Retention
D1, D7 and D30 Retention Explained
Updated 2026-09-145 min read
Short answer
Take the people who first used your app on one day. D1 retention is the share of them who came back the next day, D7 the share back on day seven, D30 on day thirty. It is a percentage of that original group, not of everyone.
Active users tells you how many people showed up. Retention tells you whether they stayed. It is the difference between a busy shop and a shop with customers, and it is measured by following one group forward rather than by summarising a window.
What a cohort is
A cohort is the set of people who first used your app on the same day. Retention always refers back to one specific cohort, which is why the denominator never changes as time passes — the group is fixed on day zero.
Day 0: 500 people install and open the app Day 1: 140 of those 500 come back → D1 = 28% Day 7: 65 of those 500 come back → D7 = 13% Day 30: 30 of those 500 come back → D30 = 6% The denominator stays 500 throughout.
The calculation
Retention rate is returned users divided by the size of the original cohort, expressed as a percentage. Nothing more complicated than that — the difficulty is in defining the cohort consistently, not in the arithmetic.
| Metric | Question it answers |
|---|---|
| D1 | Did the first session leave them wanting a second? |
| D7 | Did the app survive a week of ordinary life? |
| D30 | Has it become something they actually use? |
Why each one matters differently
D1 is mostly about onboarding. If people never return the next day, the first session failed to show them why the app is worth keeping. D7 tests whether the app fits into a normal week rather than a moment of curiosity. D30 is the one that says something about habit, and it is the hardest to move.
Note: These are not steps in a funnel. Somebody can be absent on day 1 and present on day 7. Retention on each day is measured independently against the original cohort.
Small cohorts lie
This is where most retention analysis goes wrong. A cohort of twelve people produces retention figures that swing wildly on the behaviour of one or two individuals, and those swings look exactly like real changes.
Monday cohort: 12 installs, 3 return → D1 = 25% Tuesday cohort: 12 installs, 6 return → D1 = 50% That is a doubling, caused by three extra people. Nothing about the product changed.
Careful: Before reacting to a retention number, look at the cohort size behind it. A percentage without its denominator is not information.
"Not yet" is not 0%
A cohort that installed five days ago cannot have been active on day seven. GA4's cohort report can still return a figure for it — 0% — which reads as "nobody came back" but means "not yet". Only read a day-N number from cohorts old enough to have finished day N.
Read the curve, not one number
One retention figure says how many stayed; the shape across D1, D7 and D30 says why. Compare cohorts at the same day — never a new cohort's D1 with an old cohort's D30.
| Shape | Usually means |
|---|---|
| Steep drop to D1, then nearly flat | The first session filters people out, but those who stay, stay — work on onboarding |
| Steady decline that never flattens | No lasting habit forms — work on the core value, not the first session |
| D7 higher than D1 | Weekly usage, or a small cohort — each day is measured separately against the original group |
| A new cohort below older ones on every day | Something changed for new users: a release, a traffic source, the store listing |
When daily cohorts are too small
Group new users by week. A weekly cohort of 300 people says more than seven daily cohorts of 40, and it smooths the weekday effect that makes a Monday cohort look different from a Saturday one. For apps used occasionally rather than daily, the article on utility app retention covers which windows to use instead.
On benchmarks
You will find published "good" retention figures for various app categories. Treat them carefully. They depend on how the source defined a cohort, which platforms and countries were included, and what kind of app was being measured. The comparison that reliably means something is your app against itself last month.
How this looks in Appnaly
Appnaly reads retention from the cohort data Google Analytics already produces rather than recomputing it from raw events, so the figures follow GA4 definitions. It groups new users into weekly cohorts, reads each day only from cohorts that have had time to reach it, compares the latest cohort with the average of the previous four weeks, and offers no judgement below 30 users — the honest reading of a smaller cohort is that there is not enough data yet.
If you want to work a retention number out by hand for a cohort you have counted yourself, the free retention calculator does the arithmetic, shows the formula, and shows how wide the range is for a cohort of that size.