Many steps in the transaction approval process aim to model the causal factors underlying user behaviour—for example, which forms of authentication or authorization lead to less friction in payments for specific user groups. Estimating such causal effects is often done through A/B testing, which is costly and typically involves only a limited subset of users. The results of these tests are rarely reused or combined with insights from earlier experiments or observational data. This project focuses on reusing data from previous A/B tests and designing more targeted, efficient, and cost-effective experiments, with the goal of obtaining causal insights into user behaviour in a reliable and robust manner.
