Niche dating · 14 A/B tests · 6 months
Dating App Retention Mechanics: What 14 Tests Taught Us
Two dating products, 14 retention experiments. The pattern that emerged changed how both teams measured engagement.
What every dating app gets wrong about retention
Most dating apps measure retention with the same metrics as social products: DAU/MAU, session length, swipe count. None of these are the real retention driver in dating.
The single best predictor of D30 retention we found, across two unrelated dating products and over 400K users of cohort data, was whether a user had at least one mutual conversation that lasted 4+ messages on each side within their first 7 days. We call it the 'one good match' threshold.
How we tested it
We built a model that artificially boosted matching for new users in their first 7 days (without telling them). The test cohort hit the 'one good match' threshold 38% of the time vs 24% in control.
D30 retention in the test cohort was 18% higher. D60 was 14% higher. The lift persisted through D90 with no further intervention — once a user crossed the threshold, their retention curve permanently flattened.
What did not work
Push notification frequency tests: net zero on retention, net negative on uninstalls past a certain threshold.
Gamification (streaks, badges, daily rewards): no significant retention impact in either product. Dating users are not the same as Duolingo users.
Email re-engagement: marginal D14 lift, no D30 effect. Email is a churn-recovery channel in dating, not a retention channel.
What changed in the product
Both products restructured their new-user matching to bias toward conversation-likely matches in the first 7 days (not 'highest quality' matches by their old definition).
Both teams replaced their primary retention KPI with the 'one good match' rate as a leading indicator. D30 followed.
Running tests like this on your dating app?
We've shipped hundreds of A/B tests inside dating products. If your team is debating any of these levers — onboarding, paywall, retention, activation — we can probably tell you in a 30-minute call which test to run first.
Start a conversation →