28 July 2026
Retention curves without the noise
An N-day retention chart is a polite average of several journeys. Some people found the product useful. Some people cannot leave because their files live there. Some people opened it once from a push they did not want. The line treats them as a single travelling party.
In the workshop we draw a landscape instead. The vertical is not “percent retained”. It is presence under a definition you can say aloud: opened a valued action, not merely launched. The horizontal is calendar time bounded to a cohort whose inclusion rule survives a TV burst. If March contained a celebrity campaign, March is not a cohort; it is weather.
Smoothing is where honesty dies. A seven-day moving average can turn a Tuesday cliff — the day the trial ended — into a slope that looks like learning. Show the cliff. Annotate it. If the cliff is your business model, say that in the readout rather than hoping nobody notices the second week.
Unbounded windows are another costume. “Retention” that includes people who last appeared fourteen months ago because they still have an account is a museum visitor count. Decide what absence means. Write it next to the chart. The Analytics Atlas calls this a legend; dashboards usually omit it because legends make tiles less pretty.
You do not need a more expensive tool. You need a sentence: these people, this first valued action, this window, this reason we believe they are comparable. If the sentence is long, the landscape is mixed. Split it. Two modest charts beat one heroic curve that nobody in finance trusts.
The mild unhappiness this causes in rooms is useful. Teams that have been celebrating a gentle line often discover they have been averaging a hostage population with a delighted one. That discovery is the work. A prettier PNG is not.