Skip to content
Log in
← Statistics glossary

retention plateau

Also called: retention curve, retention curve flattening, flattening retention curve, retention curve plateau

The level where a cohort's retention curve stops falling: the users who stay for good. A plateau is a strong product-market-fit signal; a curve that keeps falling to zero is a warning.

A retention curve plots one cohort's row from a cohort table: periods since sign-up on the x-axis, the share of the cohort still active on the y-axis, starting at 100% in period 0. Almost every curve drops steeply at first, because many people try a product and never return. What matters is what happens next.

  • Plateau (flattening): the losses shrink until the curve runs almost level, say at 30%. That flat part is the core of users for whom the product fits into life. It is the clearest early sign of product-market fit.
  • Steady decline to zero: every period loses a similar share, and the curve heads towards the axis. However many users you acquire, almost none stay; growth would require refilling a leaking bucket forever.
  • Smile: the curve flattens and then rises slightly as lapsed users come back, often because the product improved or network effects kicked in (friends joined). Rare and very good news.

To call a plateau, look at the change between periods, not at the level. A practical rule: the curve has flattened when it loses less than about 1 percentage point per period for two or three periods in a row. The height of the plateau depends on the category and usage rhythm, so compare it with your own segments and earlier cohorts rather than with a universal number. Read curves in the product's natural period (weeks or months for a weekly product) and per segment: a blended curve can keep falling while one segment has already flattened.

Example

Monthly retention of Halves' January cohorts, by group type:

Month
1
2
3
4
5
6
Couples (2,000 users)
55%
46%
42%
41%
40%
40%
Trips (3,000 users)
30%
14%
8%
5%
3%
2%
All (5,000 users)
40.0%
26.8%
21.6%
19.4%
17.8%
17.2%

Couples: the month-to-month losses are 9, 4, 1, 1 and 0 pp, so from month 3 on they stay at or below 1 pp. Couples plateau at about 40%.

Trips: the curve keeps sliding toward zero. That doesn't have to mean a bad product: a trip ends, and the group has no more bills. It's worth checking how many trip users start a new trip group later.

Blended: month 6 = (0.40×2000+0.02×3000)/5000=860/5000=17.2%(0.40 \times 2000 + 0.02 \times 3000) / 5000 = 860 / 5000 = 17.2\%, still falling by 0.6–2.2 pp a month (21.6 → 19.4 → 17.8 → 17.2). The average curve hides a clear plateau in one segment and a seasonal pattern in the other.

Common mistakes

  • Calling a plateau too early. Two flat points after a big drop can be noise; wait for several periods with small losses, on cohorts large enough to trust.
  • Reading only the blended curve. Segments with different jobs (couples vs trips) can have opposite shapes that average into a misleading one.
  • Using rolling retention to draw the curve. Rolling retention can only fall and is revised as data arrives, which makes curves look flatter and changes them after the fact.
  • Comparing plateau height with another category. A 40% monthly plateau is strong for a bill-splitting app and weak for a messenger.
  • Counting automatic events as activity. Server-created recurring bills can create a fake plateau of users who never open the app.