Session batching yields consistent breaks between early sessions

We’ve not analyzed the impact of batching extensively. In fact, it was basically broken from mid-August through mid-November without us really noticing! But since we suspect churn is more about # of sessions than # of calendar days, improvements to batching are a big lever for increasing how many people reach high levels of retention.

Now that it’s been a month since I’ve fixed that issue, it’s worth examining what’s happening with batching in practice. In aggregate, it seems to be working pretty well. More than half of users got a >= 5-day break between sessions 3 and 4, and between sessions 4 and 5. About half of users got a >= 3-day break between sessions 2 and 3.

Of course, those are aggregate numbers. They don’t let us see all the silly things that are happening to individual users. But it’s encouraging that batching’s having such a meaningful impact so early in the experience.

Following up, I wonder: what’s the distribution of these session sizes? Are sessions 4 and 5 mostly small-ish? If so, should we batch even more?

See new BigQuery view, logs.batchingStats, and new visualization on dashboard.