Advanced energy benchmarking in residential buildings, using data-driven modeling, provides a fast, accurate, and systematic approach to assessing energy performance and comparing it with reference standards or targets. This process is essential for identifying opportunities to improve energy efficiency and for shaping effective energy retrofit strategies. However, building professionals often face barriers to adopting these tools, mainly due to the complexity and limited interpretability of data-driven models, which can negatively affect decision-making. In order to contribute in addressing these issues, this study combines data-driven modeling with Explainable Artificial Intelligence (XAI) techniques to advance energy benchmarking analysis in residential buildings and enhance their usability by also non-expert users. The proposed process focuses on estimating primary energy demand for space heating and domestic hot water in residential building units, extracting knowledge from about 49,000 Energy Performance Certificates (EPCs) issued in the Piedmont Region, Italy. The effectiveness of five machine learning algorithms is assessed to select the most suitable estimation model. Then to ensure the trustworthiness of the selected model, a XAI layer is implemented to identify and remove input variable domain regions that demonstrated to be critical for the robustness of the inference mechanism learnt in the training phase. Moreover, the study assesses the model capability to evaluate building energy performance, examining both the current state and potential scenarios for energy retrofitting. A second XAI layer is then introduced to provide local explanations for model estimations of both pre- and post-retrofit conditions of a building. The final aim is to enable an external benchmarking analysis, by extracting from the analysed EPCs reference groups of similar buildings, that facilitate a performance comparison for the investigated retrofit scenarios. This energy benchmarking process promotes transparent and informed decision-making, aiming to instill confidence in final users when leveraging data-driven models for energy planning in the building sector.