A problem with raw comparisons, like those you've just made, is that the observations in one small-sample test are likely the result of pure chance. Before launching a redesign that might be quite expensive, how can the e-commerce site assess the reliability of what it observed?