Machine Learning (ML) technologies open up enormous potential to be unlocked through entrepreneurial activities in organizations, causing countless novel business models with ML at their core to emerge in the market. As ML technologies differ significantly from other digital technologies both in their characteristics and their effect on organizations, little is currently known about the complexities of the realization process for business models driven by ML and why only some organizations execute it successfully. By building on a qualitative study grounded on cross-industry insights from 20 expert interviews, this paper contributes to a greater understanding of the realization process by identifying ML-specific complications, before aiming to determine the underlying reasons for successful business model realization. We adopt a dynamic capabilities perspective and conceptualize eleven microfoundations that explicate how organizations build, implement, and transform business models driven by ML.