Accurately predicting the variability of thermal runaway (TR) behavior in lithium-ion (Li-ion) batteries is critical for designing safe and reliable energy storage systems. Unfortunately, traditional calorimetry-based experiments to measure heat release during TR are time-consuming and expensive. Herein, we highlight an exciting transfer learning approach that leverages mass ejection data and metadata from cells to predict heat output variability during TR events. This approach significantly reduces the effort and time to assess thermal risks associated with Li-ion batteries.