|
|
- Built a custom analysis workflow for omni-shopping backend experiments, enabling statistically rigorous evaluation of offline conversion lift at scale, supporting 4-8 experiments per month, and saving significant analysis time while accelerating model improvements.
- Unblocked statistically significant experimentation for an offline model by correcting measurement issues, expanding eligible traffic (~2.5X), and designing a fallback holdout approach, ultimately enabling the first valid results and supporting general rollout.