One of the first steps in reducing energy demand is accurately predicting energy requirements in urban areas and improving the thermal performance of building envelopes, especially in the existing residential sector. As an indicator of the as-built thermal transmittance and airtightness of the building fabric, the heat loss coefficient (HLC) can be assessed using data collected in operation coupled with data-driven statistical models. Utilizing non-intrusive monitoring technologies, such as smart meters and online weather data can provide the opportunity to develop assessment methodologies applicable to a large number of dwellings without requiring expensive equipment setups and on-site inspections. Several authors recognized the need to characterize HLC using limited existing measurements, but access to high-quality data for testing the reliability and accuracy of such methods is often limited. Therefore, this study aims to demonstrate the applicability of dynamic black-box statistical modelling when considering a sample of real-life inhabited dwellings. The key findings of this research revealed challenges in selecting appropriate monitoring periods due to constraints in favourable weather conditions and gaps in sensor data. Moreover, the dynamic of the building as well as the presence of occupants, which had a significant impact on internal gains and losses, added complexity in selecting the monitoring frequency and quantifying the HLC from in-use data.