Introducing surrogate metrics

Statsig now incorporates surrogate metrics into experiments, enabling faster decision-making without compromising focus on long-term outcomes. These metrics provide quick feedback by estimating long-term results that are otherwise difficult to measure during an experiment. To use surrogate metrics effectively, users must generate unit-level data with their own predictive models, inputting the model's mean squared error (MSE) for accurate error adjustment in p-values and confidence intervals. Best practices for implementing surrogate metrics include validating predictive models, measuring non-surrogate metrics alongside them, and using holdouts to verify decisions. Statsig's platform allows for the setup of surrogate metrics by accounting for prediction error and providing a confidence interval that reflects both observed variance and prediction error. This approach ensures that surrogate metrics remain unbiased estimators, reducing false positives and aligning short-term experimentation with long-term goals.