Stylized representations of actually diverse populations of private households are ubiquitous in research and policy. This is at odds with evidence that the transition towards a low-carbon energy system hinges on questions of social inequality, distributional impacts, fairness and justice. We apply machine learning on micro-panel data to present a detailed empirical analysis of heterogeneity in Germany's residential energy consumption. The results enable good governance by helping to predicting the distributional impacts of public policies or other events.