Case Study: PayPal achieves reduced customer churn and rapid predictive modeling with H2O.ai

Solving Customer Churn with Machine Learning

PayPal, a global payments platform with over 173 million active accounts, faced a critical business challenge: accurately identifying if and when customers would churn so marketing and product teams could act quickly. Traditional time-window churn reports mis-timed churn dates and lacked per-customer insight, so PayPal’s data scientists turned to H2O.ai, using the H2O machine learning platform (random forests and GBMs) integrated with R and Python and later deployed on Hadoop.

Using H2O.ai, PayPal built a production “predictive modeling factory” that ran large-scale, rapid modeling and scoring across the entire customer base. Model training times fell from hours (≈6 hrs) to about 10 minutes and scoring from ≈72 hrs to about 5 minutes, enabling targeted, timely win-back campaigns, improved churn metrics and accuracy, and a new program built around H2O.ai’s machine learning outputs.

PayPal

Julian Bharadwaj

Senior Data Scientist