Randomness is a reasonable policy until proven inferior. It's efficient and ensures no systematic failure. I'd prefer an experiment comparing any selection strategy against randomness before using that strategy. Even then, it'd be helpful for future experiments to continue to use some random selection indefinitely.

I've worked with many large "successful" companies over the years. At every one of them I've found bureaucratic procedures which are worse than random. They're the kind of convoluted process which started out as a good idea and then morphed over the years into a tradition for its own sake -- "that's the way it's done!" -- and gathered false post-hoc rationalizations.