In the current practice, the design of energy refurbishment interventions for existing buildings is typically addressed by performing time-consuming software-based numerical simulations. However, this approach may be not suitable for preliminary assessment studies, especially when large building portfolios are involved. Therefore, this research work aims at developing simplified data-driven predictive models to estimate the energy consumption of existing school buildings in Italy and support the decision-making process in energy refurbishment intervention planning at a large scale. To accomplish this, an extensive database is assembled through comprehensive on-site surveys of school buildings in Southern Italy. For each school, a Building Information Modelling (BIM) model is developed and validated considering real energy consumption data. These BIM models serve in the design of suitable energy refurbishment interventions. Moreover, a comprehensive parametric investigation based on refined energy analyses is carried out to significantly improve and integrate the dataset. To derive the predictive models, firstly the most relevant parameters for energy consumption are identified by performing sensitivity analyses. Based on these findings, predictive models are generated through a multiple linear regression method. The suggested models provide an estimation of the energy consumption of the “as-built” configuration, as well as the costs and benefits of alternative energy refurbishment scenarios. The reliability of the proposed simplified relationships is substantiated through a statistical analysis of the main error indices. Results highlight that the building's shape factor (i.e., the ratio between the building's envelope area and its volume) and the area-weighted average of the thermal properties of the building envelope significantly affect both the energy consumption of school buildings and the achievable savings through retrofitting interventions. Finally, a framework for the preliminary design of energy refurbishment of buildings, based on the implementation of the herein developed predictive model, is proposed and illustrated through a worked example application.Worth noting that, while the proposed approach is currently limited to school buildings, the methodology can conceptually be extended to any building typology, provided that suitable data on energy consumption are available.