The Italian National Statistical Institute is collaborating with the Department of Public Administration (DFP) on a National Recovery and Resilience Plan (NRRP) project aimed at assessing the impact of reforms on Italian public administrations (PAs), on administrative procedures and on staff-related activities. To achieve this goal, the project plan combines several data sources: census and survey data, administrative records, and unstructured data extracted from PA PDF documents, the Integrated Plan of Activities and Organisation, called PIAOs. The paper describes an innovative use of PIAOs, analyzed through advanced automatic classification processes, powered by Large Language Models (LLMs) and informed by the expertise of public administration specialists. Initial results show the high accuracy of the methodology, suggesting the use of this source to produce estimates. Future statistical developments aim to make inference to all Italian public administrations and to produce comprehensive estimates of the impact of reforms on administrative improvements and staff-related measures. The estimation procedure will take into account the selectivity of the available PIAOs and the prediction errors of the automatic classification.