A/B testing without fooling yourself

Most A/B tests fail on sample size and on testing trivial variables. Define the winning criterion up front, test one thing at a time, and watch downstream metrics.