Statistical Modeling

Mine More Value From Your Survey Programs with Statistical Modeling

Businesses that store demographic or transactional data on their customers can significantly enhance the value of their customer survey results using statistical modeling, says Ari Kapur, Ph.D., of the CustomerSat Professional Services team.

In this article, we draw on Ari's expertise to provide a brief, non-technical introduction to statistical modeling and how it can help your organization leverage survey results.

How can businesses benefit from statistical modeling?

Any company that stores demographic, transactional (behavioral), or financial data on their customers can use statistical modeling to benefit such areas as:

·         Direct Marketing:  Targeted messaging can be developed given the category into which a customer falls (loyal vs. not loyal, highly profitable vs. unprofitable, etc.)

·         Customer Service:  Based on statistical modeling, identify and contact those customers in your database who are potentially dissatisfied (but have not necessarily taken a survey) and properly address their issues/concerns.

·         Retention:  Identify those customers in your database who are statistically most at risk (based on behavior) and serve them accordingly depending upon their contribution to overall revenue/profitability.

·         Customer Acquisition:  Based on profiling your most loyal customers (using survey results, behavioral data, or a combination), is there potential for finding, prospecting, and acquiring “look-alikes”?

What are some sample applications?

A telecom company Ari worked with was preparing to roll out a new service. The company wanted to know which service areas would prove to be the most profitable. They surveyed a sample of customers nationwide asking a number of questions including whether they would be interested in purchasing the new service. The company was then able to model intent to purchase among customers within this sample against demographic and behavioral variables (such as previous purchase habits) already in their database.

This model for predicting purchase intentions was then applied to customer data from different service areas. Service areas were scored to identify those with the greatest numbers of customers likely to purchase the new service.  Results were used to roll out the campaign by service area in the way that yielded the greatest profits for the least marketing expense. 

In another example, a financial services firm examined characteristics of customers who were most likely to fall into each of the major customer loyalty categories used by CustomerSat. The project manager learned that small, low-tech institutions were more likely to fall in the most disaffected group (“Terrorists” or “Subversives”) – those who will tell their horror stories about your company to 15 or 20 of their closest friends.  This information raised important questions about account management and prospecting to institutions of this type.

How does statistical modeling work?

Statistical modeling is a mathematical tool for predicting business outcomes by identifying characteristics of customers whose outcomes are known and extrapolating to those whose outcomes are not yet known.

As the outcome variable, statistical models can use an actual behavior, an interest, or an intention:

·         How much they spent

·         How much they increased spending

·         Whether or not they upgraded within N months

·         Whether or not they defected

·         Whether or not they would recommend your products

·         Whether or not they were interested in a new product or service.

The model is an equation that indicates the probability of a desired outcome (dependent variable) based on customer characteristics (independent variables). When the customer characteristics are available but the outcome is unknown, the equation can be used to categorize customers according to their potential for achieving the desired outcome.

Statistical modeling used to identify drivers of revenue is often referred to as “revenue modeling.”  The right revenue model tells us in which areas and for which customers we can apply resources to most positively impact revenue. The model can also be used in marketing campaigns to target prospects most likely to become profitable customers.

What steps are involved in statistical modeling?

We begin by selecting an outcome variable from the survey such as repurchase intent or willingness to recommend.  We classify respondents into two groups, for example, those who Definitely or Probably will Recommend and those who Definitely or Probably will Not Recommend. CustomerSat clients who use the “Apostle Model” could classify respondents into “Apostles” and “All Others.” (More groups can be used if the sample size and characteristics of the data permit.)

We next identify customer variables (company size, location, type of business, annual revenue, etc.) we believe indicate whether or not a particular customer will fall into the more desirable category.

Using logistic regression, we generate an equation that gives us the probability of membership in the desired category as a function of the customer variables we selected. 

We validate the resulting model by “scoring” a different set of respondents than we used to generate the equation.  Scoring gives us the probability that a given respondent will fall into the target category.  We then compare the “predicted” category for each respondent with their actual categorization. When our model is sensitive enough to give us a satisfactory level of accuracy in making our predictions, we are ready to apply it to a new dataset for which the outcomes are unknown.

Note that modeling does not depend on survey responses, only on the characteristics of the individuals or firms we wish to classify. This means we can apply the model to existing customers who have not participated in the survey.  It also means you can build prospect lists based on very specific knowledge of the characteristics of your ideal target market.

In other words, you can increase the business intelligence you gain from both your survey results and your existing customer data.

Ari Kapur has applied his statistical expertise to business modeling problems, including Customer/prospect segmentation and profiling, since 1996. He earned his Ph.D. in Economics with specialization in Marketing and Consumer Demand Analysis from Texas A&M University.

CustomerSat Professional Services provides a variety of Statistical Modeling Services.  For more information, contact your CustomerSat project manager or Account Executive, or send email to info@customersat.com.