The Solvency II Directive from 2009 requires from life insurance companies to derive the full probability distribution forecast for one-year losses. Since no analytical formula for one-year losses exists and a full Monte Carlo nested calculation is computationally infeasible, life insurers utilize either the Standard Formula or a proxy methodology. Using the ideas from Least-Squares Monte Carlo proxy method, we explain how neural networks can be used to reliably predict the one-year losses opening the door to finally use risk models for other important value generating applications like asset liability management, strategic asset allocation and product strategies.