Churn Prediction

Churn prediction is the application of data science techniques to identify customers who are at elevated risk of ending their relationship with a brand—either by canceling a subscription (voluntary churn), allowing a payment to fail (involuntary churn), or simply not making another purchase (passive churn).

Effective churn prediction models analyze multiple signals. For e-commerce, key predictive features include: days since last order (weighted at ~40% importance), order frequency decline over recent periods (~25%), engagement signal changes like email open rates and site visits (~20%), and LTV relative to the customer's cohort (~15%). The model outputs a risk score (Low, Medium, High, Critical) along with the primary risk factor driving the prediction.

The value of churn prediction lies in enabling proactive intervention. Rather than reacting after a customer has already churned—when win-back success rates are typically below 10%—brands can take action while the customer is still active. Interventions range from personalized email outreach and targeted discounts to subscription flexibility options and direct customer success contact for high-value accounts.