Machine learning (ML) is key to achieve complex automation like in self-driving cars: implementation of implicit requirements and faster time-to-market are just two promises. Despite technological advances, research questions remain open about improving the level of trust and quality (quality in terms of ISO 25010) that can be placed on such ML-based systems. Their quality depends on the quality of the data used for training and appropriate verification and validation. This data quality - and with it the confidence in ML - relies on a systematic and structured process incorporating hierarchical requirements engineering for the quality and composition of data sets. This paper presents the data-driven engineering process (DDE process) as a new systematic and structured approach for leveraging future application of ML in industry. The DDE process includes hierarchical requirements engineering to link the operational design domain with the requirements and semi-automated generation of data sets. We describe the DDE process as a V-model that is fully integrated with other engineering processes. It represents a consistent approach that harmonizes development abstraction levels and DDE for ML as a third technology next to hardware and software (section III). Furthermore, the DDE process allows process automation leading to automated data set compilation. Applicability of the DDE process is shown by an application example using a convolutional neural network for traffic light detection (section IV). A summary and next steps are concluding the paper (section V).