Most “cohort dashboards” we inherit are beautiful and abandoned. They mix acquisition week with activation week, hide small-n cells, and require three filters before a founder can see whether March’s customers still show up. Product teams do not keep what they cannot paste into a Monday note.
A table that fits in a document
Rows are activation weeks, not signup weeks, unless you are explicitly studying top-of-funnel. Columns are D7, D30, D90 retained-after-first-value. Show raw counts next to percentages. Hide nothing under n = 20; write “n too small” instead of a confident 80%.
One extra column: the definition of “retained” in twelve words. If that sentence changes, the historical columns must be marked stale. Silent definition drift is how a team argues about a chart that no longer measures the same behaviour.
What to retire
Heatmaps that use five colours for a 12-person cohort. Rolling windows that quietly include users who activated after the column’s end date. Any view that cannot be explained to finance in one minute. Those belong in a sandbox, not in the weekly paragraph.
If the table cannot survive a screenshot into a chat app, it is not an operating artefact yet.
North-Star Architecture spends module three building this table with your numbers. Bring two quarters. Leave with something ugly enough to trust.