Predicting Churn
91%
accuracy in predictive churn model
£4m+
additional revenue from inactive customers in the first month
£13m
of incremental revenue was generated in the first month
The Ask
A multinational trade retailer was concerned about customer churn. It was unclear to the business which customers had truly lapsed and which were following standard lifecycles, meaning identifying customers to target with ‘win-back’ campaigns was particularly challenging.
Naturally, not all customers followed the same purchasing cycles, some bought every three days and some every twenty. A generic approach would not cater those who had lapsed and those who may return of their own volition.
Our Solution
We needed an individual prediction for each customer to understand where in their expected purchase window they stood.
Our experts created a machine learning model to analyse natural rates of return for each customer at an individual level, clearly illustrating where they were in their lifecycle. We labelled customers; ‘neutral’, ‘active’, ‘at risk’ or ’churned’. This allowed us to identify 475 thousand out of two million customers at high risk of churn. These customers were targeted with our ‘win-back’ campaign that included five different sets of incentives.
Impact
The predictive churn model’s accuracy was 91%, meaning more than half of reactivated customers were still active after six months.
In the first month, more than £4m of additional revenue was generated from previously inactive customers and £13m of incremental revenue was generated.
Sound good?