In manufacturing, a typical way to ensure correctness of an assembled product is to execute a test at end of the manufacturing process that simulates behavior under real-world conditions. As these tests are time-intensive and prune to pseudo-errors, the stations running the tests often constitute a bottleneck in the assembly process, which is difficult to remove due to the high-dimensionality and the many possible component interactions of the underlying process step. In this talk, we present how we managed to speed up the testing procedure of our electronic stability programs (ESPs) using machine learning. Beside the training pipeline and infrastructure, we also shed some light onto the internal mechanisms of the feature engineering. When machine learning is deployed to production, many things can go wrong and we also present the obstacles and learnings we experienced when making our algorithms production-ready