The "some confidence level" is the problem - how do you come close to quantifying and removing all the other effects that are involved? This stuff always just becomes something that can be debated away if the advertising lead is popular, or otherwise used as an excuse to fire him/her.