Due to increasingly strict legal requirements and rising customer expectations, powertrains are becoming more complex. This complexity is evident in greater computing power and a rising number of control units in modern passenger cars. To handle the increased parameterisation effort, model-based approaches are increasingly employed to calibrate control unit functions. However, creating detailed physical models is time-consuming and expensive. Empirical modelling, grounded in real-world measurements, presents a promising alternative. This paper presents a process for model-based, data-driven calibration of control units managing dynamic driving manoeuvres and applies it to a longitudinal load-change manoeuvre of a hybrid powertrain. Data is generated using vehicle test bench trials. Employing Design of Experiments and objectification algorithms ensures the collection of relevant information about system behaviour with minimal testing effort. Supervised learning is then used to train global vehicle models that predict comfort and performance based on control unit calibration, thereby enabling targeted optimisation of driveability. The effectiveness of the methodology in reducing development time and improving development quality is demonstrated by the application example. Finally, possibilities for further development of the methodology are discussed.