Pitfalls
Cohort charts that quietly mislead product reviews
Cohort heatmaps are easy to love and easier to misuse. Color scales can flatter tiny samples, and truncated axes can turn noise into a narrative.
The vanishing small cohort
When a late-month cohort has twelve users, a single retained account paints a triumphant cell. Require a minimum n before a cell earns saturated color, or show counts beside percentages.
Comparing unequal starts
Paid acquisition weeks and organic weeks are not siblings. If you must show them on one grid, split the legend or facet the chart. Mixing them invites false “product” conclusions.
Color without a legend contract
Teams inherit palettes from design systems that were never meant for sequential retention. Lock a single hue ramp and publish the breakpoints in your metric dictionary.
A fix we teach in critique studio
Before presenting, answer three prompts aloud: who is in the cohort, what action counts as retained, and what decision this grid will change this month. If any answer wobbles, the chart is not ready.