Journal · 12 March 2026
Retention curves that flatter the first week
Day-one retention is a polite guest. It arrives early, looks healthy, and leaves before anyone asks whether the household will open the app on a bill day.
In the Cohort Signal Lab we meet the same chart every autumn: a steep, handsome D1 line, a gentler D7, and a board pack that treats the first as proof of habit. The product is usually a utility — bill-pay, insurance, a shared calendar — opened when life requires it, not when a push lands.
Vendor templates default to “opened the app within 24 hours of install.” That definition is not wrong. It is simply a different question from “will this household still complete a job next fortnight.” Mixing the two is how a growth lead ends up defending a number that finance already distrusts.
Pick an entry that is a job, not a launch
Install is a logistics event. First session is a curiosity event. First successful payment, first saved item, first completed booking: those are jobs. The Observatory asks guests to name one job and refuse to draw retention from anything earlier, even if the earlier line is prettier.
Teams resist this because the sample shrinks. Good. A smaller honest cohort is cheaper than a large one that cannot be acted on. In Circle 38, the Cardiff bill-pay pair moved entry to first successful payment and watched the D1 line collapse. They kept the collapse in the pack. That was the lesson, not a new colour on the chart.
Windows that match the calendar
If your users pay rent monthly, a seven-day window is a fashion. Fourteen or twenty-eight days will look “worse” and tell you more. We do not romanticise long windows; we require a sentence that explains why that many days matches how the product is opened.
A useful test: if you paused all lifecycle messaging for a month, would the window still mean something? If the answer depends on a Tuesday push, you are measuring a campaign, not a habit.
What this essay will not do
It will not give you a universal “good” retention rate. Those figures travel badly across categories. It will also not argue that D1 is useless. D1 is a quality check on onboarding. Treat it as such, and stop putting it in the same slide as habit.
Further reading in the house: the Lab syllabus, week four, and the Observatory glances.