“I hate when we do tests based on percent of user base I hate it... we need to test 1% and you got to put in the bill rights No percents No percents... if you want to do like a paint a door test you should be able to do that on 100 people or 10 people”

Test to 100 absolute users — never test by percent

Chris Hulls calls percent-based A/B testing one of his biggest pet peeves. At 100M users, 1% is 1M people — far too large for a test of something genuinely new. Run fake-door and exploratory tests on absolute Ns (100–1,000 people); if you can't see a signal at that scale, your change is too small to matter. Reserve percent-based rollouts for scaling proven wins (1% → 5% → 10% → 50%), not for discovery.