Soft launch retention benchmarks: why 40/20/10 is a top-10% bar
The soft-launch gate most studios inherited is 40 / 20 / 10. Day 1 retention 40%, day 7 20%, day 30 10%.
It is not a health bar. D1 40% is the top 10% of the market.
I checked, because I have used that gate myself.
What the market actually measures
GameAnalytics measured 16,262 live mobile games for its 2026 benchmarks report (published January 2026, minimum 1,000 MAU, iOS and Android, 9 regions, measuring 2025).
| The gate | Median | Top 10% | Top 1% | |
|---|---|---|---|---|
| D1 | 40% | 20% to 22% | 40% | 64% to 68% |
| D7 | 20% | under 4% | 11% to 12% | 25%+ |
| D30 | 10% | 0.69% to 0.79% | n/a | 13% to 15% |
Read the D1 row again. The gate is the top-10% line, exactly. The D7 row is worse: a 20% gate sits above the top-10% band, and you need to be in the top 1% to clear 25%.
We built a gate out of the 90th percentile and called it the minimum.
The number that changes the decision
Median D30 is 0.69% to 0.79%. Top-1% D30 is 13% to 15%. That is a twentyfold spread on one metric in one dataset.
There is no average game to be average against. Benchmark to the median and you calibrate to a body count. Benchmark to the top decile and elite becomes the entry price.
A benchmark is a good map and a terrible gate
As a map, the data just told you something useful: the bar you inherited is an elite bar. As a gate, it tells you nothing about your game, because it was never measured on your game.
Match-3 sits higher than the market median, and yes, the market median is irrelevant to a match-3 studio. That is the point. A genre band is still someone else's distribution.
No pediatrician has ever seen a child in the 40th percentile and recommended shutting the child down. They plot that child's own curve and watch whether it holds its line.
A better kill-or-scale framework
Gate 0. Write the kill rule before the first cohort lands. A rule written after the data arrives is a negotiation, and you always win it.
Gate 1. Read the shape, not the score. D1 is a read on first-session clarity, nothing more. The decision lives in the D7 to D1 ratio and the decay curve. Retention usually decays like a power law. Fit it with three points, not two, so a retention cliff shows up instead of hiding inside a clean-looking line.
Gate 2. Check the number's age before you check the number. Thirty mature D30 cohorts take at least 60 days. Anything read sooner is a projection. Projections are allowed. Projections labelled as measurements are not.
Gate 3. Tier the prediction, do not gate on it. Predict D90 ROAS from the first three days, then act by tier. Under 50%: auto-pause at day 3. 50% to 100%: throttle to low spend and re-learn for 3 more days. Over 100%: scale, adding 25% to 30% budget every 3 days. One of the three outcomes is a kill. Most frameworks only offer two, which is why salvageable games get shot.
Gate 4. Your reference class is your own earlier cohort. Not the median, not the top decile, not your genre band. Geo one is the benchmark for geo two. It is the only gate that gets more accurate the longer the soft launch runs.
The kill rule
Kill when the curve fails its own line twice in a row, after a fix has shipped. Not when it misses someone else's number once.
Bottom line
Stop borrowing the bar. Measure your own. Kill on your own curve.
Sources: GameAnalytics, 2026 Mobile and PC Gaming Benchmarks, as reported by Game Dev Reports and Gamigion. Cohort maturity math from a16z, Play to Win: Mobile Game Soft Launch Best Practices (2023).
FAQ
What is good D1 retention for a mobile game?
In the 2026 GameAnalytics benchmarks, median D1 retention was about 20% to 22%, the top 10% sat at 40%, and the top 1% at 64% to 68%. A 40% D1 target is a top-10% result, not a minimum.
Is 40/20/10 still a good soft launch target?
It is a useful picture of what top-decile games achieve, but a poor pass or fail gate. Most games that miss it are not failures. Judge your game against its own earlier cohorts and the shape of its retention curve.
When should you kill a game in soft launch?
When its retention curve fails its own target line twice in a row, after a fix has shipped. A single miss against an industry benchmark is not enough.
How long should a mobile game soft launch last?
Long enough to read mature cohorts. Thirty mature D30 cohorts take at least 60 days. Anything read earlier is a projection and should be labelled as one.