Reducing churn for a subscription business

Problem :

• 30% annual churn (after 1-year subscription).

• Losses stemming from discounts given to them.

Data Science Challenges :

• There were multiple tables. Joining tables required a good understanding of data, and we had to pull data from lots of sources.

• Lots of duplicates in the database and it wasn't straightforward to remove them.

Approach/Solution :

1. We identified 5,200 customers at immediate risk of churn (next 3 months). These customers had 90%+ probability of churning. We came up with personalized promotions for them.

2. Explained signature behaviours before attrition. Some behaviours could be completely changed or avoided, reducing churn without doing anything; for example, it was noted that long waiting times towards the end of the subscription, resulted in churn. It can be avoided by prioritizing calls of people at the end of their subscriptions.

3. Notify the marketing team every time somebody is about to churn. We integrated our modelling results in their CRM, so that concerned team gets automatic notifications every time this event is about to happen.

4. Design different promotions for specific segments. Different people respond to different promotions. Some like 20% off, while some like few extra books. We segmented all customers by the kind of promotions they respond to.

5. Gave a list of 2,100 customers who had churned in the last 365 days but had high customer future value left. These people have $50+ future value left.

Results :

1. Out of 5,200 customers about to churn, 3,422 did not, $38,000 potential loss saved.

2. 1,123 churned customers re-subscribed, accounting for $12000 surplus.

3. $7,400 saved in losses from personalized promotions.