This session explores deep learning and how it can be relevant to actuaries. We begin with an overview of concepts in neural network architecture and model training – just enough to have a working vocabulary to discuss applications. Together with the participants, we develop simple models to solidify our understanding. We then demonstrate a neural architecture for forecasting losses using publicly available schedule P data and show that this novel approach outperforms a benchmark chain ladder model. We conclude by discussing, with the audience, possible extensions to the framework, additional data that can be utilized, and other potential applications in insurance of deep learning.