Virtual Quantitative Marketing Seminar

Upcoming Talks

• Monday, April 6, Noon ET - Eric Schwartz (Michigan)

What Public Utilities Can Teach Us About Customer Targeting Methods: When Bandits, Active Learning, and Tree Ensembles Combine

Abstract: Customer targeting in marketing typically asks, "Which customers should receive some intervention?" and the best marketers continuously learn how improve their targeting decisions over time. We show this general sequential decision-making problem governs infrastructure remediation at public utilities, a massive, data-rich domain, with high customer spend, and high stakes, yet with virtually no business school research attention. We introduce Random Forest Thompson Sampling to marketing, which extracts asymptotically valid posteriors from cross-tree variance to enable principled explore–exploit balancing in sequential customer targeting. We then show that this and other common sequential targeting algorithms are parameterizations of a single Selection Function Spectrum, unifying a wide range of disparate methods. We show validation evidence these methods (and approximations) through lead pipe replacement in Flint, MI, where a federal court mandated adoption of our model (80% hit rate with our approach versus 15% without it). Plus, we show cross-site validation with a New Jersey utility and report on implementations in dozens of cities nationwide. But this talk is not about Flint or lead pipes per se; this infrastructure work, which I began 10 years ago, inspired methodological innovations with broad applicability across management sciences. They apply whenever manager must sequentially allocate scarce resources under uncertainty, from churn prevention and B2B sales qualification to product recalls, but has the ability to learn over time to improve those decisions.

• Monday, April 13, Noon ET - Ayelet Israeli (HBS)

• Monday, May 4, Noon ET - Xu Zhang (LBS)

• Monday May 11, Noon ET - Dante Donati (Columbia)