CUSTOMER CHURN PREDICTION 

USE CASE

THE CHALLENGE

An e-commerce market leader with a little room for new acquisitions had to switch from growth strategy to protective approach of their existing customers. Their standard 'carpet bombing' marketing campaigns had been delivering poor ROI and so they asked us for help.

OUR SOLUTION

A good definition of churn is crucial for building an actionable predictive model. In this case, we linked the definition to the main KPI of the business  (churn = no purchase in 12 months). Machine learning algorithm that utilised basic demographic and transactional data then helped to target right customers at a right time.

RESULTS

We were able to decrease churn by c.30%. Additionally, during the model development we discovered various insights that helped to desing marketing offers (e.g., that relative discount size matters more than absolute one) and supported the company understanding of cusomer churn (customer memory is about six months long - what happened earlier does not matter).

Joint project with BizzTreat and Keboola