When I worked on ads at Google we had many A/B tests that had been running that long, generally holdbacks where a feature was almost entirely but not 100% launc

When I worked on ads at Google we had many A/B tests that had been running that long, generally holdbacks where a feature was almost entirely but not 100% launched. It was relatively rare that the holdback would show markedly different results than the initial A/B test we used in deciding to launch. If that had happened more often we would have run more long tests and been slower to move to launch.