Next generation batteries utilizing lithium (Li) metal anode could represent a 10x improvement in anode specific capacity compared to currently used graphite anodes however they still face major challenges associated with their evenness of plating and interfacial stability with the electrolyte. The capability to quickly evaluate and optimize liquid electrolyte formulations that can form stable solid-electrolyte interface (SEI) layers and bring out the best battery performance across a range of relevant performance outcomes will be critical to the realization of these next generation batteries. Integrating AI into battery research and development workflows has the potential to significantly increase the rate of new materials discovery, however, while machine learning models can successfully predict expected specific capacity or cycle life of the battery based on electrode designs, the complexity of liquid electrolytes used in batteries needs more contextual information than a machine or deep learning models can capture from simple structure-performance correlations. The lithium metal anode SEI layer is a formed out of the complex physio-chemical interactions between the lithium metal and the electrolyte and models will likely need to be able to capture those molecular attributes within the electrolyte composition that are decisive in electrolyte-Li interfacial chemistry. For this purpose, we will discuss the scope of two AI approaches that can predict battery performance based on electrolyte composition. The first approach showcases the capability of a deep learning model that is designed specially to represent electrolyte formulation based on structure-to-performance relationship. The second approach improves upon the first by incorporating transfer learning of quantum chemical attributes of electrolyte molecules and building a property-to-performance correlation. We will also discuss how transfer learning of ‘selective’ quantum chemical properties of electrolyte molecules to the capacity predicting deep learning model broadens the scope of the model towards recognizing new beneficial additives to the electrolyte formulation. Using this approach, the model retains impressive predictive accuracies despite limited experimental training data and predictions are validated and improved with experimental feedback for new electrolyte formulations in coin cell tests.