Tinder · B2C · dating app
Use a prediction model to show each user just the one plan they'll likely buy
Estimated multimillion-dollar yearly revenue gain in an A/B test
Workedpaywallpricingexperimentation
What they did
Problem: many tiers (Plus/Gold/Platinum) x plans (weekly/monthly) plus a la carte items caused decision overload; some users bought nothing, others bought Platinum but only used Gold features. The ML team built models predicting willingness to pay; at paywall time the paywall asks the model which paywall/product to recommend for that user, instead of showing everyone the same static paywall. Rolled out small first (a couple of features/paywalls, not the whole scope) to limit cost and risk. A/B test: control = historical static paywall, treatment = model-chosen paywall; measured conversion and total revenue. Counter-metrics: repurchase, renewal and cancellation of users nudged to higher tiers; several model variants compared against each other and against no-model paywall.
What happened
Multimillion-dollar annual revenue increase (guest's estimate 'based on our prediction'); ongoing optimization of multiple models
- Stage
- Scale (scaled consumer app)
- Effort
- Dedicated ML team; staged rollout across a subset of paywalls with A/B test
In their words
Sub Club by RevenueCat · Dynamic Paywalls That Drove Millions in New Revenue – Shawn Gong, Tinder“we can try to build ML models and then use that to predict users willing to pay and then we can surface the best product they're most likely to buy”
“our paywall will ask the ML model, "Hey, which paywall should we show?" And then the paywall will decide, okay, which product we should recommend to David”
“we want to check, hey, is David going to come back? Is David going to buy Platinum again? Is he going to cancel?”
Related topics
Get tactics like this every Monday
New tactics from the week's founder interviews, each linked to where it was said.
More from Tinder
All 22 Tinder tactics- Put a toggle in the top bar that opens a new paid tierAbout $50M in revenue from one small A/B test
- Keep dating free for most users so there are plenty of people to match85-90% never pay; the 10-15% who do fund the business
- Sell a paid 'super like' that stands out from a normal likeSuper Likes are 3x more likely to get a match
- Give each paid feature its own paywall and check its sales every morningShowed 2-3 features drive most new subs; weak ones were made free
- Offer three subscription levels and sell extra boosts on topFormer CPO calls it one of the best ways to price consumer apps; no revenue split given
- Let people start swiping in two minutes using Facebook login and photosOnboarding cut from ~20 minutes to under 2; brought younger people into online dating
More from this episode
- Tinder: Sell the travel feature on its own for 1 day, 3 days or a weekConversion jumped, but it pulled buyers away from the cheapest subscription
- Tinder: Raise the price of the standalone feature so it stops stealing subscriptionsFewer lost subscriptions, slightly lower conversion, higher total revenue
- Tinder: Price the 7-day feature pass the same as a 7-day subscription, shown side by sideLess cannibalization and more revenue, but still not good enough
- Tinder: Offer the subscription first; show the one-off purchase only to people who say no
- Tinder: Add an even pricier top tier for the biggest spendersDidn't fit a mass-market brand; being scaled down