From Joule to Carnot: ex-post electricity savings achieved through residential air-to-air heat pumps

Electrification and energy demand reduction represent two primary pillars of the energy transition, alongside climate change adaptation (ESABCC 2026). Within this framework, we conduct an ex-post evaluation of electricity savings resulting from replacing direct resistive electric heating (REH) with air-to-air heat pumps (AAHPs). The dataset comprises 88 residential dwellings selected from a survey led by the French Agency for Ecological Transition (ADEME). As energy savings vary significantly across households, we combine weather-normalised daily consumption models with multivariate statistical techniques (clustering and distributional regression) to evaluate overall electricity savings and explain their variability using survey data.

Daily smart-meter data collected pre- and post-retrofit were weather-normalised using local degree-day adjustments. Statistical methods (Multivariate Adaptive Regression Splines, Factor Analysis for Mixed Data, and Generalized Additive Models for Location, Scale, and Shape) were applied to quantify energy savings and identify key explanatory factors, such as floor space, secondary heating systems, and usage patterns. The spline regressions provided baseline temperature-dependent consumption profiles. Factor analysis and hierarchical clustering revealed four distinct household archetypes based on secondary equipment and dwelling type. Finally, distributional modelling demonstrated that even detailed survey data cannot capture all factors driving energy savings, highlighting the inherent complexity of predicting real-world performance.

The empirical results show substantial variation in actual energy savings across households. Furthermore, cooling consumption during summer remains low and does not offset the net heating savings. These findings demonstrate the value of AAHPs for decarbonizing space heating while illustrating the statistical limits of predicting individual-level energy savings from field survey data.