Seminars

Abstract:
Funding pressures have forced public radio, as well as other not-for-profit organizations, to reduce their reliance on mass-marketing efforts, e.g., pledge drives, and increase the volume and sophistication of their direct marketing activities. In contrast to pledge drives, direct-marketing is viewed as a less disruptive (to programming), lower cost, and potentially higher margin alternative. In this talk, we present a model-based probabilistic framework to segment the donor population of a public radio station in the Midwest of the United States. The managerial aim of this work is to understand the distribution of member contributions, identify behavioral drivers that influence contributions (eg., suggested membership levels), and capture the behavior of members (eg., how contributions evolve over time) in different segments. Collectively, we believe, these behavioral insights can allow the station to develop more successful direct marketing strategies.

We derive and implement two instances of the EM Algorithm to cluster members based on annual contributions, and based on dynamic giving-behavior over time. In the latter segmentation problem, members are assumed to belong to classes with distinct giving characteristics, each with its own Markovian behavior. The EM Algorithm is used to obtain parameter estimates of the generative model, i.e., to update the population’s mixture proportion, improve the model’s characterization of the cluster/behavior types (eg., transition probabilities), and cluster each member into a behavior type via a Bayesian scheme based on the posterior probabilities of cluster membership. The clustering of members based on annual contributions is formulated as a finite mixture model with unknown mixture proportions. Through an extensive empirical study, we show that framework is able to capture unobserved, but systematic differences between individuals, a key limitation of classical ``tiling'' approaches to segmentation. We also show that failure to capture heterogeneity in class behavior can lead to the improper characterization giving behavior, which, in turn, can lead to ineffective targeting of members. The two clustering approaches are also shown to be synergistic, and provide a means for the station to exploit segment-level behavior, such as loyalty, persuadability, interests, etc. in the devising its direct marketing strategy.

Speaker Biography
Elizabeth J. Durango-Cohen is an assistant professor of Operations Management at the Stuart School of Business at the Illinois Institute of Technology. She completed her Ph.D. in 2002 in Industrial Engineering and Operations Research at the University of California, Berkeley where she worked in the area of production planning and inventory control in the context of supply chain management. Dr. Durango-Cohen’s current research efforts focus on the interface of Marketing and Operations. In addition to her work on direct-marketing optimization for non-profit broadcasting institutions, Dr. Durango-Cohen is also interested in modeling the effect of capacity on pricing decisions for supply chains with competing National and Store-Brand products.

For more information, contact Prof. Homem-de-Mello, thmello@uic.edu

• Fall 2009 Seminar Series