All of this madness is elegantly tackled by Bayesian statistics. There are usually two roadblocks to using a Bayesian approach: 1. It's computationally expensi

All of this madness is elegantly tackled by Bayesian statistics. There are usually two roadblocks to using a Bayesian approach: 1. It's computationally expensive. 2. You need a prior. #1 is not an issue at all here since the problem is tiny. Companies like Visual Website Optimizer are in an excellent spot for #2: because they have data from many customers, they can find a very good prior.