The Experiments That Made Hily a 40M+ User App

Because dating depends on user density, the 2017 launch concentrated performance-marketing spend in California, then added Texas for more user variety and a chance at lower CAC. Only after finding Snapchat did they scale to the whole US.

In the early struggle, launched something every day and met the next day to review results and decide next steps. Tests covered onboarding, paywalls, ad channels, UGC and creatives. Now: analytics on almost everything; every A/B test compares control vs test on many core and 'healthy' guardrail metrics.

Among early 'bold moves', tested full premium-only access: users could enter the app only after payment. No specific result given.

While also running Facebook, Google Ads and Apple Search Ads, spotted (from non-dating apps) that working with content creators to reach a younger audience worked on Snapchat, which few dating apps used. Built Snapchat-specific UGC/creator creatives and made it the primary channel; once it worked, expanded from California/Texas to the whole US. After COVID, competitors copied the approach, CAC rose with flat LTV, and budgets were cut.

Among the first dating apps to add long quiz onboarding collecting lots of user info, combined with paywall mechanics and upsells to front-load monetization and reinvest in acquisition. Onboarding answers drive personalization and matching; an MVP can skip some of it.

Tested (a) value-proposition screens before registration and (b) letting people into the product without an account to have a glance, against the simple signup-first flow. Signup-first had higher conversion to registration and from registration into the product.

Animated interstitials sprinkled through onboarding ('Nice to meet you', 'Tell us a little bit more about you', 'Awesome', 'It's time for you to shine' before photos) give a sense of accomplishment.

After post-COVID CAC rises, researched fast-growing streaming apps and combined dating with livestreams. With a bigger team and slower processes it took long to build and to find any product-market fit.

From early 2022, dropped livestreams and aimed all product changes and communication at dating quality: matching quality, quality after the match and real success outcomes for users.

Hily had two subscription groups (Hily Boost for visibility; Hily Premium for unblur, advanced filters, message before match). To reduce confusion they consolidated to one subscription, but tests showed bundling Boost into Premium worked worse than keeping Boost as a consumable bought from time to time. Right after onboarding, users see 'Be seen 10 times more. Get more matches. It's just a one-time purchase' (host says ~$3). Buyers signal willingness to pay for later offers.

The main feed shows a 'Low' visibility indicator; tapping it explains visibility is low and offers a fix, with social proof ('1,000+ women are swiping right now'). Principle: pair a felt problem (want to be noticed) with a solution and social proof rather than throwing generic purchase screens.

From time to time, change the icon: add a present, a seasonal theme or an event-related element.

Years ago, the icon artwork included a notification-style badge. Stores now prohibit it.

Tried viral/referral mechanics in the app (share links). Earlier contact-list messaging worked but is no longer possible.

Treat the permission ask as critical: test the moment, the wording and the explanation of what the user gets. Asked during onboarding (host saw an 'all notifications or only important ones' choice). More active push tokens = more users you can bring back.

30+ push logics by user segment and behavior, tuned by tests to raise opt-in and click-through without annoying users. Example: if a user doesn't click 4 consecutive pushes, enter a hold period with no pushes, then try other strategies.

Promotional pushes ('come back, you have gifts, check the people') perform far worse than real events (a like, a message). Next best: personalized content such as new people nearby matching the user's search criteria. Bombarding with generic texts loses users.

Behind the scenes, the matching algorithm raises the visibility of users Hily wants back (e.g. favorable users who didn't open the app today) in other people's feeds, so they receive genuine activity that gives them a reason to return.

Recently launched a central Discover tab with collections of users by parameters (e.g. people who like the same movie today, same music tomorrow), letting them experiment with collections and send personalized push notifications that lead back to that section.

Show basic info; cover additional info (blur) until the user uploads photos or fills in more of their own profile. More photos and info improve personalized matching.

Many CPPs for different ad networks and campaigns, with different versions for female and male audiences (e.g. a Snapchat ad targeting women links to a women-oriented CPP), and likewise per Apple Search Ads keyword; lots of tests.

Show the rating request after the user takes meaningful actions (a like, a message) rather than during onboarding, since more investment signals real interest.

For a Google Play client app, ran in-app events (Google Play 'promotional content'), which appear as a separate search result from the organic listing, plus custom store listings.