Even though the coordination of kids’ activities is largely successful, the modern dual income family still regularly experiences breakdowns in their practices. Families often rely on routines to help them coordinate when plans prove less effective. Routines, however, are rarely documented, challenging to express in detail, and frequently evolving, making them cumbersome to manually describe and so largely unavailable to computational systems as input. This work proposes that this disconnect can be overcome, and argues that unsupervised models of family routine can be learned using a single, lightweight sensor. This way, the successful but tacit knowledge of the routine might be captured and exploited by learning systems, providing a new kind of information for families and computational systems alike. A method is proposed to develop a Bayesian Network to reason about the state of family coordination. This model relies on learned routines of pickup and drop-off at kids’ activities.