Achieving energy savings in buildings is vital to restrain global warming and reach the targeted reduction of CO2-emissions. Large-scale studies on the gap between the real and regulatory energy use in buildings have shown that the regulatory calculation overestimated the real energy use, inflated true energy savings and undermined national energy policy making. Data-driven stochastic grey-box models, characterised by their simplicity and reliance on lumped physical parameters, have demonstrated superior predictive accuracy. They prove to be among the most promising approach to increase the reliability and reduce the discrepancy between bottom-up physics-based models and real energy use. Yet, so far, grey-box modelling has focused only on parameter identification, building performance characterisation and prediction of real energy use, using various modelling approaches with historic, often artificial, data. This research aims to leverage the capabilities of stochastic grey-box models outside academia. The candidate believes that the lumped physical variables, should be trained on real hourly data, mapped with variables from existing registries and broken down into subcomponents to allow for accurate predictions of energy-saving measures. By transferring the capabilities of scenario analysis - a unique feature of white-box models - to grey-box models, the short-fall in terms of energy-savings can be reduced and the true impact of energy-saving strategies predicted.