This research evaluates the performance and business applicability of Marketing SmartAudience360, a high-dimensional lookalike modeling framework engineered to identify client-like households for Savant Wealth Management. Leveraging Savant’s internal CRM ecosystem and third-party enrichment from the Adstra national consumer database, we developed and compared multiple supervised machine learning models, including Logistic Regression, XGBoost, Stochastic Gradient Boosting, and Feedforward Neural Networks, within a Positive–Unlabeled (PU) learning context. The models predict the degree to which external households resemble Savant’s validated seed audience of 39,859 client accounts, classifying them into ranked “Lookalike Bands.” Findings show that tree-based ensemble methods, particularly stochastic gradient boosting, offer superior ranking stability and operational precision. The resulting scoring architecture provides Savant with a scalable and repeatable system to prioritize high-value prospects, reduce acquisition costs, and enhance marketing effectiveness.
Watch the team present this project at 06:10 in the session recording here.
Keywords: Lookalike modeling, XGBoost, Adstra, PU Learning, Wealth Management, Predictive Marketing
