So, when advertising agencies run A/B tests, they don't publish the result in scientific journals. They just use the data internally along with art direction and relationship/political factors to make advertising decisions.
This isn't an open source field with a handy guide. This is a 'knowledge through experience' field. You learn from those people, not through academic research or Google search.
In fact, a good advertiser can instantly find holes in published research based on the many variables that a paper doesn't take into account, but an advertiser is already familiar with. "Brand value correlates with celebrity status... but inversely with the celebrity nationality.. with a correlation with the accent.. but also with season of the year.. except when using this font... and based on president's party" etc.. All these variables may have a nonlinear relationship.
To get a good introduction to what you're dealing with, just watch "Mad Men".
