It's difficult to remove that randomness from the process. As I see it the only reliable approach for both sides, given that structure, is volume - you just do/conduct a lot of interviews and evaluate/get evaluated on the top percentile.