Overview of this campaign
There is significant revenue potential in retaining customers. Not only because of direct loss of revenue from churning customers, but also because we would like to cover costs already invested in acquiring our customers. But how can we know exactly which customers will churn so we can effectively communicate with these customers? We wanted data from different sources to tell us, and then use data analytics to build a churn prediction model.
In January 2015, we started by recruiting a churn analyst. His first task was to design and build a churn prediction model – specially customized to take into account the relevant data and parameters for our print and digital subscriptions respectively.
The objectives we would like to achieve were the following:
- Predict in advance, and with high precision, which customers are going to churn the coming period.
- Identify which customers to target with retention measures.
- Identify what specific actions that would have the greatest retention impact on each customer, or on defined customer segments.
The output of the churn prediction model would be a “churn score” assigned to each customer, as well as churn segments with corresponding retention activities based on each customer’s respective churn drivers (the most relevant parameters identified to score the customers at risk of churning). Finally, we could then provide the most efficient retention activities to each customer, and reduce actual churn.
Results for this campaign
By December 2015, we are proud to say that we are able to predict churn by close to 90 % accuracy on our print subscribers, and about 80 % accuracy on our digital subscribers. Through 2015, we have experienced a reduction in monthly churn rates by 5 % across all subscription categories – which obviously contributes significantly to our subscription revenues.
After 4-5 months with data experimentation and evaluation of different data models, we ended up with a churn prediction model based on a series of prescriptive and predictive statistical analysis – fed by data from several different data sources:
- Descriptive analysis: In order to describe the effect on each parameter (or group of parameters) separately.
- Survival analysis: To understand where in the lifecycle our customers tend to churn on a general basis.
- Logistic regressions: To test the parameters, and their relative relevance and strength.
- Factor and cluster analysis: To segment customers based on their respective churn drivers.
Based on the predictions, each customer is assigned with a churn score (the likeliness to churn the next period). The customers are then split into four segments (secure, relatively secure, uncertain, likely to churn), based on their churn scores. Finally, proper retention activities are provided based on the customers respective churn drivers.
We now run churn predictions on our print and digital subscription base on a monthly basis. The model is continuously tested and modified, in order to improve churn predictability. The model is highly scalable and flexible and we will be able to refine the model – both based on actual churn and new data sources that become available.
Next steps will be to feed the model with even more data, especially data generated through digital customer behavior. Furthermore, we will look into how we can move in the direction of automated 1:1 customer retention measures, to see even higher timeliness and impact on retention retention.