Duolingo's Android Performance Case Study and DAU Growth
Article Summary
Chenglai Huang and Michael Huang from Duolingo reveal how 39% of entry-level Android users were abandoning their app before it even loaded. Their fix? A masterclass in performance optimization that saved hundreds of thousands of DAUs.
Duolingo's Android team ran 200+ A/B tests in 2024 to tackle a critical problem: slow app performance was driving away millions of learners in emerging markets. They focused on conversion metrics over latency, built automated tracing tools, and systematically optimized everything from network requests to perceived wait times.
Key Takeaways
- App open conversion jumped from 91% to 94.7% on entry-level devices
- Users waiting 5+ seconds dropped from 39% to just 8%
- Automated method tracing via ASM transform eliminated 20-minute rebuild cycles
- Deferring ads loading saved 20,000 daily users from quitting
- Showing celebration screens immediately cut perceived latency by 60%
By prioritizing conversion metrics over raw latency and optimizing for perceived performance, Duolingo gained hundreds of thousands of DAUs while making education accessible to users on the slowest devices.
About This Article
Duolingo's course models had ballooned to multi-megabyte files. Downloads were slow, and deserializing the data took several seconds, which created real performance problems throughout the app.
The team split courses into smaller pieces on both the backend and client side. Instead of loading everything upfront, users only fetched the specific content they needed at that moment.
Breaking courses into sections improved performance across the board and increased DAU. It also made room for new course features that had been turned off because of performance issues.
