How do you read cohort retention?
Read retention curves and cumulative ARPU per install cohort to project LTV.
Loading…
Related resources
Read retention curves and cumulative ARPU per install cohort to project LTV.
Related resources
A cohort is the group of users who installed at the same time. Use per-cohort retention curves and cumulative ARPU to project LTV and judge a campaign's true value.
Cohort analysis proceeds in 3 steps: (1) diagnose the retention curve's shape, (2) project the cumulative ARPU curve, (3) compare LTV by channel. Two cohorts with identical D7 retention can differ 2× in D30 ARPU — never judge on retention alone.
| Day | Commerce avg | Games avg | Social avg |
|---|---|---|---|
| D1 | 32% | 38% | 45% |
| D3 | 22% | 27% | 33% |
| D7 | 17% | 20% | 26% |
| D14 | 12% | 15% | 20% |
| D30 | 8% | 10% | 15% |
| D60 | 5% | 7% | 11% |
| D90 | 4% | 5% | 9% |
Retention curves usually follow power-law decay. Fit retention(t) = a × t^(-b) to project future points. R² ≥ 0.95 indicates a trustworthy projection.
LTV = cumulative ARPU × retention adjustment. D90/D365 can be projected from D7-era data.
{
"cohort_install_date": "2026-04-01",
"channel": "Meta",
"campaign": "AAP-iOS-KR-Purchase",
"observed": {
"d1_arpu": 0.15,
"d3_arpu": 0.38,
"d7_arpu": 0.62,
"d14_arpu": 0.95,
"d30_arpu": 1.40
},
"predicted": {
"d90_arpu_low": 2.10,
"d90_arpu_mid": 2.50,
"d90_arpu_high": 3.05,
"d365_arpu_mid": 4.20,
"ltv_confidence": 0.87
}
}The D7→D90 ARPU multiplier varies by category: commerce ≈ 4.0, games ≈ 6.0, subscription SaaS ≈ 8.0. Use these multipliers for quick LTV estimates.
Traffic quality differs on that channel. Never judge by D7 retention alone. Also compare ARPPU (revenue per paying user) and payment conversion — check whether one channel brings in more high-value payers.
Under 50 installs/day, noise dominates. Aggregate by week or by network. But never merge OSes — iOS and Android retention patterns are fundamentally different.