Edition 03 – Data Science

Dive into the world of smart bidding for mobile game advertising: leveraging machine learning for customer segmentation and Flink stream processing to maximize auction win rates whilst maintaining profitability.

Understanding the impact of product features, especially when traditional A/B testing isn’t feasible (e.g., with fully rolled-out features), can be addressed through encouragement designs. These designs use instrumental variables to estimate treatment effects. We’ll delve into three experiment types, emphasize the critical role of assumptions, and discuss the nuances in interpreting encouragement design estimates versus standard A/B tests. Effective encouragement designs rely on well-chosen instruments and a sizable sample size for reliable results.

In this talk, I’ll explore the evolution of digital experimentation methods, emphasizing the transition from traditional A/B testing approaches to advanced causal inference techniques. I’ll show how these methodologies can uncover deeper insights and drive more informed decision-making in digital environments. Additionally, the talk will address the challenges and opportunities in implementing causal inference with a small use case.