Wouldn't finding a well converting copy and targeting combo throw all the statistical significance issues out the window? I mean you can start experiments with

Wouldn't finding a well converting copy and targeting combo throw all the statistical significance issues out the window? I mean you can start experiments with statistically significant results branching from that, but when you are tight on budget, and just want to start out, it might be better to just randomly find something which can get the ball rolling?