I think this hints at a concept I find very hard to articulate about letting metrics guide your design. You set out with metrics that seem plausible, you work to improve them, and if they do improve, you call the design a success. But then it actually turns out the metrics were no good in the first place because they “improve” for scenarios where you’ve most certainly made the user experience worse. Metrics proponents would argue that you should have had better metrics, but then I start to feel that the problem of selecting the right metrics becomes an art form just as subjective and complicated as designing a good UI in the first place, at which point they’re no help.