Modeling Customer Memory to Improve Prediction of Customer Purchasing Behavior
Statistics Department; Faculty Advisers: Peter Fader, Shane Jensen
Models for consumer purchasing have become increasingly useful in industry as a way to predict future demand, identify high value customers, and more. What three of the most well known and well used models do not account for is the intuitively appealing fact that customer preferences may change over time. For example, a customer may like but then grow out of a particular clothing style, causing purchases to rise and then fall. This customer’s behavior would not be consistent with models that assume preferences do not change.
Our research goal is to build into these models the possibility that customer preferences to change in a statistically principled, data-driven way, and to create metrics and tools to help businesses operationalize this new information. We do not want to propose entirely new models – rather, we want to take the existing models and insulate them against changing consumer preferences.
We believe this research has three managerial implications. One, we can make so-called “big data” smaller by shrinking the amount of data that companies need to keep on hand, to the extent that this information is no longer relevant. Two, we can better estimate the lifetime value of customers, because we have a more current estimate of customer preferences. Three, we can alert companies more quickly to changing preferences in their user bases, making it possible for companies to more quickly respond to those changes.
Publications, Presentations & Awards
McCarthy, Daniel and Peter Fader (2017), “Valuing Non-Contractual Firms Using Common Customer Metrics,” Working Paper, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2923466
“V(CLV): Examining Variance in Models of Customer Behavior”, Marketing Strategy Meets Wall Street Conference, Singapore (January 2015)
“V(CLV): Examining Variance in Models of Customer Behavior”, Marketing Science Conference, Baltimore (June 2015)