Since churn is so important, wouldn’t it be useful if we could predict in advance which customers were most likely to churn?

That way we could put our best customer service reps to work in an effort to save the situation. It turns out that we can do that by instrumenting our SaaS applications and tracking whether our users are engaged with the key sticky features of the product. Different features will deserve different scores. For example if you were Facebook, you might score someone who uploaded a picture as far more engaged (and therefore less likely to churn), than someone who simply logged in and viewed one page.

Similarly if you sold your SaaS product to a 100 person department, and only 10 people were using it, you would score that differently to 90 people using it. So the recommendation is that you create a Customer Engagement Score, based on allocating points for the particular features used. Allocate more points for the features you believe are most sticky. (Later on you can go back and look at the customers who actually churned, and validate that you picked the right features as a predictor of who would churn.) And separately score how many users are engaged with specific scores.