It is well-documented that coaches in the National Football League are frequently too conservative on fourth down. However, few decision-making tools exist for demonstrating and explaining the optimal decision on any given fourth down. In coordination with The Upshot and the New York Times, Kevin Quealy (NYT), Josh Katz (NYT), and I built the most recent iteration of the Fourth Down Bot. Trained on 13 years of play-by-play data, the bot uses live data during every game of the NFL season to determine which call (punt/field goal attempt/go for it) maximizes the probability of winning the game and provides the rationale for why this decision is optimal. We open sourced all of the code for readers to submit their own modifications and adjustments. I'll discuss the challenges of building the infrastructure, training, validating, and deploying a popular product that is, at its heart, a relatively simple statistical model. You can read more about the model here: http://www.nytimes.com/2015/10/02/upshot/a-better-4th-down-bot-giving-a….