In this talk, we show how both novel data sources and a new machine learning technique can be used for predicting defaults of loans made to Swiss small and medium-sized enterprises (SMEs). Advanon is a Swiss fintech startup that operates a platform on which SMEs can obtain loans by pre-financing open invoices. We highlight some of the practical challenges when applying machine learning to the task of default prediction and present possible solutions. In particular, we introduce a novel model, called the Grabit model, which is obtained by applying gradient tree boosting to the so-called Tobit model. We obtain a large improvement in predictive performance compared to other state-of-the-art approaches. See Sigrist and Hirnschall (2017), https://arxiv.org/abs/1711.08695, for more information.