Methodology

Where these numbers come from, and what they cannot tell you

This library exists because benchmark numbers travel badly. A figure gets quoted without its denominator, loses its cohort, changes statistic, and ends up compared against something it has nothing in common with. What follows is how this dataset tries not to do that.

Primary sources only

A source is only used when the publisher has direct access to the data it reports — subscription infrastructure that processes the transactions, an attribution SDK that observes the sessions, billing analytics that sees the invoices. RevenueCat, Adapty and ChartMogul are the platforms the money moves through. Adjust sees app opens.

Articles that quote another publisher's figures were rejected even when the figures were correct, because a second-hand number arrives without its methodology and usually without its statistic. Where a secondary article was the only route to a number, the number was left out.

Metric identity is definitional, not lexical

Two publishers can both write "D30 retention" and mean different things. Adjust means: of everyone who installed, who opened the app on day 30. RevenueCat's retention means: of paid subscriptions, which ones accumulated enough renewals to cover the elapsed time. The first is app usage by installers; the second is subscription survival among people who already paid. They differ by more than an order of magnitude and neither is wrong.

So this dataset gives them separate metric entries, records the denominator and cohort on each, and refuses to compare them. The same applies to trial conversion measured over installs versus over trial starts, to per-cycle renewal rates versus cumulative retention, and to revenue per install versus lifetime value per payer.

Statistics are not interchangeable

App performance is heavily skewed: a handful of apps earn most of the revenue. In that shape a mean sits well above a median, so the statistic matters as much as the segment. RevenueCat reports medians and quartiles across apps. Adapty aggregates conversion as the average of per-app rates. AppsFlyer's published methodology describes a trimmed mean with the top and bottom 10% removed.

Those three approaches produce different numbers from the same underlying reality. Where this dataset holds the same metric from two publishers using different statistics, the comparability verdict is downgraded and the reason is named.

Ranges stay ranges

Publishers often report a band across categories — "median 14–26%". Collapsing that to a midpoint would print a figure nobody published. Those observations are stored as ranges and rendered as ranges, and a value inside one is reported as inside it rather than above or below.

No industry average

Nothing on this site averages across sources. The function that could combine observations refuses unless every pair is directly comparable, which in practice almost never happens across publishers, and there is no code anywhere that averages incomparable numbers. When several sources measure the same thing, they are shown side by side.

This is the main thing the library does that a search result does not. A single blended number is easier to quote and tells you less.

Vocabulary is not normalised

Adjust's "APAC" and RevenueCat's "Asia-Pacific" cover different country lists. Adjust's "Social" and RevenueCat's "Social & Lifestyle" cover different app sets. Renaming one to match the other would make two incomparable numbers look like two readings of the same thing, so publisher labels are kept exactly as published and the comparability engine treats them as different segments.

The same reasoning applies to category buckets. RevenueCat's "Productivity" includes Graphics & Design and developer tools; other publishers draw that line elsewhere.

Editions, versions and breaks

Every edition of a report is a separate source. Older observations are never replaced by newer ones, and a metric is only shown as a time series when the method held still between editions.

One break is worth stating plainly. RevenueCat's 2025 edition reports 44.1% median Year 1 retention for annual plans; the 2026 edition reports 28% for the same nominal metric. A 16-point move in one year is not a market event, it is a change in how the figure is built. Both numbers are in this dataset, flagged, and deliberately not joined by a line.

Data quality found along the way

Two findings from the extraction are worth recording, because they affect how much weight the numbers can carry. Adjust's retention article uses "median" for its platform-level figures and "average" for its vertical figures without explaining the change, so the vertical numbers are marked at lower evidence quality. And Adapty's category blog pages quote three different global install-to-trial rates — 10.9%, 11.2% and 13.8% — for the same report, so only the report page itself is used as a source of values.

Neither publisher is being careless with the underlying data. Both illustrate how quickly a number degrades once it leaves the document that defined it.

What this dataset is not

It is not complete. It covers business-to-consumer mobile subscription apps well, install-cohort retention adequately, and business-to-business SaaS thinly. Several strong publishers put their numbers behind an interactive dashboard or an email gate, and those were left out rather than worked around.

It is also not advice. A benchmark tells you what other products reported under conditions that are never quite yours. It cannot tell you what your number should be.

Source inventory

Every accepted source, what was taken from it, and what limits it. A local archive of each source was kept while the data was extracted, so every figure can be checked against the document it was read from.

RevenueCat
State of Subscription Apps 2026
Published
2026-03-19
Observations
164
Metrics
17
Archived
Yes
Accessed
2026-09-09
Sample
Over 115,000 apps using RevenueCat for in-app subscription management, covering more than $16 billion in revenue and more than a billion transactions. Apps must have active subscription revenue and meet a minimum install or revenue threshold.
Aggregation
App-level metrics aggregated across apps and reported as medians and quartiles (Q1/P25, median/P50, Q3/P75, P90). Each app is a data point regardless of size.
Limitations
- Covers only apps that chose RevenueCat, which skews toward independent and mid-size subscription developers rather than the largest publishers.
- Retention is measured at subscription level, not user level, so a user with two subscriptions counts twice.
- Two charts in the State of the Market chapter come from Appfigures rather than RevenueCat.
- Segments with too few apps are omitted or merged, so absent breakdowns are not the same as zero.
Metrics: Download-to-trial (30 days) · Trial-to-paid conversion · Download-to-paid (35 days) · 6-month subscription retention · 12-month subscription retention · Renewal rate by billing cycle · Active renewal rate · Refund rate · Cancellations by reason · When annual cancellations happen · Reactivation within 12 months · Revenue per install, day 14 · Revenue per install, day 60 · Realised LTV per payer, 1 month · Realised LTV per payer, 1 year · MRR growth, year on year · Median subscription price
RevenueCat
State of Subscription Apps 2025
Published
2025-04-01
Observations
15
Metrics
5
Archived
Yes
Accessed
2026-09-09
Sample
Apps using RevenueCat for in-app subscription management, segmented by category, platform, region, price point and engagement strategy.
Aggregation
App-level metrics reported as bottom quartile, median, upper quartile and top 10% of app performance.
Limitations
- The published edition does not state a sample size, so the population behind each figure is unknown.
- Retention figures are markedly higher than the 2026 edition's for the same nominal metric, which indicates a definition or population change between editions rather than a real market move.
- Same RevenueCat-customer selection bias as later editions.
Metrics: Trial-to-paid conversion · Download-to-paid (35 days) · 12-month subscription retention · Refund rate · Revenue per install, day 60
RevenueCat
State of Subscription Apps 2024
Published
2024-04-01
Observations
19
Metrics
6
Archived
Yes
Accessed
2026-09-09
Sample
Apps using RevenueCat for in-app subscription management, broken down by pricing, packaging, localisation, conversion and retention.
Aggregation
App-level metrics reported as quartiles and medians across the app population.
Limitations
- No stated sample size.
- Download-to-paid is reported on a 30-day window here and a 35-day window in later editions, so the series is not continuous.
- Same RevenueCat-customer selection bias.
Metrics: Download-to-paid (30 days) · Trial-to-paid conversion · Download-to-trial (30 days) · Renewal rate by billing cycle · 12-month subscription retention · Cancellations by reason
Adapty
State of In-App Subscriptions 2026
Published
2026-01-01
Observations
16
Metrics
7
Archived
Yes
Accessed
2026-09-09
Sample
Subscription data from more than 16,000 apps representing $3 billion in subscription revenue.
Aggregation
Conversion benchmarks are computed as app-level rates (trials or purchases divided by installs) within a segment, then aggregated as the AVERAGE of per-app rates. Medians and percentiles are reported alongside averages on some charts.
Limitations
- Averages of per-app rates are pulled upward by high-performing outliers, so Adapty's conversion figures are not directly comparable with median-based figures from other publishers.
- Because plan duration is unknown at install, install-to-trial and install-to-paid are computed across all durations and the same value is repeated for every duration bucket — a duration-specific reading is unsupported.
- The renewal series is presented both as a 'user journey stage' funnel and as a table headed 'Global Average Retention'. This dataset reads it as cumulative retention because the series falls monotonically, but the publisher's own labelling is ambiguous.
- Adapty's category blog pages quote three mutually inconsistent global install-to-trial figures (10.9%, 11.2% and 13.8%) for the same report. Only figures from the report page itself are used here.
Metrics: Install-to-trial (all durations) · Trial-to-paid conversion · Retention after the Nth renewal · Refund rate · 12-month subscription retention · Install LTV, 12 months · Median subscription price
Adjust
Insights into what makes a good mobile app retention rate 2024
Published
2024-04-16
Observations
29
Metrics
4
Archived
Yes
Accessed
2026-09-09
Sample
Adjust's measurement data across its client apps, all verticals combined for the platform figures and broken out by vertical and region elsewhere. The published article does not state a sample size.
Aggregation
Stated as median retention rates across verticals for the overall and platform figures. Vertical breakdowns are described in the article as 'average', so the statistic is not consistent throughout.
Limitations
- No stated sample size and no stated measurement window, so the figures cannot be pinned to a period.
- The article uses 'median' for platform-level figures and 'average' for vertical figures without explaining the change, so vertical numbers are recorded here at lower evidence quality.
- 'Day' is defined as a rolling 24-hour period from install, not a calendar day — this differs from tools that bucket by calendar date and reads slightly higher.
Metrics: Day 1 install retention · Day 7 install retention · Day 14 install retention · Day 30 install retention
Adjust
Mobile app retention benchmarks for 2023 2023
Published
2023-01-01
Observations
20
Metrics
3
Archived
Yes
Accessed
2026-09-09
Sample
Adjust measurement data covering calendar year 2022, reported globally and by region and vertical. No sample size stated.
Aggregation
Explicitly stated as median global retention rates, with 2021 medians given for comparison.
Limitations
- No stated sample size.
- Reports D1, D14 and D30 but not D7 at the global level.
Metrics: Day 1 install retention · Day 14 install retention · Day 30 install retention
ChartMogul
SaaS Retention Report 2023
Published
2023-03-01
Observations
12
Metrics
3
Archived
Yes
Accessed
2026-09-09
Sample
Anonymised, aggregated data from over 2,100 SaaS businesses using ChartMogul. Only companies active for the full twelve months are included in the aggregates.
Aggregation
Benchmarks are reported as top-quartile values within ARR and ARPA bands, alongside the share of companies clearing a threshold. Company-level, equally weighted.
Limitations
- Most headline figures are top-quartile rather than median, so they describe good performance, not typical performance.
- Data is now several years old; SaaS retention fell broadly in 2022 and has not been re-benchmarked here.
- Growth-rate comparisons exclude companies below $3m ARR, so those cuts describe post-product-market-fit companies only.
- B2C figures are inferred from low-ARPA bands rather than measured as a separate population.
Metrics: Net revenue retention (SaaS) · Gross revenue retention (SaaS) · Customer (logo) retention (SaaS)

Rejected sources

Publishers that were researched and not used. Listing them matters: an absent source is usually a judgement, not an oversight.

AppsFlyer — Industry Benchmarks
Excellent published methodology, and the only one of these publishers to document a trimmed mean explicitly, but the figures themselves load into an interactive dashboard rather than appearing in the page. Its methodology is cited above; no observations were taken.
Mixpanel — State of Digital Analytics 2026
A real dataset (3.7 trillion events, 12,000+ companies), but the public page's carousel pairs industry headings with chart titles that contradict them — a "Fintech" heading above a chart labelled "Blockchain & Crypto companies" — and repeats each value with and without a unit suffix. Industry attribution could not be established with confidence, so nothing was extracted.
Amplitude — Benchmarks
2,600 companies and a clearly stated statistical approach, but the values are only available inside the interactive tool; the page itself carries none.
Superwall
Publishes product analytics features, not a benchmark dataset. There is nothing to cite.
Adapty — category benchmark blog posts
These pages quote three mutually inconsistent global install-to-trial figures (10.9%, 11.2%, 13.8%) for the same underlying report. The report page is used instead; the blog pages are not.
Business of Apps — App Subscription Trial Benchmarks
Cites RevenueCat's State of Subscription Apps, which is already in this dataset as a primary source. A second-hand copy adds no evidence.
CleverTap, Airship
Messaging and push-campaign benchmarks rather than the product metrics in scope here, and the figures are gated.
Sensor Tower, data.ai
Download and revenue estimates rather than first-party retention or conversion rates, and the State of Mobile report is gated.
General SEO articles on app retention benchmarks
A large number of pages rank for these queries by restating Adjust's or AppsFlyer's figures, frequently changing the statistic, dropping the year, or attributing a median to "average". All were rejected.

A note on language

The interface and the explanations are translated. Source quotations, report titles, sample descriptions and publisher segment labels are not — a translated quote stops being evidence, and translating "Asia-Pacific" into another language's regional shorthand would imply an equivalence with a different publisher's region that does not hold. The benchmark values themselves are language-neutral and shared across all locales.

All benchmarks Compare your product