> If we wait until the new thing is better than the old thing, we’re taking a big risk. By definition, are we not taking risk in switching to the new thing whi

> If we wait until the new thing is better than the old thing, we’re taking a big risk. By definition, are we not taking risk in switching to the new thing which is admittedly worse than the old thing, before it's ready? Also, the threshold 77% seems so random. In many systems, it is hard to even quantify how effective is the new one compared to old. Suppose you do have metrics, then why 77%?