What is a session?

The timeout defines the metric

A session has no natural end, so one is imposed: after a period of inactivity, the next action starts a new session. Thirty minutes is the common default, inherited from web analytics.

That threshold determines the numbers entirely. A short timeout produces many short sessions; a long one produces fewer, longer sessions from identical behaviour. Session count and average session length are therefore not comparable across products, or across the same product before and after a settings change.

See time of inactivity for the same mechanism applied to reattribution eligibility.

What sessions are useful for

• Frequency. Sessions per user per period, which is usually more informative than length.
• Depth. What happens within a session, rather than how long it lasts.
• Comparing cohorts and sources, where the definition is at least held constant.
• Detecting regressions, since session patterns shift quickly after a bad release.

Why session length misleads

Longer is not better. A user spending ten minutes to complete something that should take two has had a worse experience, not a better one, and the metric records it as improvement.

Session length is only interpretable against what the product is for. Entertainment products want longer; utility products want shorter and more frequent. Optimising it without that context reliably produces a worse product.

In conversational products the concept strains further: an exchange may span hours with long gaps, and a timeout-based definition will split one continuous task into several sessions. This is why conversational frequency caps are set per conversation rather than per session.