9.1. Setting the Scene

Cellularplex is a small cell phone provider that is poised and anxious to expand its customer base. Jeremy Halls, the director of marketing, has been hearing about predictive analytics and how it can be successfully used to focus marketing strategies. He believes that statistical and data-mining models of customer characteristics and proclivities could greatly enhance his ability to run marketing campaigns that are targeted and well-timed, reaching potential customers with the right offers and using the right marketing channels.

Mi-Ling Wu has recently been hired by Cellularplex. Her previous position was with a medical research firm, where she conducted statistical analysis relating to clinical trials. Prior to that position, Mi-Ling worked with a retail firm doing predictive analytics in the marketing area.

Mi-Ling is a key member of a team that Jeremy forms to design a test campaign aimed at identifying new customers for a specific offer that Cellularplex will market in the coming year. The goal of the test campaign is to identify demographic characteristics of individuals who would likely purchase the offer and to determine the best delivery method for various combinations of demographic characteristics. The knowledge gained will be employed in designing a subsequent large-scale campaign to attract new customers.

The team members brainstorm a large number of characteristics that they think are indicative of people who might respond positively. Then they work with a data vendor to obtain a list of people with these characteristics. They also obtain the contact details for a small random sample of people who do not have the characteristics that they have identified, realizing that information from a group outside their chosen demographics could yield information on customers who might otherwise be overlooked. The team also determines different delivery methods for the offer, and with Mi-Ling's help they include these in their design. Finally they agree on how to measure customer response—for each contact, they will record whether the result was a customer inquiry, purchase of the offer, purchase of a different offer, or rejection.

The duration of the test campaign is set at three months, starting in the second quarter. Mi-Ling will support the effort, but in the meantime she wants to devote some thought to how she will analyze the kind of data it will generate. She will need to use the numerous measured characteristics and the response of each individual to classify each into one of the four possible categories, and thus determine those likely to become customers and those unlikely to become customers. Mi-Ling starts learning about JMP, which is used by the engineers at Cellularplex, and soon finds out that it has powerful visualization and modeling capabilities, which she hopes to put to good use when the real data arrive.

Knowing that she will have to undertake this analysis quickly once the data become available, Mi-Ling looks for a published data set that she can use for practice, both to learn how to use JMP and to see how it performs relative to other software that she has used. Specifically, she would like a data set with a large number of descriptive characteristics where classification of subjects into two or more categories is of primary interest.

Given her previous medical background, she easily finds and downloads an appropriate data set—the Wisconsin Breast Cancer Diagnostic Data Set. Her plan is to use various techniques in JMP to fit classification models to this data set. Realizing that some of the Cellularplex engineers are experienced JMP users, Mi-Ling connects with a few of her associates to ask if they would be willing to help her if necessary. James, an experienced JMP user, spends a couple of hours with Mi-Ling, giving her an introduction and offering to help her with further questions as they arise. What James shows Mi-Ling impresses her and gives her a good starting point for learning more on her own.

In this case study, you will join Mi-Ling as she works with some of the JMP capabilities for exploring and modeling high-dimensional data. She begins by using visualization techniques to help build an understanding of the Wisconsin Breast Cancer Diagnostic Data Set. After dividing the data into a training set, a validation set, and a test set, she applies three different modeling approaches to the training data—she fits a logistic model, a partition model, and two neural net models. Then she compares the performance of these four models on the validation set, chooses one of these as her final model, and uses the test set to assess its predictive performance.

This case study uses the principles of Visual Six Sigma to construct knowledge. This type of knowledge can eventually be used to guide sound business decisions. By its nature, this case study continues through the Model Relationships step of the Visual Six Sigma Data Analysis Process, but it does not involve the Revise Knowledge and Utilize Knowledge activities. These activities will become relevant once the classification scheme that Mi-Ling eventually develops for the marketing data is implemented as part of the formal marketing campaign.