Flipkart's Madman: An Open-Source Ads Framework for Android
Original post Madman: An Open-Source Ads Framework for Android
Article Summary
Anirudh Ramanan from Flipkart reveals how Google's IMA library was causing 15% of users to bail before videos even started. Their solution? Build their own ads framework from scratch.
Flipkart's engineering team open-sourced Madman, a video ads framework for Android, after hitting performance walls with Google's IMA library. The move came as Flipkart entered video streaming and needed better control over ad playback and UI customization.
Key Takeaways
- Madman loads pre-roll ads ~700ms faster than IMA by using native views instead of WebView
- Production rollout showed 1 second improvement in join time for videos with ads
- 15% user exit rate before video start dropped sharply after switching from IMA
- Full UI control enables custom CTAs and countdown timers impossible with closed-source IMA
- Works with any VAST-compliant ad server and integrates easily with ExoPlayer
Flipkart cut video join time by 1 second and reduced pre-video exits by replacing Google's IMA library with their own open-source ads framework built on native views.
Continue reading
What This Article Covers
Flipkart's video platform ran into performance issues with Google's IMA library. The library used WebView rendering, which caused black frames when ads started up. This made it impossible to build custom features like Rich CTAs or reliable ad countdown timers.
Flipkart built Madman, an in-house ads framework that uses native views instead of WebView. This gave them full control over the UI and let them support VAST-compliant ad servers. The framework also has extensible components for custom network and XML parsing layers.
In production testing, Madman loaded pre-roll ads about 700ms faster than IMA. When they rolled it out gradually, they saw improvements in video streaming reliability and user retention across Flipkart.
This page summarizes “Madman: An Open-Source Ads Framework for Android” by Flipkart, published Apr 1, 2020. All credit for the original work goes to the authors.
