From a Stumbling and Mumbling post:
A key insight here comes from Marko Tervio. He argues that what matters isn’t so much talent as proven talent. Many hirers would prefer the known quantity who is just above a threshold of competence to the unknown one who might be brilliant but might also be a duffer. In hiring a factory manager, you want someone who isn’t going to blow the place up. In hiring a journalist, you want someone who can be relied upon to file something literate and on time. And so on.
This situation goes both ways - choosing predictability over variance reduces your chances of hiring a duffer, but also of hiring a genius. Growing companies can afford more variance, and benefit more from the stars, so tend to take more chances.
I'd say there is also a factor where many hiring managers don't actually have a good concise way of describing the job they actually need done. Part of the reason automation isn't going to suddenly replace everyone is that there is a lot of ambiguity in many jobs, at all levels, but that same ambiguity opens the door for all manner of biases.