Category: Analytics

A/B tests need a clear unit, a primary metric, and honest peeking rules. A green checkmark with +3.2% lift can still be calendar-driven science if you stop early.

Cohort retention charts need honest periods and clear incomplete cells. Build the table first, then the heatmap, so calendar scars and survivor tricks do not hide.

Pipeline observability means early signals: freshness, volume, schema, and a business check. Red, yellow, green without theater, plus a human who gets paged.

Cohorts group users by a shared start and measure them on the same clock. Define active clearly, prefer acquisition cohorts for retention grids, and label behavioral cohorts as…

Run a metrics cadence that decides: weekly for levers, monthly for outcomes. Bring roles, a pre-read, and a decision log so the hour is not theater.