Decarbonizing the building sector is crucial for a carbon-free society, as buildings are responsible for 30 % of global greenhouse gas emissions. It has been recognized that occupants’ decision-making in buildings has a substantial impact on carbon emissions. Previous studies have underlined the potential of behaviour-change intervention strategies, such as smart thermostat functions and eco-feedback through interfaces, in helping occupants make environmentally conscious decisions while meeting their needs. However, developing an effective intervention strategy for optimal behaviour change is challenging due to the difficulty of discovering the direct causal impact of each factor on behaviour change. There are two reasons that make the quantification of causal impacts burdensome: (i) the challenges of conducting controlled experiments in real buildings and (ii) the limitations of conventional statistical methods in estimating causal effects. To address these difficulties, causal inference approaches have been proposed to enable the investigation of causal structures from observational data. This study aims to develop a causal discovery algorithm and demonstrate the reliability of a causal model with a toy dataset by assuming a hypothetical situation. Firstly, causal relationships between factors observed in the toy dataset will be identified with a constraint-based Bayesian causal discovery approach. Secondly, a causal model will be developed based on the discovered causal structure with Bayesian modelling. The robustness of causal models will be assessed using in-distribution and Out-of-distribution datasets and compared with an association-based prediction model that lacks causal structures. This study will enable the use of causal estimates for reliable building energy solution development and provide an essential causal inference methodology to achieve a carbon-neutral society.