It is Monday morning. You open your dashboard, and the first thing you see is a big number going up. Downloads. Maybe revenue.
It feels good. I do the same thing.
The problem is that neither of those numbers can tell you the one thing you actually want to know, which is whether the people who paid you last month are still here.
And here is the uncomfortable bit. Most of the numbers that can tell you that are not on your dashboard at all. Not because you did something wrong. Because nobody puts them there by default, and the analytics tools show you what is easy to count rather than what is worth counting.
So this is the list I wish somebody had handed me. Which numbers are worth watching, what counts as a normal result for each one, and where the number is quietly lying to you.
What follows is nine numbers, each with a benchmark from 115,000+ apps to hold yours up against, and the two that mislead almost everybody who reads them. Fair warning: there is a lot of arithmetic ahead. I promise it pays for itself.
Of the people who downloaded your app on Monday, how many opened it again on Tuesday? A week later? A month later? Those three answers are D1, D7 and D30.
Track them. Just do not go hunting for an industry average to put next to them.
Here is why. RevenueCat’s State of Subscription Apps 2026 covers 115,000+ apps and publishes no install-retention curve at all. None.
The D1/D7/D30 figures you see in blog posts come from somewhere else entirely, with a different mix of apps and a different idea of what “came back” even means.
The comparison that actually tells you something is you, last month.
One more thing. These are averages, and averages lie.
Say your D30 is 18%. That single number could be iOS at 22% and Android at 14%. Or paid users at 35% and organic at 12%. Same 18%, three completely different businesses underneath, and only one of them tells you what to fix.
So split them by platform, by where the user came from, and by type, before you read anything into them.
It is worth knowing what the apps that live on this number build against it. Looking for pre-emptive retention mechanics across language learning, wellbeing, relationships and screen time, I found streaks and close to nothing else, and every mature one ships in two halves.
Each turns “came back today” into a record the user does not want to break, then names an action small enough to do while reading the notification. Speak: “Don’t Lose Your Record! Do just one lesson today to keep your record!” Alma: “Track one item to keep your streak.”
Paired: “Answer a conversation today to keep your communication strong!” Numo points at the record itself: “It’s your longest streak, don’t stop!”
The loss frame is only half of it. The single named unit, one lesson, one item, one conversation, is the other half, because a loss frame with a vague ask just produces avoidance.
Duolingo spends the freeze automatically and tells you afterwards, “You protected your 5 day streak with a Streak Freeze!” and only then offers the refill. Finch forgives you first, then offers: “It’s okay to miss a day. The important thing is you’re here today, cheep!”, then “Repair your 2 day streak? Repair for 2,000 — FREE! (1ST TIME OFFER)”.
Alan hands one out for finishing a level. Deepstash accrues them automatically and gives Pro subscribers two a week. Numo ships a “Pause & preserve”; Alma has a vacation mode that ends with “Vacation Ended! Your streak is now active again.”
Far fewer apps ship the valve than ship the streak, and that gap is the thing to measure. Pressure without forgiveness turns your most engaged users into your most abrupt churners: a 200-day streak does not decay, it ends with a deleted app. Opal shows the other failure, invoking the streak against an ordinary settings change, “Disable this rule? If you disable this rule you may lose your streak.”
This is your retention curve with money attached to it, and it is the one number RevenueCat does publish.
The first renewal is where subscriptions die. For whoever survives it, the number climbs steeply. Here are all nine figures in one place.
| Billing period | 1st renewal | 2nd | 3rd |
|---|---|---|---|
| Weekly | 35–58% | 67–75% | 77–81% |
| Monthly | 53–61% | 65–77% | 73–82% |
| Annual | 23–40% | 44–64% | 56–70% |
Read it left to right and every row has the same shape. The first renewal is a cliff. Everything after it climbs.
Read it top to bottom and annual looks frightening: 23 to 40% surviving the first renewal. It is not as bad as it looks. An annual subscriber already paid you for a whole year before that number applies to them.
The jump from the first renewal to the second is worth 18 to 27 percentage points, depending on the billing period. That is the biggest single move in the whole table.
So if you only measure one thing in this article, measure that jump.
Of users who start a trial, what percentage become paying subscribers? The cross-store median is about 32.5%, and it is almost identical on both stores (32.6% on the App Store, 32.5% on Google Play). Where it does move is category and geography.
Median trial-to-paid runs 43.5% for Travel, 37.7% for Health & Fitness, 25.0% for Gaming and 22.2% for Photo & Video.
By geography it goes from 34.2% in North America down to 15.2% in India and Southeast Asia.
Those are medians, not typical rates. So find your own category before you decide your number is good or bad. A 24% in Photo & Video is a win. The same 24% in Travel is a problem.
Before you compare it to anything, though, split it in two.
Ekaterina Gamsriegler, a B2C growth advisor who used to run growth at Mimo, turns a weak trial rate into two separate questions. How many people reach the paywall at all? And of those, how many buy?
Those are two different problems with two different fixes, and one number hides both.
If only half your installs ever see the paywall, you can lift trial starts by fixing onboarding drop-off without touching the paywall. Most dashboards carry only the second number.
What share of paid subscriptions are still active 6 months after converting? Note the unit: RevenueCat measures this per subscription, not per user.
Monthly plans sit at a 14–26% median, with the best quarter of apps reaching 30–50%. Weekly plans do far worse, though the report’s own page disagrees with itself about how much: its headline says 3–6% and its benchmark list on the same page says 5–15%. Take the direction, not the decimal.
Stretch the window to a year and the shape is unambiguous. Yearly plans retain at a 20–40% median against 6–14% for monthly and 1–2% for weekly. If your model rests on weekly billing, that is the business model rather than a bug in it, and everything depends on how fast you monetise.
What percentage of new users complete your activation event, the specific action that predicts long-term retention, in their first session? If this is low, your onboarding isn’t getting people to the value moment fast enough.
How long does it take from app open to the first meaningful value experience? Track the median, and look at the spread too, not just an average. If the median is four minutes but a large share of users never arrive at all, you have a funnel problem, not a content problem.
Of the trial cancellations you get, what share land on the same day the trial started? RevenueCat’s 2026 data: 55.4% for 3-day trials, 39.8% for 7-day, 35.7% for 14-day and 31.1% for 30-day. Read the total carefully, because it is the number most people get wrong.
That is the share of cancellations, not the share of trial users. Widen it by a day and 84% of 3-day trial cancellations and 64% of 7-day ones have already happened. Track your own Day 0 share, and if cancellations cluster there, the problem sits at the paywall moment rather than in the trial length.
One caveat on reading it, from Gamsriegler again: a cancellation inside the first day is often not a verdict on the product. In her experience it is someone who does not want auto-renew left on and would rather choose when they pay next, a control problem, and immune to a discount. The cancellations that cluster around the trial-reminder email are the ones that say the value never landed.
Which means the number is only as good as the reasons you let people give. Most exit surveys make that impossible.
Four apps in four categories carry a reason a churn dashboard almost never has. Duolingo, in language learning: “I subscribed by accident” and “I wanted to try Super temporarily”. SiriusXM, in audio: “I didn’t intend to subscribe after my trial ended”. Panera Bread, in food: “I don’t mean to sign up or don’t remember signing up.”
Photoroom, in photo editing: “The free tools are enough for my needs” and “I prefer not to say”.
Every one of those separates an accident, or a deliberate short stay, from a rejection of the product. Which is exactly the split a day-0 spike hides.
Across the eleven reason lists in my teardown the median is seven options, and every single one includes a price reason. So the default survey is built to report that you are too expensive.
Just do not add the option and then route it to the same discount as everything else. You would be collecting the difference and throwing it away.
What share of your cancellations were billing failures rather than someone deciding to leave? Billing errors account for 32.2% of Google Play cancellations against 15.2% on the App Store, a 17-point gap. Voluntary unsubscribes are 66.3% of cancellations on Play and 82.9% on the App Store; everything else, including developer-initiated cancellations, price increases and pauses, stays at or under 0.5% on both platforms.
If you are not separating these, you are reading a payments problem as a product problem.
And the fix for the payments half is mostly settings.
Dan Layfield, founder of Subscription Index, puts it at about ten minutes in the payment processor:
He puts the recovery at roughly 5% of churn, his own figure on his own clients, not a benchmark.
What share of the people who churned come back within a year? This is the number that decides whether a win-back campaign is worth building, meaning the emails or notifications you send to people who already left. And it depends almost entirely on which plan they were on. Monthly churners reactivate at 20%, weekly at 9%, annual at just 5%.
Note the population, because it is easy to misread: these are churned subscribers, people who paid and then cancelled. The figures say nothing about users who never converted in the first place.
The category spread is wide (weekly 3–16%, monthly 6–36%, annual 3–8%), with Productivity’s 36% monthly reactivation at the top and Gaming flat at 6–8% across every duration. Price tier moves it as well: high-priced monthly plans win back at 28.9%, high-priced annual at 4.4%. Geography barely moves it at all, with monthly reactivation sitting in an 18–24% band worldwide.
The practical read: if your churned base is mostly annual, a win-back campaign is too small to matter and the money is better spent before the cancellation happens. Dan Layfield, founder of Subscription Index, makes the same argument in that report.
People come back when the problem comes back, not when your win-back email lands.
So the lever is letting them pause instead of cancel, and not making them dig out a credit card to return.
His own example is Clear, about $300 a year on the family plan. He went to cancel. They offered him a two-month pause instead.
Two months was long enough to land him back in a week where he was travelling. They kept him roughly another year.
Three products in three categories pause something other than an invoice.
Atoms, a habit app, raises a sheet headed “Snooze your habit”, with the line “Take a break from this habit and return to your schedule when ready”. Four durations underneath: until tomorrow, for a week, for a month, until turned off.
Kit, a family money app, confirms with “Are you sure you want to pause PayDay? Money will no longer be debited to Alex on a regular basis. You can reactivate PayDay at any time.” and then keeps the state visible afterwards as a page heading, “PayDay paused · Edit PayDay to reactivate”.
Hims snoozes the shipment rather than the plan, and only then asks: “We’ve snoozed your shipment. Mind letting us know why you snoozed?”
And forget the kindness argument for a second. The measurement argument is stronger.
If they cancel, you get a reactivation rate you can only read a year later. If they pause with a date on it, you get a return date you can track right now, plus a group to compare against the people who left for good.
Kit’s persistent “PayDay paused” heading is the part worth copying, because a pause the user forgets they set churns on the same curve as a cancellation and you will not see it coming.
All six produce an identical screen after cancellation, because it is Apple’s and not theirs: “You have cancelled your subscription. Your subscription ends on 4 March”, a full-price Renew button, and a link to see all plans. The app supplies its name, its icon and its price, and nothing else.
If you bill through the App Store, that is where your reactivation rate quietly gets set. No offer, no context, no reason. It is a receipt and it converts like one, so spend the effort earlier in the flow or later in a channel you actually own.
Each of these either bills outside the App Store or built a second surface inside the app, which is the prerequisite. Panera Bread is the most aggressive: “WE DON’T HAVE TO SAY GOODBYE… You can still enjoy subscription benefits until 08/27”, then the whole membership benefit list under “I’d like to Resubscribe”.
Oportun answers rather than sells, “After your plan end date, you will no longer have access to Set & Save. You can restart your subscription at any time”, a “What happens after you cancel?” link, and a Resume button with no discount attached.
Strava does the least. “Your subscription is canceled and will expire on 2 Jul 2026”, and one Resubscribe Now. It sits in the normal Membership screen, so the door is visible every time the user comes back.
Which one you copy is a question your reactivation number answers. If monthly churners come back at 20% anyway, Strava’s quiet permanent door is enough. If your base is annual and comes back at 5%, this moment is not worth designing at all and the money belongs earlier, in the pause.
Single-number retention metrics miss the dimension that matters: time. A product change, a new acquisition source or a price rise can move retention a long way, and in an aggregate the effect is invisible. You only see it by comparing the subscribers you acquired in January against the ones you acquired in April.
Track at minimum: monthly cohort retention (subscribers by month of acquisition, tracked over 12 months) and new subscriber retention by acquisition source.
Realized lifetime value per payer after Year 1 is the net value of an average paying user over their first 12 months, counting initial subscriptions, reactivations, expansion and one-time purchases.
The global median is $23 per payer, and it splits hard by where the developer is based:
$44 or more puts you in the best quarter worldwide.
It is the one number where pricing, retention and trial strategy show up together. Below the median for your region you have one or the other, and the metrics above are how you tell which.
Retention metrics are only useful next to a reference point. These come from RevenueCat's 2026 report, covering 115,000+ apps, and each one changes what you would do about it.
55.4% of 3-day trial cancellations happen on day 0. On 7-day trials it is 39.8%, on 14-day 35.7%, on 30-day 31.1%. A share of cancellations, not of users.
That is a paywall-moment problem rather than a retention problem. Over half the people who quit a 3-day trial do it within hours, before the product has had a chance to do anything.
On its own that is not an argument for a longer trial, since the same report finds longer trials convert better. It does mean the screen that introduces the trial is carrying most of the weight.
Median trial-to-paid: Travel 43.5%, Health & Fitness 37.7%, Gaming 25.0%, Photo & Video 22.2%. By geography, 34.2% in North America against 15.2% in India and Southeast Asia.
Comparing your rate to a cross-category average is close to meaningless. Compare within your own category and your own market, or do not compare at all.
Trials of 17–32 days convert at a 42.5% median. Trials of 4 days or less convert at 25.5%. RevenueCat frames that gap as 70% better, about 1.7×, or roughly 17 percentage points at the median. The 5–9 day bucket sits between them at 37.4%.
And almost nobody uses the long bucket. Trials of 17–32 days are 5.0% of apps and falling, while trials of 4 days or less have risen to 46.5%. Worth reading alongside the day-0 figure: short trials lose people before the product is ever opened.
Hard paywalls convert 10.7% of downloads to paid by day 35 against 2.1% for freemium, a 5× gap. But 23% of freemium conversions do not happen until six or more weeks after download.
Pull a conversion or ROAS report at day 7 and freemium will always look broken. The window is doing as much work as the model. Retention is a separate question, and there the two are level: 28% of freemium yearly subscribers are still there at twelve months against 27% for hard paywalls, a gap the report itself calls noise.
The practical order: measure day-0 cancellations first, because it is the cheapest thing to fix and the one most likely to be quietly large. Then instrument the first renewal, which is where the curve actually breaks. Then pick your comparison set by category and region, not by industry average.
Trial-to-paid, the day-0 share of trial cancellations, renewal rate at each billing cycle, and realized lifetime value per payer. Almost everything else is a slice of one of those. The one most teams never look at is day-0 cancellation: 55.4% of 3-day trial cancellations land on the day the trial started, which points straight at a gap between what the paywall promised and what the first session delivered.
There is no single number worth chasing. The cross-store median is about 32.5%, but the median moves by category from 43.5% in Travel to 22.2% in Photo & Video, and by geography from 34.2% in North America to 15.2% in India and Southeast Asia. Compare against apps in your own category and market, and more usefully against your own previous cohorts.
Long enough for a full cohort to reach its renewal point. Judging a change to a monthly product on two weeks of data measures noise. The cohort view exists precisely because calendar-month averages hide the shape of what is happening, and the first renewal, the checkpoint where subscriptions actually break, is 18 to 27 percentage points below the second.
It depends on which plan your churned subscribers were on. Within a year, 20% of monthly churners reactivate, 9% of weekly and only 5% of annual. Those are shares of people who paid and then cancelled, not of users who never converted. If your churned base is mostly annual, the money is better spent before the cancellation, on a pause option rather than a win-back email. It also depends on who owns the screen: if you bill through the App Store, the post-cancellation screen is Apple's rather than yours, and Headway, SiriusXM, Feeld, DailyArt, GoPro Quik and Insight Timer all show the same one, a full-price Renew button with no offer attached.
Conversion design for subscription apps.
Screenshots, onboarding and paywalls, connected around one goal: getting more users to pay.