This project focuses on identifying "silent churn" within a B2B wholesale giftware business. Because wholesale clients don't formally "cancel subscriptions," this project utilizes an intelligent Machine Learning approach to track purchasing behaviors, flag irregular account inactivity, and predict which high-value accounts are at risk of leaving.
The business relies heavily on a small group of high-revenue clients (the top 100 customers generate ~36.8% of all revenue). When these key accounts quietly stop purchasing without warning, the revenue impact is massive. A proactive strategy was needed to protect these vital relationships before they were permanently lost.
- The 90-Day Churn Threshold: Statistical analysis revealed that 90% of returning customers place a subsequent order within 115 days. A strict 90-day inactivity window was established as the baseline for customer churn.
- Routine Over Revenue: The Random Forest model revealed that Recency and Frequency are vastly stronger indicators of loyalty than total lifetime spend.
- Model Performance: The model successfully achieved an 81% Recall rate for the churned class, meaning it accurately detects 8 out of 10 deflecting customers, giving the marketing team a crucial window to intervene.
Based on the data, the following business actions were recommended:
- Launch a 60-Day Intervention: Automate retention emails/discounts specifically triggered at the 60-day inactivity mark.
- Reward Frequency: Redesign loyalty programs to incentivize smaller, more frequent orders rather than single, massive purchases.
- VIP Retention Protocol: Assign dedicated account managers to manually monitor the Recency metrics of the top 100 accounts weekly.