Abstract: Traditional A/B testing methods can be resource-intensive, as each experiment requires its own control group. In this talk, we introduce the concept of shared control groups in A/B testing, which allows multiple experiments to share a common control group while preserving statistical power and confidence intervals. This approach can significantly boost experimentation throughput, enabling organizations to to test more ideas in less time. We present the theoretical framework for shared control groups and demonstrate how it has helped in dramatically increasing experimentation throughput at Meta.
Bio: Sonam Saxena is a seasoned Product Manager with over two decades of experience in the technology sector, having contributed to leading companies such as Microsoft, Google, and Meta Platforms. At Meta, he is part of the Applied AI team and leads the AI/ML Experimentation Team, focusing on enhancing advertiser and customer outcomes through advanced AI and ML models. Sonam's expertise encompasses Experimentation, Artificial Intelligence, ML Infrastructure and advertising.



