What is retention — and why the best apps lose users just as fast

There's one study worth seeing once and remembering forever. Quettra, together with Andrew Chen, looked at anonymised data from over 125 million phones and worked out what happens to the average app after install.
In three days it loses 77% of its active users. In thirty — 90%. In ninety — over 95%.
That's not about bad apps. That's the average. And the more unpleasant finding comes next: the best apps' curves fall roughly as steeply.
What retention is
Retention is the share of people who came back N days after their first visit.
One number, one question: did they return?
If 100 people installed your app on Monday and 25 opened it on Tuesday, day-one retention is 25%. If a week later 12 of that same hundred showed up, day-seven retention is 12%.
Note the "that same hundred". The denominator never changes — it's the original group. Not "how many people were here yesterday" but "how many are left of the ones who arrived back then". Otherwise the numbers stop meaning anything.
This is the other half of the picture. A funnel tells you whether someone reached the value in a single visit. Retention answers something else: was that value worth coming back for. You can have a great funnel and zero retention — that means you're good at persuading, but the product isn't needed.
D1, D7, D30 — and the confusion hiding there
The standard measuring points are day one, seven and thirty. Written D1, D7, D30.
But there's a fork that makes two people mean different things by "D7":
- Day-exact. Did the person show up on day seven specifically? The number comes out low and jumpy — people don't live on a schedule.
- Range. Did they show up at least once between day one and day seven? That number is noticeably higher.
Neither is "correct". The correct one is the one you don't change. If you count it one way today and the other way next month, you'll see growth that never happened. Pick one and write it down.
The curve has to flatten
Here's the main test, the reason anyone computes this at all. Plot retention by day: D1, D3, D7, D14, D30. You'll get a falling curve. Don't look at its height — look at the shape of the tail.
- The curve falls and then settles onto a shelf, stops declining at some percentage. That means there's a core of people who genuinely need the product. Your job from there is to grow that core.
- The curve falls toward zero and doesn't slow down. There's no core: sooner or later everyone leaves. However much traffic you pour in, the bucket has no bottom.
The second case can't be fixed with ads. It's fixed with the product — and usually it means you're solving the wrong problem. What helps here isn't analytics but talking to real users: five honest conversations say more than a chart.
Why the best apps lose users just as fast
Back to the study. You'd reasonably assume top apps hold people better at every interval — nudging, winning them back, reminding.
The data says otherwise: their falloff from D1 to D30 is about the same as everyone else's.
The whole difference is the starting height. They win day one. After that their curve drops almost parallel to the average one — it just runs higher.
The practical conclusion is surprising and very useful: "we miss you" pushes, win-back emails and exit discounts are work on the flattest part of the curve, where the payoff is tiny. The real lever is the first few hours. People decide whether to keep an app within the first three to seven days, and mostly on first impressions.
What to actually do
- Define what counts as "came back". Opened the app — weak. Better: did something useful — created a record, sent a message. Otherwise you're measuring curiosity.
- Compute D1 and D7 by hand against your own database. Two points are already enough to see the slope.
- Find what the returners did on their first visit that the leavers didn't. It's often one specific action.
- Make new people reach that action faster. That is the work on starting height.
- Recompute per group of arrivals rather than as one overall average — otherwise your old loyal users will paper over a collapse among new ones. More on that effect in cohorts.
What counts as good retention?
It depends on how often the product genuinely needs to be used. For a daily tracker, a D7 of 20% is decent. For something people honestly use once a month, a low D7 means nothing — you'd look at a monthly grid instead. Compare yourself to yourself a month ago, not to numbers from someone else's blog post.
How is retention different from churn?
Same thing from opposite sides. 30% retention means 70% churn. "Churn" is used more where there's a subscription and you can say precisely that someone left by cancelling. Where there's nothing to cancel, "left" is always a convention: hasn't shown up in N days.
Can I measure retention without an analytics service?
Yes. You need two things in your database: each user's signup date and the dates of their actions. After that it's an ordinary query: how many people who signed up on a given day did anything on day N afterwards. A dedicated tool saves time, but the maths here is school-level.
The AI Architect 2027 course: step by step from computer basics to your own app and first customers. 543 short lessons on your phone.





