Hot stamping is one of the key technologies to produce lightweight components for the automotive industry. Numerical simulations are an important tool to design hot stamping tools. However, the hot stamping simulations are highly complex due to the deformation as well as heat transfer and phase transformation mechanisms taking place simultaneously. An accurate prediction of these mechanisms leads to high computational cost. Conventionally, an elastoplastic material model with temperature and strain rate dependence is used to describe the forming behavior, while austenite decomposition models are used to predict the kinetics of the phase transformation. The time spent on calculating the phase transformations can reach up to 40% of the total calculation time. To overcome these computational time issues, the current study introduces an alternative fast approach, to reduce the total calculation time of the hot stamping simulation during the die design stage. The simulation is sped up by replacing austenite decomposition models with surrogate models. A machine-learning (ML) model is used as a candidate for calculating the final phase fraction of hot-stamped parts. To train the model, various simulations models with varying degrees of deformation and cooling rates are set-up in the commercial FEM program LS-DYNA®. The ML model learns from the deformation histories and the final phase fractions to achieve the desired accuracy. For validation, the proposed approach is tested and compared with fully integrated phase transformations models. The final phase fraction and the total amount of floating operations are used as indicators to evaluate the performance of the fast approach. The results indicate that the ML model performs well in predicting the final phase fraction after hot-stamping and a significant reduction of the calculation time evaluated from the counting of floating operation FLOPS is reduced from 5713 to 12.