Why Duolingo’s Engagement Strategy Won’t Work For Every App — Asya Paloni, Welltory

Realized mass-market users have no daily 'job' or real-life trigger to open a wellness app (unlike athletes or sick people). Targeted the boredom trigger instead (standing in line, on the couch before Netflix): made the app as entertaining as a social feed, showing something new on every open, and removed goal-setting and shaming metrics that make users feel worse. Keep accurate information, but not as the only thing users see.

Before generative AI, built an in-house 'infinite content generation system' that changes the text and visuals users see as their data changes: about 45 message types, each with near-infinite variants (enough text for four volumes of Lord of the Rings). The guest advises using generative AI today rather than building this by hand.

Instead of charts and metrics, packed a rigorous heart rate variability analysis into a liquid visualization that changes color and behavior with the body's state: red and bubbling over when stressed, green when energized, ocean blue when calm and focused. Data is pulled automatically from the Apple Watch, so users get it by doing nothing.

Users can connect 1,000+ data sources, e.g. a developer's GitHub commits. An algorithm constantly scans for statistically meaningful correlations and sends personal ones such as 'when you run in the evenings, your GitHub commits the next day go up', instead of distant risks like heart attacks in 10 years.

For years insights lived only in a feed, which worked but users expect an organized home screen from a health app. Built a more static home screen around 'sedentary stress' that reacts to every step and heartbeat the watch tracks and explains what it means, instead of showing numbers like 'stress 48%'. Apple Watch first; other trackers later because data formats differ.

Built 'more scientific' rings than Apple's: personalized cardio zones and smarter move goals per user.

Offered personalized programs where users do things during the day, then check off what they did, like a 'Duolingo for health'.

Much of the content is shareable, e.g. the color-changing 'fuel tank' visualization ('my tank is blue today because I'm super calm'), so users can post it to their social media. Modeled on Duolingo's shareable badges.

Packed most of a weather app into an interactive home-screen widget and prioritized it over standard features competitors have. Reasoning: many people check two or three weather apps, so one delightful feature can earn the subscription even if users check other apps for the rest. Now using the users and runway gained to add the must-haves before rivals copy the widget.

For new core features (not monetization tweaks): 1) does it fit the mission (improve well-being with science, personalized); 2) which segment is it for, keeping the main persona in mind ('the Dude' from The Big Lebowski, with AI-generated images of him at the office, on the couch); 3) what trigger brings them to it; 4) what immediate and high-level job it does; 5) which of seven retention drivers it uses: magic, novelty, relief, personalization, gamification, bragging rights, social; 6) is it a must (competitors have it, fulfills an ad promise, absence causes switching or extreme frustration); 7) is it a delighter (completes a known job delightfully or creates a wow moment). Tried giving 10 points per box and building the top scorer; dropped it. Use as prompts, not a scorecard, including for competitor analysis.