Two failure modes of A/B tests that I've seen Big Tech popularize:

- Statistically insignificant results are not results.
- Experiments without a believable hypothesis is not a result. It's a starting point.

Problems arise from treating these as guidance/dogma in future debate.