According to 'Thinking Fast and Slow' which I'm currently reading expert predictors of performance are generally worse than even quite simple statistical models. This being due to all the flaws in human decision making - if it's a sunny day, if the player 'looks the part' if the scout happens to view them on a day when they're feeling lucky. It's a very good book, but it suggests that in all fields of prediction you should actually rely on combinations of statistical metrics over expert judgment.

Pretty hard to believe I accept.