Data Science, Analytics & Solutions
This multi-channel retailer knew that many online customers were not returning and therefore was looking for solutions to proactively identify those customers most likely to churn.
What We Did
In addition to data collected to support the segmentation process, we used features such as seasonal profiles to understand propensity to churn Using machine learning, we ran over 40 different models (varying customer features, historical data, segment and predictive model used) to identify the models that best explain the propensity to churn and a packaged model that can identify the key features driving churn by segment.
Key behavioural traits were identified that could be addressed through targeted product placement, pricing and delivery offers. These included: Page views that were between 10 and 30 were less likely to churn; The optimal price range is between £20 and £60; Customers not returning any items and those with very high return rates were more likely to churn. Analysis of the customer base identified a £9.6m revenue increase through reducing churn.