We offer a novel perspective on how to manage inductive risk in the context of machine learning (ML) models used for societal purposes. Since ML models can produce erroneous results to varying extents, whether to accept a given ML model requires taking account of the inductive risk associated with its use. Deciding which levels and types of inductive risks are acceptable in the application contexts is a value-laden activity. While ML models are increasingly used for societal purposes, their epistemic features are more strongly connected with their societal purposes and underlying value judgments. We suggest that the epistemic and societal value judgments should be understood as coupled in the sense of mutually influencing each other, rather than as two separate kinds of factors, in the construction and assessment of societal ML models. We provide a coupled societal-epistemic analysis of the inductive risk in the context of societal ML models. Based on a case study concerning the societal use of an ML model by the City of Amsterdam, we suggest that finding the right balance of acceptable inductive risk is a collective quest, in the sense that this risk is prioritized and interpreted differently by the relevant stakeholders of the model.