Why mobile retention curves are getting flatter, and what it means.
A small but consistent pattern in the published data from twenty-three mid-stage app companies. We think it is real, we think it matters, and we have some theories about what is causing it.
The retention curve, in its classic shape, drops sharply in the first few days after install, levels off somewhat by the end of the first month, and then continues to decay slowly thereafter. This shape has been, for most of the decade since the major analytics platforms started publishing benchmarks, the most consistent pattern in the published mobile-app data. The exact numbers differ by category — a meditation app's day-30 retention is structurally different from a casual-game app's — but the shape of the curve is broadly similar across categories.
Over the past eighteen months, the shape has started to change. Across the twenty-three mid-stage app companies whose published retention data we have been able to reconstruct, the curves are getting flatter. The day-1 drop is, in most categories, becoming less steep. The early-week decline is also softening. The day-30 number, paradoxically, is in many cases stable or even slightly lower than it was eighteen months ago — but it is being reached more gradually, by a curve that is less steep on the way down.
This is, on its own, a small change. It is also, in our reading of the data, a real one. We have looked at it carefully enough to be reasonably confident it is not a measurement artefact. The question worth asking is why it is happening.
What the curves actually look like
Let me describe the change concretely. In a typical productivity app in mid-2024, the day-1 retention figure was around 38 per cent — which is to say, about 62 per cent of installers did not return to the app the day after they installed it. In the same productivity app in mid-2026, the day-1 retention figure is around 44 per cent. The day-30 retention, in both periods, is around 13 per cent.
The reason this is interesting is that the curve between day 1 and day 30 has, in the same period, become substantially less steep. In 2024, the path from 38 per cent at day 1 to 13 per cent at day 30 involved a sharp drop over the first week and a slower decline thereafter. In 2026, the path from 44 per cent at day 1 to 13 per cent at day 30 involves a much more gradual decline throughout the period. The end point is, broadly, the same. The shape of how the app gets there has changed.
This is, I want to stress, the average pattern across the twenty-three companies in our sample. Individual companies vary considerably. Some companies are seeing the flatter curve combined with a higher day-30 number; some are seeing the flatter curve combined with a lower day-30 number. The flattening of the curve is, however, the common feature across almost all of the apps in the sample.
Why we think it is happening
We have three working theories. None of them is mutually exclusive with the other two, and our suspicion is that all three are contributing to the pattern to varying degrees in different categories.
Theory one: the new onboarding norms. The mobile-app industry has, over the past three or four years, broadly converged on a set of onboarding norms that ask the user for more upfront engagement — questionnaires, intent-detection flows, the longer onboarding sequences we have written about elsewhere in this issue. Users who complete these onboardings are, by the structure of the onboarding itself, more invested in the app by the time they reach day 1. The day-1 number, accordingly, looks better. The day-30 number does not improve in proportion because the underlying engagement curve, beyond the onboarding period, has not changed as much as the onboarding has.
Theory two: the change in acquisition channel mix. The marketing budget shifts we have written about elsewhere have, as a side effect, changed the average source of new installs. Apps in 2024 were acquiring a much larger fraction of their users through paid social channels, where the user typically arrived with relatively weak intent. Apps in 2026 are acquiring a larger fraction of their users through organic search, content, and product-led referral channels, where the user typically arrives with stronger intent. Stronger-intent users have, on average, flatter retention curves. The shift in channel mix is, therefore, mechanically producing flatter curves at the aggregate level.
Theory three: improvements in the first-week experience. The mobile-app industry has, in our admittedly anecdotal observation, become noticeably better at the first-week experience over the past three years. Lifecycle messaging is better. Activation prompts are more carefully designed. The product surfaces that a user encounters in the first week are, in most apps we have audited, more thoughtfully built than they were three years ago. The cumulative effect of these improvements is to soften the early-week decline that used to be the steepest part of the retention curve.
"The retention curve is getting flatter because the apps are getting better at the first week. This is a small piece of good news in an industry that does not often get to celebrate small pieces of good news."
What this means for the rest of us
The flattening of the retention curve has, we think, several practical implications worth flagging.
The first is that the standard benchmarks — "good day-1 retention is around X per cent for this category" — are quietly becoming less useful. The benchmark numbers were set, in most cases, on data from a period when the curves had a different shape. Comparing your current day-1 number to a benchmark set in 2022 or 2023 will, in many categories, make your current performance look better than it actually is. The relevant comparison, increasingly, is not the day-1 number but the shape of the whole curve.
The second is that the relationship between day-1 retention and day-30 retention has weakened. It used to be a fairly reliable rule of thumb that a 10-point improvement in day-1 retention would produce a 2-to-3-point improvement in day-30 retention. In our sample, that relationship has broken down. Day-1 improvements that are largely driven by better onboarding do not, in most cases, translate proportionally into day-30 improvements. Day-1 improvements that are driven by better channel mix or by better first-week experience tend to translate better. The mechanism matters as much as the metric.
The third — and most strategic — is that the work of improving retention has, in 2026, become considerably more product-centric than it was three years ago. The big retention wins are no longer, in most categories, available through marketing-led levers like lifecycle messaging or notification timing. The big wins are, increasingly, available through product changes — improvements to the experience of the first session, the second session, the first week. The retention curve is being shaped by the product team in ways that the marketing team has limited leverage on.
What we will be watching
The next data refresh on this study is scheduled for early autumn. We expect to have, by then, a clearer picture of whether the pattern we have described is continuing, accelerating, or reversing. Our working assumption is that it will continue, but we will revise that assumption if the data argues otherwise.
In the meantime, we would say to anyone reading their own retention data in 2026: pay more attention to the shape of the curve than to any single point on it. The curve is, increasingly, telling a more interesting story than the headline number ever did.
Methodology, briefly
The sample consists of twenty-three companies whose retention figures we have been able to reconstruct from a combination of: published quarterly reports, public investor materials, conferences talks given by the relevant teams, and data shared with us on background by the marketing leads themselves. The sample is not representative of the broader app industry; it is biased towards companies that publish more rather than less data about their retention figures. We have, however, controlled for category in our comparisons, and the pattern is visible across all the categories represented in the sample. The data and the methodology are available on request to bona fide researchers; we do not, by agreement with several of the contributing companies, publish the company-level data.