A White Box Framework for Estimating Long-Term Experimental Impact via Short-Term Metrics

Experiment decisions often require not only long-term treatment effects, but also insights into the underlying causal mechanisms. We present a white box framework that estimates long-term treatment effects by linking short-term key metrics to a long-term north star and its driving mechanisms. We also introduce validation methods using pseudo-experiment data and actual experiment data. We further outline the estimation and deployment considerations on an experimentation platform.