Seems more natural to take Evan Miller's approach and use the random variable defined by the difference in conversion counts between the treatment and control populations. See:<p><a href="http://www.evanmiller.org/sequential-ab-testing.html" rel="nofollow">http://www.evanmiller.org/sequential-ab-testing.html</a><p>I could be wrong though, I only skimmed the paper.