Our longest sessions had the lowest satisfaction.

We assumed engagement was good. Then we correlated session length with post-session survey scores across 1,800 responses.

Sessions under 2 minutes: 4.1 / 5
2–10 minutes: 4.4 / 5
10–30 minutes: 4.2 / 5
30–60 minutes: 3.6 / 5
Over 60 minutes: 3.1 / 5

The longest sessions were the least satisfying. When we looked at what those users were doing, it was repetitive manual work — exporting, reformatting, re-importing, searching through long lists. They were not absorbed. They were stuck.

We had been reporting "average session length" as a health metric. It was hiding the fact that our most engaged users were our most frustrated ones. We started tracking "time to complete core task" instead, and it became the number we optimized.

What are you calling engagement that is actually just difficulty?