This project includes an outreach‐optimization engine that maximizes CMS Star Ratings improvements with cost and capacity constraints. First, we predict each member’s response likelihood to calls, emails, and assessments using machine learning (XGBoost, random forest, MLP). These predictions form a conversion probability table. Alongside, we estimate outreach costs, capacity, and expected Stars gain per action. A mixed-integer optimization model then identifies the most cost-effective strategy mix. The project outputs (1) optimal outreach recommendations, (2) Stars impact summaries, and (3) an interactive dashboard for scenario analysis and member-level insights—all aimed at maximizing impact within resource limits.
Watch the team present this project at 52:21 in the session recording here.
Keywords: Conversion probability, optimization, resource allocation, engagement analytics, machine learning, outreach strategy
Faculty Advisor
Jonathan Williams has been working in statistical consulting and data science education for fourteen years and currently teaches full-time at the University of Chicago. Previously, he managed data science teams at Civis Analytics, working on behalf of public sector clients, and before that he worked as a vice president at Compass Lexecon, providing and supporting expert reports and expert testimony for litigation. Jonathan earned his BA and MS degrees in statistics from the University of Chicago (’07, ’08) and is also an alumni of the Master of Science in Analytics (now known as MS in Applied Data Science) program (’16). His focuses include regression analysis, data visualization, technical writing, financial valuation, mortgage portfolio modeling, and damages estimation.