So you’ve created a churn prediction model, and you know which customers are likely to leave you. Now it’s time to accept the facts: some churners are going to churn no matter what you do. What next?
It’s easy to get over excited about churn prediction and start immediately empowering your marketing team to enact an all-encompassing prevention plan. But when thinking about building an end-to-end churn strategy, perhaps the most important component is the output, or what you’ll do with the data - it deserves as much attention as the beginning stages, like choosing initial data sets to use or fitting and choosing a model.
Enter: Uplift Modeling
Since you know marketing efforts will not change the mind of every potential churner, the final step of churn prediction is uplift modeling - a secondary prediction after your initial one. Basically, finding of your potential churners which ones are likely to respond positively to marketing messages so that in the end, you don’t waste time or money targeting the wrong people (or making matters worse in the case of the “Do Not Disturbs”).
It’s easy to get excited about churn prediction. Instead of immediately reaching out to potential churners, consider a more strategic approach.
Uplift modeling is very nuanced and isn’t specific to churn (read more detail on the subject), but in summary, it looks something like this, looking at the increase in likelihood of churn with marketing intervention as compared to the outcome without:
Predictive vs. Prescriptive
This concept points to a much larger point when thinking about predictive analytics: they don’t provide value unless they are actionable. “This great objective of data science, to intelligently drive day-to-day business decisions based on data, is the purview of uplift modeling,” says Data Scientist Mike Thurber. This also ties in with the idea of deployment into production, something that we emphasize strongly and touch on often in our blog.