Retention is the percentage of users from a cohort who return to the app N days after their install. D1 = day-1 retention (returned the day after install), D7 = 7-day, D30 = 30-day. It's the leaky-bucket metric — a product with great acquisition and terrible retention burns through audience faster than it can replenish, while a product with decent acquisition and great retention compounds. After install volume, retention is the single most important non-revenue metric in mobile.
That distribution is the truth most retention articles obscure. The median catalog app retains under 4% of installs by day 30 — most "good benchmark" numbers floated in industry posts describe top-decile or top-quartile performance, not the actual middle. If your D30 sits in the 5-10% bucket, you're already top-quartile. If you're above 10%, you're in the top-decile band where consumer-social, productivity, and finance apps live.
Industry benchmarks by category (rough 2026 anchors):
- Messaging / banking / habit apps: D1 50-70%, D7 30-50%, D30 25-50%. Long-tail anchored by structural daily need.
- Consumer-social / productivity: D1 40-50%, D7 20-30%, D30 10-15%.
- Streaming / media: D1 35-45%, D7 15-25%, D30 8-15%.
- Casual games: D1 35-45%, D7 12-20%, D30 5-10%.
- Hyper-casual games: D1 25-35%, D7 5-10%, D30 1-3%. Designed for fast monetization, not retention.
- Utilities: enormous spread — task-specific (calculator, flashlight): D30 < 5%; recurring (period tracker, weather): D30 > 30%.
Compare to your own category and your own historical baseline. Cross-category comparisons mislead.
Retention curves follow a universal shape: steep drop in the first week (most installs churn fast), slower decay over weeks 2-4, then a nearly-flat long tail. The mathematical shape is "exponential decay + asymptote" — the asymptote is the fraction of users who become "permanent" users, who'll be active months or years from install. The asymptote is the number that really compounds. A product that retains 10% of installs indefinitely scales fundamentally differently from one that retains 3% — over time, the permanent-user base accumulates from every cohort, and that compounding is the engine of organic growth.
The distribution above is heavily right-skewed: tens of thousands of catalog apps cluster between 1-5% D30, a long tail of strong apps clears 10%, and a tiny strong-app tier crosses 20%. The implication for product teams: D30 = 5% feels mediocre against blog-post benchmarks but is actually top-quartile across the measurable market. Compare against your category's median in the table below, not against headline numbers from messaging or banking case studies.
D30 retention calculator
Enter how many of an install cohort were still active on day 30 to get your D30 retention rate, then see where it lands.
Enter your numbers to see your result and how it compares to the catalog.
Benchmarks: MWM data, US, apps with ≥1,000 d30 downloads. Compare to your category median too.
Three retention measurement methods to know:
- N-day retention (classic): the user returned on EXACTLY day N. Strict, low number, easy to compute. Standard for D1 / D7 / D30 benchmarks.
- Rolling retention (looser): the user returned on day N or any day AFTER N. Higher number, smoother curve, useful for low-frequency products.
- Range retention: the user returned at least once during a day-N to day-N+M window. Used when daily granularity is too noisy.
Different products use different defaults — what matters is consistency. Mature analytics platforms (Amplitude, Mixpanel) expose all three and let you pick.
Levers that move retention (in rough order of impact):
- Onboarding completion rate — users who complete onboarding retain 2-3× higher than those who don't. Single biggest D1 lever.
- Day-1-to-day-2 reactivation — push notification, email, in-app prompt timed to bring the user back within 24 hours. Lifts D7 substantially.
- Habit-loop mechanics — daily streak, daily-content drop, daily-routine integration. Drives the long-tail asymptote.
- Product-market fit — at the macro level, retention is the truest expression of fit. If retention is structurally weak across all cohorts, the answer is usually product, not marketing.
D1 / D7 / D30 retention benchmarks by category (2026)
| Category | D1 | D7 | D30 |
|---|---|---|---|
| Messaging / banking / habit apps | 50-70% | 30-50% | 25-50% |
| Consumer social / productivity | 40-50% | 20-30% | 10-15% |
| Streaming / media | 35-45% | 15-25% | 8-15% |
| Casual games | 35-45% | 12-20% | 5-10% |
| Hyper-casual games | 25-35% | 5-10% | 1-3% |
| Utilities (task-specific) | 15-30% | <5% | <5% |
The asymptote — long-tail flat retention rate that survives beyond D90 — matters more for LTV than any single D1/D7/D30 number. A product retaining 10% of installs indefinitely scales fundamentally differently than one retaining 3%.
Median D1 / D7 / D30 retention by category (US, Q3 2025, MWM)
| Category | D1 | D7 | D30 |
|---|---|---|---|
| Social & Communication | 31.9% | 12.3% | 5.9% |
| Lifestyle & Well-being | 23.6% | 9.6% | 4.8% |
| Productivity & Tools | 23.0% | 8.9% | 4.5% |
| Education & Knowledge | 24.9% | 8.6% | 3.6% |
| Media & Entertainment | 24.9% | 8.0% | 3.4% |
| Game | 36.6% | 9.3% | 2.9% |
Two facts the category table makes vivid. Games have the highest D1 retention in the catalog yet the lowest D30 — the classic decay curve, where flashy onboarding wins day 1 but habit-loop mechanics fail by month 1. Social & Communication leads at every horizon — the long-tail asymptote is where structural-network products dominate. If your retention curve looks like the Games row (high D1, collapsing D30), the lever is mid-funnel re-engagement; if it looks like the Education row (modest D1, steady decay), the lever is value-density and progression depth.
Does retention vary by country?
Common assumption: retention is structurally lower in emerging markets because users have lower intent, more storage pressure, more aggressive app churn. The catalog data contradicts this myth.
Median D1 / D7 / D30 retention by country — Tier-1 vs emerging markets (Q3 2025, MWM)
| Country | D1 | D7 | D30 |
|---|---|---|---|
| United States | 27.3% | 9.2% | 3.9% |
| United Kingdom | 27.2% | 9.0% | 3.9% |
| Germany | 27.4% | 9.0% | 3.8% |
| France | 27.3% | 9.0% | 3.8% |
| Japan | 27.4% | 9.1% | 3.9% |
| South Korea | 27.4% | 9.0% | 3.8% |
| Brazil | 27.1% | 8.9% | 3.8% |
| India | 27.5% | 9.1% | 3.7% |
Median D1/D7/D30 retention is essentially flat across major markets. Within a half-point of the US baseline you find the UK, Germany, France, Japan, South Korea, Brazil, India. The "emerging-market apps churn faster" narrative doesn't hold up — at the median, retention shape is structurally similar across geographies. The variation between MARKETS is dwarfed by the variation between apps within any single market. Localization, language coverage, and content cadence move retention more than country mix does.
Common retention mistakes
Most retention problems aren't measurement problems — they're predictable misreads of the curve. The recurring ones:
- Chasing D1 while the D30 asymptote rots — a flashy onboarding can inflate day-1 without building the habit loop that drives the long-tail rate. D1 looks good in the dashboard; the asymptote is what compounds into LTV.
- Treating retention as a re-engagement problem when it's a product problem — if retention is structurally weak across *every* cohort and channel, no amount of push or email fixes it. Weak retention everywhere is the truest signal of missing product-market fit.
- Reading blended retention instead of cohorts — one blended curve hides which acquisition source, platform, or onboarding variant is leaking. The blended number can hold flat while a paid channel quietly collapses underneath it.
- Over-notifying to juice D7 — hammering users with push to drag them back lifts short-horizon retention but drives notification fatigue and uninstalls, hurting the very asymptote you're trying to grow. Frequency discipline beats volume.
- Benchmarking against the wrong category — comparing a casual game's D30 to a messaging app's is meaningless (the catalog spans 1% to 50%+). Compare to your category median, and above all to your own historical baseline.
The textbook positive counter-example is the streak-plus-reminder loop (Duolingo being the canonical case): a daily-return mechanic with a visible cost to breaking it, backed by well-timed re-engagement — which builds the long-tail asymptote rather than just the first-week numbers.