I would go about it with a lot of precaution. There is a lot of human bias in the creation, execution and interpretation. Secondly, with regards to the sample, size is important for increasing statistical signifigance but when we are talking about sales there is a lot that can skew the signifigance of results. Finally you can’t A/B test emails on the same population, only very similar populations which again, are subject to more influences outside your control than they are within your control which can make A/B testing of emails more astology then physics.