on an equal level with the logistic regression models we aforementioned.

Conclusion

In general, ANN for Marketing can be viewed as a modelling tool that helps marketers to take decisions. ANN can recognize patterns, pick up key information that are not easily identifiable and develop relationships among them. Marketers are expected to fully take advantage of the ANN modeling in order to solve their target selection problem faster and more efficiently.

Read more »

Intro to Data Mining for Marketers - Part 2

Data Preprocessing

This stage is the most time-consuming stage of the data mining process. Data is never clean and in a form suitable for data mining. There are few typical data corruption problems in business databases such as duplication of the records, missing data fields, and presence of outliers. The preprocessing step involves integrating data from different sources and making choices about representing or coding certain data fields that serve as inputs to the data discovery stage. Such representation choices are needed because certain fields may contain data at level of details not considered suitable for the pattern discovery stage. For example, it may be counterproductive to represent the actual date birth of each customer to the data discovery stage. Instead, it may be better to group customers  into different age groups and  the chosen age groups should have some significations for the research goal. It is important to remember that the preprocessing stage is a crucial step. The representation choices made at this stage have a great bearing on the kinds of the patterns that will be discovered by the next stage of data discovery.

 Patterns & Market Segmentation

Since there are so many ways we, human beings, are different, it should not be surprising that we would differ in our needs for automobiles. While there are many factors/variables that contribute to these differences, we are considering the following factors for presenting our data mining framework for the aforementioned example: vehicle image (Table 1), customer anticipated feelings (Table 2),  and  demographics (such as age, sex, income, occupation, education etc). The demographic factor plays an important role in the proposed analysis. 

For example, consider how customer needs and preferences for an automobile change as one moves demographically from college student to management trainee; changes in income, occupation, and educational status each contribute to a changing set of customer needs for a variety of products such as an automobile. Many other variables can be incorporated as well.

There are mainly three different techniques to perform market segmentation:

•             Clustering: this approach implies data grouping or partitioning

•             Association: this approach seeks to establish associative relationships between different variables in the database

•             Visualization: this approach consists of providing the user with an immersive virtual reality environment so that the user can move through this environment discovering hidden relationships

Evaluation, Interpretation & Knowledge Discovery

To test how well the identified segments perform when predicting preferences for new customers, two approaches can be considered: train and test error estimation, and cross validation.

After the prediction accuracy is verified by one of the above methods, the segments will be evaluated by the business people in order to determine the usefulness of the segments. The evaluation of usefulness of the market segments should be made by the business team with respect to the following characteristics:

• Substantiality: (segment size): The market segments are large or profitable enough to serve.
• Measurability: (segment profile): The market segments can be identified and measured in terms of data already available. The segment identification is very important: Segments that are based on meaningful differences in customer needs but lack clear segment identification will fail because the segment identity will not be known and an actionable marketing strategy cannot be developed.
•  Actionability: Effective programs can be designed for attracting and serving the segments. The market attractiveness depends on market opportunity, competitive environment and market access.

If a segment fits the company’s objectives, the company must decide whether it possesses the skills and resources needed to succeed in that segment. If the company lacks the strengths needed to compete successfully in a segment and cannot readily obtain them, it should not enter that segment. Even if the company possesses the required strengths, it needs to employ skills and resources superior to those of the competition in order to really win a market segment. Once the company has decided what segments to enter, it must decide on its market positioning strategy - on which positions to occupy in its chosen segments[D. Raicu, DePaul University).

Conclusion

A theoretical, qualitative data mining framework for automatic gathering of relevant and unbiased data was proposed. As a result, the initial investment of producing a new product vehicle without being certain that it will be satisfying people’s needs will be eliminated. Discovering a-priori segments of people being interested in a certain product will also help the managers focus their advertising, promotion, and sales efforts on those categories of people and thus, the time and costs will be significantly reduced.

Read more »

Intro to Data Mining for Marketers - Part 1

Data mining can be defined as the process of "discovering patterns, meaning and insights in large datasets by using statistical and computational methods". Data mining works to analyze data stored in data warehouses that are used to store that data that is being analyzed. That particular data may come from all parts of business, from the production to the management. Managers also use data mining to decide upon marketing strategies for their product. They can use data to compare and contrast among competitors. Data mining interprets its data into real time analysis that can be used to increase sales, promote new product, or delete product that is not value-added to the company.

History

Data mining was born in the fields of Statistics and Computer Science (some might say Artificial Intelligence) and may also be referred as “Statistical Learning”. From a statistical perspective, most early and recent advances coming from Statistics have come from the Stanford Statistics department school of thoughts like  Bradley Efron, Jerome H. Friedman, Trevor Hastie and Robert Tibshirani. By the way, don’t forget that Stanford University is only 7 miles away from Google.

Stanford University ©

Data Mining Framework

Using data mining techniques, we, marketers, need to master an approach that will provide the decision makers with  a-priory knowledge about customers’ preferences and needs. Since there are many different kinds of customers with different kinds of needs and preferences, a simple, solid approach is meant to be a tool for performing market segmentation: divide the total market, choose the best segments, and design strategies for profitability serving the chosen segments better than the company’s competitors do. The example developed below is described for product development in auto industry, but it can be successfully implemented for any other applications where it is necessary to  find the correlations between the customer feelings or perceptions and the physical characteristics of a product. Yes, correlations, even through our statistics lenses. 

Yes,arithmophobia is over, my friend!

Understanding

Any data mining application should start by understanding the business goals of the application since the blind application of data mining techniques without  the requisite domain knowledge often leads to the discovery of irrelevant or meaningless patterns. In order to understand the target customers of an automotive company, it would be helpful to examine the relationships between the vehicle image/attributes and the customer emotional benefits that are tied to psychological needs, personality traits, and personal values. Thus, data mining can enable us to understand more completely how product specific characteristics relate to customer needs and the benefits a customer hopes to obtain from them. For instance, for many people, cars, homes, restaurants and vacations provide emotional benefits as well as rational benefits. However, for a wealthy person who has everything, the emotional benefits provided by status, prestige and superiority of an expensive automobile could outweigh rational benefits such as gas economy, lower maintenance and insurance costs, and resale value.  

A target audience perhaps? "Free to do anything, in control, confident, sporty but with family."

Therefore, it will be beneficial to have a tool that will help us to respond to questions such as: What and how many of the personality attributes used to describe the customer might be shaped by the vehicle’s image?  What kind of vehicle this customer or group of customers will buy?

Data selection

This step calls for targeting a database or selecting a subset of fields to be used for the data mining. The following issues should be considered in developing a plan for collecting data efficiently:

• Evaluation of existing data sources 
• Specification of research approaches 
• Data gathering (contact methods, sampling plans and instruments)

The survey research is a simple, efficient method to collect data. One of the advantages of the survey research is flexibility because it can be used to obtain many different kinds of information in many different situations. Furthermore, depending on the survey design, it may also provide information quicker at a lower cost compared to manual processing. The survey may be in the form of a questionnaire that is very flexible as there are many ways to ask questions. In preparing the questionnaire, only the questions contributing to the research objectives will be asked. The questions may be closed-ended, as they include all possible answers. In designing the survey, we also make sure that the questions are simple, direct and arranged in a logical order.  The first question should create interest if possible, and difficult or personal questions should be asked last so that respondents do not become defensive.

Instead of a traditional mail questionnaire, a more modern approach is the computer interviewing process, in which respondents sit down at a computer, read questions from a screen, and type their own answers into the computer at their own leisure. The beauty of this approach consists of its multiple benefits. As a first benefit, the respondents’ answers are automatically stored in a database. Furthermore, the survey is posted on the web and it can be accessible by an unlimited number of people. Filling out the survey becomes a non-time consuming task even for a busy person: the survey is on the web and it is accessible for anybody at any time; the submission of the completed survey requires only a ‘click on’ action executed by respondent, action possible  through an interactive survey implementation. 

Third, the computers might be located at different locations such as auto shows, dealerships, or retail locations. The biggest benefit is the collection of more relevant data since people present at those locations are most likely willing to answer correctly to the questions because they are interested in automobiles. The approach can be implemented such that the data is gathered from numerous computers  at different locations and stored in a unique and global database. As a fourth benefit, same survey format will be accessible to different categories of people: expert people (such as car designers) or people less familiar with auto domain characteristics. The large number of respondents and their diversity give more reliability on the results than small samples.


                                                      ...To be continued...

Read more »