With Daniel Gross, here is a (very much) shortened bit from Talent: How to Identify Energizers, Creators, and Winners Around the World, published at a16z, excerpt from the chapter on when to use talent scouts:
It is worth thinking about why the scouting model works in this context [finding supermodels]. First, the relevant talent could come from many different parts of the world, and the number of people to be scouted is very large. It is hard to imagine a centralized process getting the job done. Second, many of the scouts plausibly have a decent sense of who might make a good model. Looks are hardly the only factor behind modeling success, but they are a kind of “first stop,” and expecting the scouts to judge looks well from first impressions is more plausible than expecting the scouts to use first impressions to judge talent well for skill in, say, quantum mechanics. Third, a follow-up investigation to judge the modeling talent of the chosen candidates is not extremely costly. You can have them in for a photo shoot and see how popular they prove in the market without having to invest millions of dollars right away…
Scouting is also becoming more important as the options for self-education are rising. With more people trying their hand at various avocations than ever before, that places more and more burden on talent search. We need to be more open to the accomplishments of self-taught individuals without traditional training, and that holds all the more true for the tech world, where many of the most important founders have eschewed the institutions of traditional education.
There is much more at the link, we also consider when scouting models fail relative to centralized evaluation, and which kinds of incentives should be given to scouts.