In my last article, We Bootstrap to $1M in Two Years Yet We're Starting All Over Again, I mentioned the churn problems we've been having for our product Smiley.
For the last 5 months, we've been focusing a lot on customer success. We use health scores to predict which customers are more likely to churn, so we can proactively engage with the customers.
To calculate the health score, we give each customer's event a weight that reflects the impact it has on the likelihood churn, then multiply the weight by the event count, and sum them up.
The problem here is that we need to determine the weights by gut. 😮
We realized it is hard to determine which events contribute the most for a customer to churn, especially we are tracking a lot of them. So we applied machine learning to help us set the weights.
This way, not only we get more accurate health scores, but also we know which events/attributes are sticky. For example, by looking at the weights table below, we find that customers who care about feedbacks are most likely to stay.
This also helps us better qualify customers.
We're planning to release this as a tool to help SaaS founders get more accurate health scores.
Subscribe on ProductHunt to get early access if you are interested :)