Watch the seminar at https://cornell.mediasite.com/Mediasite/Play/ff46e80764554e9389c8fb763aa60e771d.

An automated system using machine learning methods, applied to a broad historical database, while avoiding survivorship bias, and for a variety of performance metrics, is developed and tested against actual historical human performance, for drafts, free agency, and trades, in the National Basketball Association (NBA). The resulting system is robust, comprehensive, realistic, and does not overfit information from the future. Backtested over ten years in a partial equilibrium non-zero-sum setting where only one team can benefit from its recommendations, the automated general manager would have outperformed the actual historical production of every single team, by substantial margins. From draft decisions alone, the average team lost about $130 million worth of on-court productivity relative to what they could have had with the automated general manager in total over the decade; this shortfall represents a quarter of the average franchise value. Thus, the general management of sports franchises may benefit substantially from automation.