This document discusses using machine learning and operations research methods to improve patient flow in hospital emergency departments. Specifically, it proposes using admission prediction tools at triage to help decide whether to board patients earlier than usual (called "BERT"). The goal is to balance reducing patient wait times with avoiding unnecessary boardings. A queueing model is developed and simulations are used to test policies for setting BERT thresholds. Future work may expand this approach to predict demand for different hospital units and help operations and control of the entire hospital system. The research is a collaboration between UNC and Harvard Medical School.