Improvements to sequential approaches in A/B testing

This paper introduces an enhanced application of sequential testing methods for A/B testing, focusing on practical implementation and scientific rigor. We build upon and refine the BI-SPRT (Bayesian Inference Sequential Probability Ratio Test) by addressing key limitations such as inflated false positive rates and computational inefficiency. Our main contributions include a simplified approximation of BI-SPRT intervals aligned with Z-tests, enabling integration into BI tools like Tableau and Amplitude, and empirical correction of false positive rates (FPR) through simulation and statistical adjustment. We demonstrate that the proposed improvements significantly enhance test reliability and feasibility in real-time decision-making environments, particularly for health metric monitoring in online experiments. These results fill a practical and methodological gap in current A/B testing practices, offering both statistical validity and industry-level applicability.