Not to mention that, for abstract judgements like "product quality", these can also compound in non-linear fashion. Three A/B tests could show only minor effects on perception of product quality, but then if you put the original A next to the final B, you'll get much bigger differences.