Impact Solution Challenge 1 • Pre-paid churn rates higher than post-paid • Predict churn-likely customers before next bill - Target them with remediation or marketing • 32 million prepaid transactions daily - 40 different channels - 100+ transaction types • Gained actionable insights from data • Identified leading indicators of churn for: - Pre-paid customers - Post-paid customers • Were able to proactively identify 30% of churn • Able to challenge fundamental assumptions - What was causing the churn? - How addressable is the problem? • Capture all real-time data and aggregate - Complex pre-processing process • In-depth feature engineering process: - 200 attributes collected; 80 generated • Compared 7 models – GBT selected: - Accuracy - Model confidence VERIZON WIRELESS Understand Caus

Join for free to read