The aim of this paper is to provide an extended analysis of the outlier detection, using probabilistic and AI techniques, applied in a demo pilot demand response in blocks of buildings project, based on real experiments and energy data collection with detected anomalies. A numerical algorithm was created to differentiate between natural energy peaks and outliers, so as to first apply a data cleaning. Then, a calculation of the impact in the energy baseline for the demand response computation was implemented, with improved precision, as related to other referenced methods and to the original data processing. For the demo pilot project implemented in the Technical University of Cluj-Napoca block of buildings, without the energy baseline data cleaning, in some cases it was impossible to compute the established key performance indicators (peak power reduction, energy savings, cost savings, CO2 emissions reduction) or the resulted values were far much higher (>50%) and not realistic. Therefore, in real case business models, it is crucial to use outlier’s removal. In the past years, both companies and academic communities pulled their efforts in generating input that consist in new abstractions, interfaces, approaches for scalability, and crowdsourcing techniques. Quantitative and qualitative methods were created with the scope of error reduction and were covered in multiple surveys and overviews to cope with outlier detection.