This work presents a structured methodology to improve Urban Building Energy Modelling (UBEM) through customized, data-driven building archetypes for the Italian context. The approach addresses a key challenge in Italy: the lack of integrated and standardized urban-scale data. The proposed workflow involves five steps: defining baseline archetypes using various datasets, deducing heating, cooling, and DHW systems characteristics from electric smart meter data, modeling occupant behavior (OB) through the same electric data via clustering, customizing archetypes, and scaling them to a larger building stock using classification algorithms. Validation was performed on 15 buildings with smart meter data and extended to 855 public housing buildings in Milan. The methodology led to significant improvements in simulation accuracy, with the electricity NMBE reduced from –150% to around ±10%, and similarly improved gas estimates. Results confirmed that data-informed archetypes offer better performance than default assumptions, especially when OB and HVAC systems are realistically modeled. The study concludes that the workflow is transferable to other urban contexts and supports the development of more reliable energy models for sustainable planning and policy-making.