Affinity and wealth score prediction using multi-task learning in donor analytics

Understanding and predicting donor behavior is crucial for optimizing fundraising strategies in nonprofit organizations. This paper introduces a multi-task learning (MTL) framework that simultaneously predicts donor affinity (likelihood to donate) and wealth score (capacity to donate), leveraging shared representations across tasks. Using a combination of internal CRM datacomprising donor demographics, engagement history, and giving patterns, and external socioeconomic enrichment data, the model is trained to capture both behavioral and financial indicators. Compared to traditional single-task approaches, our MTL model demonstrates improved predictive accuracy, with a 7.3% increase in AUC for affinity prediction and a 12.5% reduction in RMSE for wealth estimation. These results indicate that jointly modeling related tasks not only improves efficiency but also enhances decision-making capabilities. The proposed system enables nonprofits to better segment donors, prioritize outreach, and personalize campaigns, ultimately increasing engagement and fundraising yield.