This paper investigates hidden liquidity on U.S. equity exchanges and how to find it. Despite the National Market System’s (NMS) goal of transparent trading, we show that orders hidden from the NMS provide liquidity for 40% of the trading volume in our sample, rising to 75% of the dollar volume traded for high priced stocks. We demonstrate how price improvement on exchanges depends on interacting with hidden liquidity, especially with non-displayed orders. Leveraging big data and machine learning, we develop an algorithm that dynamically predicts where price-improving non-displayed orders are likely to appear, illustrating how AI-driven models can allow broker-dealers to more effectively meet their best execution obligations.