Retention curve
The line showing what percent of a signup group is still active N days later, and whether it flattens out or slides to zero.
See it
What it is
Days since signup along the bottom, percent of that signup group still active along the side. Most products lose a big share of a cohort early, though how early depends on how often the product is meant to be used and how you defined 'active'. The question is what happens after: a curve that levels into a plateau means some subset keeps coming back under your chosen definition, and a curve that keeps sliding toward zero means a leaky bucket where growth only ever rents users. A plateau is evidence of a persistently retained group, not proof that anyone formed a habit. The height of the flat part is the number that matters, not day one.
There are several flavors and they are not interchangeable. N-day retention counts people active on exactly that day (spiky, harsh, good for daily-use products). Unbounded or rolling retention counts anyone active on or after that day (smoother, higher, better for weekly or occasional tools). Bracket retention buckets days into ranges. Any two curves you compare must use the same one, or you are comparing definitions rather than products.
Gotcha: the curve is only as honest as your definition of 'active'. Count app opens or push-notification taps and you can draw a beautiful plateau while nobody gets any value; anchor it to the core action instead. And a flat curve at 3% is still flat, so read the shape and the level together before celebrating.
Ask AI for it
Plot a retention curve for [cohort] from day 0 to day 30. State the cohort-entry event and the retained-activity event separately (they are often not the same event), and use every member of the cohort as the denominator for every day. Use unbounded rolling retention (active on or after day N) against the retained-activity event, computed from the observed data rather than normalized: set day 0 to 100% only if the definition logically guarantees it, meaning entry and activity are the same event, and otherwise plot the measured day 0 value. Overlay two or three cohorts as separate lines with a legend, keep the y-axis linear from 0 to 100, and annotate the plateau level with a dashed horizontal line plus its value. Note the retention definition, both event names, and the cohort sizes in a caption under the chart.