Parametric or non-parametric?

In order to leverage the parametric tests for statistical significance we need to confirm that our assumptions about randomness, normality, and independence are satisfied. Most online experiments run in-practice will satisfy these assumptions for large traffic sizes. However, depending on your experimentation field and industry, it is not uncommon to come across cases where any of the assumptions is in violation. When this is the case, it is safer to go with the appropriate non-parametric test that doesn't require strict assumptions about the data.