Why users try your app once and never come back (and how to fix it)

You shipped. People signed up. Then the graph did the thing: a spike on day one and a flat line after it. One maker summed it up better than any analytics tool could: users sign up, use it once, close the tab and forget you exist.

The usual advice is a list of 21 tactics: push notifications, streaks, badges, emails, a referral program. Some of them work. But if you add them before you know why people leave, you'll be decorating a leak. So this post is a diagnosis. There are three common leaks, they happen in order, and each one has a different fix.

What is the average retention rate for mobile apps?

Low. Lower than you'd guess. The figure most often quoted is that only about 6% of mobile users are still around by day 30. If you build with AI, the picture is a little worse again.

Treat those as weather, not a target. An average across every app in a store mixes games, banking apps and things people open once a year. The only benchmark that tells you anything is your own cohort next month compared with your own cohort this month. If 6% sounds fine because "that's average", remember that average apps mostly don't make it.

Leak 1: they never got a first win

This is the big one, and it hides behind a happy number. Signups feel like progress, so it's easy to treat "created an account" as the moment someone became a user. It isn't. A person becomes a user the first time your app does the job they came for: the first report that answers a question, the first note they find again later, the first lesson they finish, the first photo that looks better than the one they took.

Notice what didn't work there: a checklist. That's not because checklists are bad. It's because most onboarding checklists end in the wrong place. "Complete your profile, upload an avatar, invite a teammate" is a list of things you want. The user wants a result. A good first-week checklist is short and its last step is the result.

How to find your smallest useful result

1. 1

   Look at the people who stayed

   Pull the users who were still active in week four. What did almost all of them do in their first session or two that the leavers didn't? That action, not "logged in", is your candidate.

2. 2

   Make it as small as it can be and still be useful

   "Built a full dashboard" is too big. "Saw one chart with their own numbers in it" may be enough. Smaller wins happen sooner, and sooner beats complete.

3. 3

   Remove what stands between signup and that result

   Every required field and every setup step before the win is a place to quit. One creator reports their signup completion went from 60% to 100% after swapping a long form for Apple and Google sign-in (self-reported, no sample size; video). Let people see value before asking for an account if you can.

4. 4

   Write the checklist backwards from the win

   Three to five steps, each one necessary for the result, the last one the result itself. If a step doesn't lead to the win, move it to week two.

There's also research suggesting the first visit and the habit after it need different help. In a large experiment with about 60,000 students in Peru, personalized text messages did better at getting students to connect at all, while streak messages did better at keeping already-connected students practicing weekly (working paper; the streak group also got twice as many texts, so the comparison isn't clean). The lesson holds anyway: decide which stage you're fixing before you pick the tool.

Leak 2: there's no reason or moment to come back

Some people get their first win and still don't return. Often the product did its job, and then simply had nothing for them tomorrow. One tester put it bluntly about a recipe app: no benefit to revisiting the page (thread). That's not a notification problem. It's a missing return occasion.

Every useful product has a natural moment when someone needs it again. Your job is to find it, not invent it. And here's the part most retention advice skips: that moment is not always tomorrow. A habit app is daily. A team dashboard might be weekly, the morning of the standup. A tax tool is quarterly. A salon is every six weeks. An insurance helper is once a year, at renewal.

That reply is the whole idea in one line. The product already knows when the user will need it next, so it can show up at exactly that moment, once, with something useful. Compare that with a generic "We miss you!" push on day three. People can tell the difference, and they uninstall over it.

• Tie reminders to the user's own data. A renewal date, a goal they set, a report that just changed, a streak about to end. "Your weekly numbers are in" beats "Come back!"
• Respect their clock. One health-app maker traced quiet dropout to reminders arriving in the wrong time zone and a confusing Android permission prompt, not to missing features (thread).
• Don't force a daily habit on an occasional product. A note-taking app maker with 100 downloads and almost no use got a sharp reply: notes aren't daily, try a weekly review instead (thread).
• If there's no natural return at all, that's a product question. Maybe it's a one-time tool, and that's fine: charge once, ask for a referral at the moment of the win, and stop measuring daily use.

Leak 3: coming back doesn't feel like progress

The third leak is quieter. People return a few times, but each visit feels like starting over. Nothing shows them that the last five sessions added up to anything. This is where visible progress, streaks and records earn their keep, as long as you use the right one for your cadence.

Streaks: only for things that are genuinely daily

Duolingo is the case everyone cites, and its own published experiments are more modest and more useful than the legend. When it let a single lesson keep the streak alive (separate from a bigger daily goal), it reported a 3.3% relative improvement in day-14 retention, with fewer learners hitting their larger goal as the tradeoff (Duolingo). A streak wager paid with earned in-app currency reportedly lifted day-7 retention by 14% against learners not offered it, and a Weekend Amulet test reported 4% more return the following week and 5% fewer lost streaks (Duolingo). None of those posts give sample sizes or confidence intervals, and they're separate tests, so don't add them up.

Two lessons survive the caveats. First, Duolingo's daily task is small and genuinely useful: one lesson is real practice. Second, the wins came from forgiving the streak (an easier bar, a weekend off), not from making it stricter. So if you use a streak, count a real completion rather than an app open, and give people a freeze for the day life gets in the way.

Records: for everything that isn't daily

For weekly or occasional products, a streak is the wrong shape: it breaks on a perfectly normal gap and punishes the user for using your app the way it's meant to be used. A record works better. HabitKit, a habit tracker whose founder reported 120,000 downloads by the end of 2023 (self-reported), shows a full grid of your history on one screen and keeps the gaps in it (about HabitKit). A missed day is a blank square, not a reset to zero. Your last 20 weekly reports, your sessions this quarter, your fifth haircut: all of these make progress visible without turning a gap into a failure.

A win-back nudge when someone is overdue

Once you know someone's usual rhythm, you know when they're late. A weekly user who hasn't been back in two weeks is overdue. A daily user gone for three days is overdue. That's the moment for one personal, useful nudge, ideally pointing at something new in their own data, and maybe a small welcome-back bonus. Once. Not a drip of seven emails.

How to measure retention so it actually tells you something

Most retention dashboards measure opens. Opens tell you the notification worked, not that the product did. Track four numbers per weekly signup cohort instead:

1. Day-1, day-7 and day-30 return, the standard curve, so you can compare with the rest of the world.
2. First win rate: the share of signups who reached your smallest useful result. This is your activation number.
3. Second useful action: the share who did the useful thing a second time, within their cadence. One thread of AI makers proposed tracking first use and second use within seven days separately (thread). This is the best early sign that someone will stay.
4. On-cadence return in week four: the share who are still doing the useful action at their natural rhythm.

If your day-30 number looks terrible but your on-cadence number is healthy, you probably have a quarterly product and a daily dashboard. If both are bad, find which leak is first: low first win rate means Leak 1, a decent first win rate but few second useful actions means Leak 2, and second actions that fade by week four mean Leak 3.

What streaks and points can't fix

Here's the honest part. Streaks, credits, badges and nudges are amplifiers. They make a good product easier to come back to. They can't make a product useful. If people don't trust what your AI did, a badge won't help; one maker found users turned off a background AI feature because they couldn't see what it did, and weekly use came from visible, editable suggestions instead (thread). If the first session is buggy, a welcome-back bonus just brings people back to the bug.

So fix in order: the first win, then the return moment, then the progress. And if you haven't got enough users yet to see any of this in a cohort, start with getting your first 100 users and talking to every one of them.

Set it up in Lurk

Once you know your first win and your cadence, Lurk handles the bookkeeping. Your app sends the events it already has (report.viewed, lesson.finished), rules decide what happens, and every change gets a receipt you can check. Your coding agent can draft all of it, and a person on your team takes it live (how agents work with Lurk).

• First week: the Onboarding or Checklist template, with the steps written backwards from your first win. Its view shows where people stop, which is your Leak 1 diagnosis in one screen.
• Daily products only: a Streak that counts the useful action once a day, plus a Streak freeze so one bad day doesn't erase a month.
• Everyone else: a Win back rule timed to your cadence.

The Win back template starts with a fixed window ("quiet for 30 days"). Change it to match your table above: 14 days for a weekly tool, 3 days for a daily one, or twice each person's usual gap if your app sends that gap along. The credits are earned, not bought, so they can expire without trouble. Then watch the template's view of who came back after the nudge compared with similar users who didn't get one. When people do come back and get value, that's the moment to ask them to bring a friend: see how to build a referral program that works.