How sticky are new users?
№ 193 · Retention curve by cohort
Definition
MetricShare of a signup or install cohort still active at day or month N
Unit% of original cohort active at day or month N
Percent of subscribers retained at month N per signup cohort. Watch the M1 cliff (trial-to-paid leakage). Build from subscription event data, counting active subs per cohort per month. Step changes follow content launches. Example: Jan cohort 78% at M1, 62% at M6; Apr cohort 84% at M1 (improving).
Benchmarks
| Bottom 30% | Median | Top 30% |
|---|---|---|
| Under 2% at month 3 with no flattening at all, so the curve heads to zero and every acquired user has to be reacquired. | About 6.5% of the cohort active at month 1, and 2.5% to 4.6% at month 3. Week 1 sits near 28% on Mixpanel's cross-industry average, down from 50% in the prior period. | Roughly 26% of the cohort still active at month 1, and 15% to 19% at month 3 depending on vertical, with a visibly flat curve from month 3 onward. The flattening matters more than the level, because a curve with an asymptote has a retained core and a curve without one does not. |
Digital products instrumented in product analytics, measured as user-level return retention, meaning did the user come back, not logo retention and not revenue retention. Amplitude n=2,600+ companies with data from September 2023 to September 2024. Mixpanel n=7,700 customers and 11.7 trillion events. Userpilot n=83 B2B SaaS companies for the month 1 figure. · Three definitions are in play and the spread between them is larger than the spread between good and bad companies. Amplitude and Mixpanel measure all users returning, and land at 6.5% month 1 at the median. Userpilot measures inside an instrumented B2B product where the denominator is closer to licensed users than to all signups, and lands at a 45.3% median for month 1. Neither is wrong and they are not interchangeable. Do not compare this curve against SaaS logo or revenue retention, which run at an entirely different level, and do not compare it against mobile install retention, where day 30 across categories typically sits in the 4% to 7% range on a different denominator again. Mixpanel's week one figure is 2023 data and is flagged as such.
| Category | Bottom 30% | Median | Top 30% |
|---|---|---|---|
| B2B technology products, month 3 return retention | Not separately published, but below 1% on the observed distribution | 2.5% | 15.6% |
| Ecommerce products, month 3 return retention | Not separately published | 2.8% | 18.9% |
| Financial services products, month 3 return retention | Not separately published | 4.6% | 19.5% |
| B2B SaaS measured inside the paid product, month 1 | Under 35%, where Healthcare sits as a category average | 45.3% | Above 57%, the level FinTech and Insurance reaches as a category average |
| Consumer mobile apps, install-denominated day 30 | Under 3% | 4% to 7% | Roughly 10% and above |
When it looks bad
The curve keeps falling through month three and never flattens, so there is no retained core underneath, only a front end that decays more slowly in some cohorts than others.
Day 1 42%, day 7 19%, day 30 6%, day 60 2.4%, day 90 0.9%. No asymptote at any point, so the entire user base has to be repurchased every quarter.
What to do about it
- Attack week one before anything further out. Amplitude finds that 7% of a cohort returning on day seven puts a product in the top quartile, and that 69% of strong day seven performers were also strong three-month performers, so the curve past month one is largely determined in the first week.
- Split the curve by acquisition source and by activation status, then stop spending on sources whose cohorts never flatten. Amplitude found no relationship between top-quartile acquisition and top-quartile retention, which means volume from a bad source cannot be fixed downstream.
- Build a scheduled reason to return, sized to the product's natural frequency rather than to a notification calendar. A weekly digest on a product used monthly trains people to ignore it, and the second-order damage shows up as a lower asymptote rather than a lower day 7.
- Track resurrection rate against new-user rate on the same chart. A flat overall active count with a falling per-cohort curve is a growth mask, and it is the specific failure that a blended retention number hides.
Sources
- Amplitude Top decile products retain 4x more users at month one than the median, 26% against 6.5%. Just 10% of products account for 79% of all user engagement. amplitude.com ↗
- Amplitude At three months, top B2B products retain 15.6% of customers against a median of 2.5%, a gap of more than 6x. There was no relationship between top-quartile acquisition and top-quartile retention. amplitude.com ↗
- Amplitude At three months, top ecommerce products retain 18.9% against a median of 2.8%, close to a 7x gap. amplitude.com ↗
- Amplitude At three months, top financial services products retain 19.5% against a median of 4.6%, roughly a 4x gap and the strongest median of the published verticals. amplitude.com ↗
- Amplitude 7% of an original cohort returning on day seven puts a product in the top 25% for activation, and 69% of top day seven performers were also top three-month performers. amplitude.com ↗
- Mixpanel Average week one retention across industries fell from 50% to 28% in 2023, with no focus industry improving over the year. mixpanel.com ↗
- Userpilot Month 1 retention average 46.9% and median 45.25%. By industry: FinTech and Insurance 57.6%, AI and ML 53.6%, CRM and Sales 52.5%, MarTech 44.7%, HR 41.4%, Healthcare 34.5%. userpilot.com ↗
- Amplitude Month 1 retention for technology B2B SaaS runs 33% below the all-industry average, and day 1 activation 35% below it. amplitude.com ↗
- Amplitude Month 1 retention for technology B2C SaaS runs 23% above the all-industry average, the inverse of the B2B position. amplitude.com ↗
- RevenueCat AI apps showed twelve-month payer retention of 9.2% on the App Store and 11.5% on Google Play in the prior cycle, with AI monthly plans later retaining 36% worse over twelve months than traditional apps. revenuecat.com ↗
- Phiture Cross-industry mobile averages in 2025 and 2026 settle near 25% to 26% on day 1, 11% to 13% on day 7 and 5% to 7% on day 30. Included to mark where the install-based denominator sits. phiture.com ↗
- UXCam Day 30 retention under 5% is the norm across most mobile app categories, with the all-category median near 4%. Corroborates that install-denominated curves cannot be read against product-analytics curves. uxcam.com ↗