We did this analysis fairly early on at Pinterest, when we were really trying to work on retention. One of the primary questions we needed answered was, "When should we intervene to try and win someone back?" Ideally, you don't want to wait until someone's dormant and been gone for 30 days. Your chances of winning them back at that point are low. But a very active user might still be in danger of churning after a few days of inactivity, whereas a light user might not churn after that same number of days inactive. So I was trying to understand by usage, after how many days of inactivity should I we get worried?

We applied this analysis to build a reengagement focus email. Based on these variables, we sent that email out to users who crossed a certain churn probability threshold.

The horizontal axis is the number of days since the individual was last active, and the vertical axis is the number of days active the user was out of the past 28 days. An individual cell is the percent probability of that user churning out and not being active the following month. (I've obscured the real data, but this gives you a sense of the trends.)

To actually use this heat map, we set a cut line of 20%. We decided that when a user's churn probability hit 20%, that's when we'd send a notification or email to try to re-engage them.

In general, this was effective at boosting retention. But we weren't as sophisticated in our data measurement back then as we were when I left Pinterest. At the time, we just ran the experiment, saw the numbers go up, and said, "Great, it's working!" I think I'd approach this more rigorously today.

We made this second heat map to understand how big the problem actually was for the areas where we were seeing high churn. We didn't want to fall into the trap of focusing all our energy on a super high churn segment if it only made up a small percentage of our total user base.

This heat map strongly suggested that most of the population fell into the 1-2 days active camp.

John Egan

We did this analysis fairly early on at Pinterest, when we were really trying to work on retention. One of the primary questions we needed answered was, "When should we intervene to try and win someone back?"

Churn Probability Analysis from Pinterest

Overall Heatmap:

Population Percentage Heatmap:

Sign up to see more