Pooled Wisdom, Unique Insights: Improving Customer Predictions with Cooperative Databases and Deep Transfer Learning
(A2025-126418)

AUTHORS

Sumon Chaudhuri, WU Vienna; Arnaud De Bruyn, ESSEC Business School

ABSTRACT

This paper explores the effectiveness of cooperative (co-op) databases in improving customer behavior prediction through transfer learning. Co-op databases are a recent concept that enables firms to pool anonymized data managed by third parties, enhancing insights without compromising data privacy. We study fourteen charitable organizations, utilizing transfer learning to create individual predictive models from shared data while ensuring confidentiality. Our findings reveal that pooling data can improve prediction accuracy and is particularly beneficial for firms with limited data or experiencing strategic shifts or market disruptions (i.e., models calibrated on pooled data are more resilient to covariate shifts and external shocks). This research underscores the importance of co-op databases in predictive analytics. It demonstrates how machine learning can be applied to maximize the benefits of data sharing across diverse operational contexts.