Add in the ability to AB test different recommendation logic to see which one coverts higher