Machine Learning (ML) is increasingly becoming a crucial asset across diverse industries. However, designers lack human-centered processes to envision and develop innovative solutions enabled by ML. By engaging in a Research-through-Design activity, we outline a new design process to generate human-centered adaptive systems enabled by data and ML. We describe and discuss the possibilities and limits of designing with ML, the need to concurrently address user experience and ML aspects, and the implications of their mutual influence. We argue that designers can envision and design human-centered ML-enabled systems if they acquire fundamental ML knowledge, although certain tasks necessitate close collaboration with ML experts. We discuss how uncertainty and risk of failure characterize the outlined process and may limit its applicability. The proposed process serves as a foundational framework for future research in human-centered design innovation through data and ML.