A considerable portion of total energy loss within the built environment originates from operational errors during the actual lifespan of a building. With the rise of fully automated commercial buildings, a large amount of sensory data is becoming available that can be leveraged to detect and predict such errors. However, processing these data on-site requires significant knowledge and effort by building operators. In this work, a combination of model-based and data-driven approaches are employed to facilitate the analysis of historical energy demand data. Using change-point models and symbolic quantisation techniques, a large dataset of heating and cooling demand profiles collected from several office buildings are transformed into a format that is easily interpreted by the building operator and is suitable for actionable anomaly detection. Further quantification of anomalies and calculation of potential savings are drawn from the results.