One of the most crucial skills of a data scientist is not only to be able to build an accurate predictive algorithm but to suggest a set of actions based on that to enhance the goal ratios.

I have build a churn rate predictor for a retail APP which accuracy after testing is about 90%. So now that I know with high accuracy which users will churn out of the APP in the next days, what can I do to try to retain them. Thinking in the overall APP environment the only idea that comes to my mind is:

  • Push notifications. CRM, use a recommender system to send push notifications based on users profile and micro-moments.

1 Answer 1


You could do the following:

  1. Find out the most important predictors used by your model; the methods used to identify importance depend on the model used.
  2. For each of the major pedictors try to find out how does each predictor segment the customer population of will-churn vs. will-not-churn. That would be the starting point for framing hypothesis of why these set of customers are churning. Generate reports/info-graphics showing how these predictors impact the tendency to churn. Sometimes predictors are not causes but side-effects of the hidden cause showing up elsewhere. So be open to skepticism about the predictors.
  3. Involve the business/users and have brainstorming sessions to go through each of these hypothesis and evolve the most practical and economic actions that could be taken to reduce churn. Before you get into a discussion, plan for a sufficient amount of time, come up with your best thought actions which you feel will address the situation and have all the details gleaned from your analysis available for sharing in the session.

This is an art more than a science and your ability to understand the domain as well as good communication will help you achieve this goal.

  • 3
    $\begingroup$ Let's say, totally hypothetical, after much hard work, you find the customers most likely to churn happen to have long hair. What are you going to do? Send a notification, ask them, pls cut your hair short ? Or when user sign up for your app, you ask the long-hair folks, hey, pls don't come...... $\endgroup$
    – horaceT
    Commented Dec 27, 2016 at 17:54
  • $\begingroup$ Those are the kind of questions solved by going through the steps suggested in #2 and #3. $\endgroup$ Commented Dec 27, 2016 at 18:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.