Mobile A/B Testing at LinkedIn: How Members Shape Our Apps
Article Summary
LinkedIn ships a single mobile binary but runs dozens of A/B tests simultaneously. Here's how they pulled it off without constant app releases.
Akhilesh Gupta breaks down LinkedIn's mobile experimentation framework that solves the classic problem: how do you A/B test native apps when you can't push updates instantly like on web? Their solution puts the server in control.
Key Takeaways
- All mobile clients render UI from view-based JSON served dynamically from backend
- XLNT platform assigns experiment buckets server-side using member attributes and context
- Bug fixes and UI changes deploy without client updates or app store approval
- Performance analyzed by device type reveals when Android outperforms iOS or vice versa
- Scaling to 100% rollout requires only tweaking experiment definitions, zero client iteration
LinkedIn decoupled mobile UI from client code, enabling rapid server-side experimentation with the measurement accuracy of traditional A/B testing.
About This Article
Native mobile apps are harder to A/B test than web platforms. Their release cycles and single shipped binary mean testing UI variations usually requires a client update.
Akhilesh Gupta's team at LinkedIn built a view-based JSON architecture where the server controls UI presentation through vType fields. These fields map to visual design elements, letting the XLNT platform assign experiments server-side without the client needing to know about it.
The XLNT dashboard automatically links experiment bucket assignments to engagement metrics, making statistical analysis possible. Teams can make confident product decisions across multiple platforms and roll out to 100% assignment without any client updates.