Depends on the tests. However a point to note is that this is not classical testing followed by hard-coded values. It's more like a multi-armed bandit-type optimization.