Explore Rice

5/15/2006

New model helps marketers identify customers with greatest lifetime value

A firm's marketing decisions, how much they should spend on a customer and what customers they should target depend, in part, on how much individual customers are worth to the firm. Knowing the lifetime value of each customer can help a company get better returns on its marketing investments. There are a number of approaches to estimating customer lifetime value, but a new model developed by researchers at Rice University and Northwestern is proving to be the most valuable.

-----------

To get more bang out of their marketing budgets, firms in recent years have looked for more accurate ways to measure their customers' lifetime value or the total revenues each customer generates for the company over his/her lifetime with the firm. A new statistical model recently developed by university researchers is the best predictor yet for calculating each customer's lifetime value and for targeting a firm's most valuable customers.

"If a firm did nothing more than use its in-house data with this model to identify and target its more valued customers, it could significantly improve the effectiveness of its targeted marketing activities," claimed Siddharth Singh, one of the model's designers and an assistant professor of management at Rice's Jesse H. Jones Graduate School of Management.

According to Singh, the types of firms that can most benefit from the new model are membership-based purchase clubs like music and book clubs, automobile associations such as AAA and membership-based retailers like Sam's Club and Costco.

In each of those settings, companies have data on when their customers have become members, when they have discontinued their association with the firm, when they purchase and how much they spend during each purchase, once these events happen. The future purchase behavior of customers, however, is not known with certainty.  

"Prior research has not focused on these particular types of firms to develop sophisticated models that could measure their customers' lifetime value," said Singh. "Ours was also the first to develop a model both to estimate customers' lifetime value in this context and use it as a means of targeting a company's most valuable customers."

In the past, many companies have targeted customers based on such measures as how recent the latest purchase was, frequency of purchase incidents or how often a customer purchases, and how much a customer spends on average. With a limited marketing budget, a company would want to focus its promotional dollars on its most valuable customers. For example, it might select a group of customers based on whether they had made purchases recently, purchased the most frequently in the past or spent the most money with the company.

"The company might focus its marketing activities based on one or more of these factors to target certain customers they estimated would provide the most revenues," explained Singh.

In their research entitled "Customer Lifetime Value Measurement," Singh, Rice University colleague Sharad Borle, as assistant professor of management at the Jones School, and Dipak C. Jain, dean of the J.L. Kellogg School of Management at Northwestern University, compare a model they recently developed to several other existing models in predicting customer lifetime value and in targeting valuable customers.  

The authors used two random samples drawn from data on customers who had joined a membership-based direct-marketing company in a specific year in the late 1990s. The information included each customer's membership initiation and termination date with the company and their purchasing behavior over their entire membership duration with the company.

The first sample contained 1,000 past customers and consisted of 7,108 purchase occasions. The second part, which contained 500 past customers, was used for predictive testing and to illustrate their model's application.

"Since customer purchase behavior might change over time, the key drivers of a costumer's lifetime value also might change over the customer's membership with the firm," explained Singh.

The researchers used a hierarchical Bayes approach to modeling a customer's lifetime value with the firm and explicitly account for a customer's expected spending pattern over time. After estimating the parameters for the comparison models, they applied each model to the second data sample and compared their predictions to the customer's actual behavior. Using the same data set, each model was also used to identify the top 50 percent of the company's highest-value customers.  

"Of all the models compared, ours provided the most accurate prediction of customer lifetime value and the closest match to the actual top 50 percent of the company's most valuable customers," said Singh.

A member of the Jones School since July 2003, Singh has also conducted research on customer loyalty programs, profitability and product returns and has collaborated on studies involving the impact of survey participation on short- and long-term customer behavior.

Singh received his Ph.D. degree from Northwestern University, his master's in business administration from the University of Illinois and an undergraduate degree in technology from Banaras Hindu University.

For more information, contact Singh at sssingh@rice.edu or Debra Thomas in the Jones School at dthomas@rice.edu.