2 May 2026

N-day versus unbounded windows

Day-7 retention asks: of the people with this birthday, how many did the qualifying action in a window that starts at the birthday and ends 168 hours later? Unbounded “returned” asks: have they done it at any time since? The second number is almost always larger. It also grows as you wait, which makes old cohorts look healthier than young ones for no product reason.

We see packs that plot unbounded return as a downward curve and call it retention. The curve falls because later cohorts have had less time, not because the product worsened. Maturity footnotes exist to stop that reading. If you refuse footnotes, at least cut the last two cohorts from the chart.

N-day windows have their own trap. A person who uses the product every Sunday can miss day-7 if their birthday was a Monday. That is not churn; it is a clock. For weekly products, age in weeks with a seven-day qualifying window inside each age band is usually kinder than a fixed day-7 from an arbitrary hour.

Pick one window per pack. Print the rule. If Finance prefers unbounded because the number is higher, give them unbounded and label it “ever returned by this date,” then show n-day beside it. Two labelled numbers beat one ambitious one.

The method page lists how we teach this on the desk. The short version: the window is part of the definition, not a display option.