When machine learning algorithms are fed a steady stream of real-time data, they can quickly detect bottlenecks, anomalies and other operational inefficiencies.