i built it. it's A/B tested to death.
we tripled our onboarding from 10 to 30+ screens.
everyone said we'd lose users.
instead:
→ 8M new users
→ 2.5x subscriptions on day one
→ +23% D1-D5 retention
100+ A/B tests later, here's the playbook for a "world-class" onboarding 👇
1/ test absolutely EVERYTHING
i vibe coded v1 in one night. but it took us 2 months to get it right.
i see stunning onboardings on X all the time. reality is brutal: the average user will probably drop on your fancy animation.
don't trust your intuition. measure every click, even if you have a small app. every new screen, every copy change is a bet. tiny changes can have a crazy impact. you'll learn a lot.
→ we added a "skip" button to our mascot's intro. drop-off didn't move. D1 fell 2.9%. we killed the button.
2/ think in STORYLINE, not screens
you can copy CalAI's screens or ours. it won't work for you. it depends on your users. what matters is the sequence. it's your story.
onboarding IS your product. every step should either a) give value now (advice, stats) or b) build value for later (personalization, setup). if you don't nail it, they'll never reach your actual product anyway.
don't show, ask: questions show you know their pain. every question should remind them why they're here.
then add confirmation screens: "we got you", "you're not alone", "here's how we'll help". they sell the value, not the features.
bonus: that storyline writes your paywall for you.
→ "get to your [x] GPA twice as fast with PRO"
3/ STOP optimising for funnel drop
the contrarian one. "37 screens? they must have a crazy drop."
shorter ≠ better. if someone drops halfway, what are the odds they'd have retained or paid? they'd have uninstalled the next day anyway. why optimise for them?
every screen is a trade-off. +2.5% D1 vs −1% onboarding completion, which wins? know what you're optimising for, and cut screens that aren't worth the drop.
trade conversion for retention: more activated users → more retention → more subs
and you can recover the drop: −10% completion at first, now back to 10-screen levels after swapping bad screens to converting ones.
4/ predict your BAD users
cross onboarding answers with long-term retention or subscription CR. the longer the onboarding, the clearer the patterns in how 'bad users' answer, so you can tell whether the ones dropping are worth it.
→ we ask "which feature will help you most?" users who pick 4+ retain 10% better. correlation, not cause, but it tells us which drops we can stop worrying about.
5/ quick wins
- account registration ALWAYS at the end
- copy Nikita Bier's X notification permission screen: +30% opt-ins for us
- personalization works crazy well: an app icon selector with our mascot alone gave +5.6% D1
hmu if you want more details :)
@caseycrogers Ah nice, yeah that's basically what we have by now
We also added a possibility to locally override the flag via local storage. Happy to share some code or discuss or through if you are interested
@caseycrogers We have more than 200 feature flags and are typically running 20 A/B tests at any given moment. The biggest pain point for us is actually killing unused feature flags.
@caseycrogers With "we" I mean the 7 devs flutter team at my company Knowunity (https://t.co/pdGPypTGLO).
We merged step 1-3 of your flow in one line of code. And the mgmt has never really been a pain for us. We just never remove feature flags from firebase and use the versioning in firebase