2 Feb 2026 · Journal
Session length is a lagging metric
Average session length arrives after the user has already succeeded or failed. It also mixes people who are reading a long article with people who cannot find the log-out button. Treating the average as a health metric is how Lifecycle Analytics for Mobile Users turns into interior decoration.
In Activation Windows we pair two measures instead. First, time-to-core-action inside the session, capped so a backgrounded app cannot inflate it. Second, a binary: did the core action happen at all. Length is demoted to an annotation, not a window definition.
There is a cultural problem here. Executives like minutes because minutes sound like attention. Faculty like completions because completions can be owned. If your product is a timer, minutes might be the work. If your product is a marketplace, minutes in search can be a tax.
When students insist on keeping session length on the exec slide, we ask them to split sessions that contain the core action from sessions that do not. The two distributions are rarely the same. Publishing only the blend is how a confusing navigation change looks like “more engagement”.
Lagging does not mean useless. Look at length after you already know who activated. Use it to hunt for stuck states. Do not let it define the cohort. That rule has survived every term since 2018, including teams who arrived with a KPI pack printed in colour.