2026-05-12
How late events distort weekly retention figures
When mobile clients buffer events overnight, Monday retention can look healthier than Friday’s cohort actually was. Here is how we isolate the lag.
Late-arriving application events are ordinary. What breaks trust is treating a cut-off timestamp as if every device had already flushed its buffer. Weekly retention reports that close Sunday midnight often still receive Tuesday payloads for the same cohort.
We start by comparing the report’s close time with the pipeline’s maximum accepted event age. If the warehouse accepts events older than the report window implies, the published figure mixes completed and incomplete cohorts. A practical check is to freeze two snapshots: one at the official close, and one forty-eight hours later, then measure the delta per cohort day.
Teams in Seoul that ship consumer apps with aggressive battery policies see this pattern often. The fix is rarely “collect faster.” It is to publish retention with an explicit lag policy, and to mark preliminary versus final figures in the same place stakeholders already look.
If your Monday pack routinely revises Friday’s numbers without a documented lag rule, the pipeline is not lying—the report contract is incomplete.