By picking two major revisions or versions of an email that we'd like to compare, and finding a hypothesis we want to test. Then finding a measurable difference between the two from post-send data.