[en] Increasing trust in energy performance certificates (EPCs) and drawing meaningful conclusions requires a robust and accurate determination of building energy performance (BEP). However, existing and by law prescribed engineering methods, relying on physical principles, are under debate for being error-prone in practice and ultimately inaccurate. Research has heralded data-driven methods, mostly machine learning algorithms, to be promising alternatives: various studies compare engineering and data-driven methods with a clear advantage for data-driven methods in terms of prediction accuracy for BEP. While previous studies only investigated the prediction accuracy for BEP, it yet remains unclear which reasons and cause–effect relationships lead to the surplus prediction accuracy of data-driven methods. In this study, we develop and discuss a theory on how data collection, the type of auditor, the energy quantification method, and its accuracy relate to one another. First, we introduce cause–effect relationships for quantifying BEP method-agnostically and investigate the influence of several design parameters, such as the expertise of the auditor issuing the EPC, to develop our theory. Second, we evaluate and discuss our theory with literature. We find that data-driven methods positively influence cause–effect relationships, compensating for deficits due to auditors’ lack of expertise, leading to high prediction accuracy. We provide recommendations for future research and practice to enable the informed use of data-driven methods.