The initiative consists in the development of an AI-based chatbot capable of analysing aggregated statistical data. The first chatbot prototype was based on a monolithic assistant, but a multimodal agent is being designed with the support of our technical partner, Società Generale d’Informatica (SOGEI). The initiative develops an AI-based chatbot designed to analyse aggregates fiscal data and statistical data and support advanced analytical tasks. It addresses the growing volume and complexity
Initiative overview
The initiative addresses the growing volume and complexity of fiscal and statistical data, which increasingly challenge traditional analytical tools and constrain the capacity of public administrations to fully leverage available information for decision-making. At the same time, it capitalises on the opportunity offered by high-quality, structured public data and advances in Artificial Intelligence (AI) to modernise analytical processes, strengthen evidence-based policymaking, and improve transparency in the management of public information.
The initiative pursues three main objectives:
- to enhance the accuracy, speed, and efficiency of fiscal data analysis through AI-based tools;
- to support data-driven decision-making within public administration;
- to promote the responsible and trustworthy use of AI, in line with principles of data protection, accountability, and human oversight.
To achieve these objectives, a prototype AI-based chatbot was developed using aggregated statistical data published by the Department of Finance. The tool supports statistical analysis and facilitates interaction with complex datasets. The initial prototype was based on monolithic architecture and limited data domain.
The initiative is now evolving towards a more advanced multimodal AI agent with enhanced reasoning capabilities and dynamic data-loading functions. In the medium term, the project aims to transition from a single-assistant model to an integrated ecosystem of specialised AI agents embedded within existing analytical workflows. Institutionalization will involve the adoption of governance frameworks ensuring compliance with ethical, legal, and security standards, as well as capacity-building for public officials. The model is designed to be scalable and transferable to other policy domains and public administrations, enabling broader adoption of AI solutions to strengthen public sector performance and trust.
Other relevant details
Results, outcomes and impacts: So far, the initiative has delivered a functional prototype, which was presented at the Italian Public Administration Conference in 2025, increasing visibility and awareness of the solution. Initially designed to support analysis of tax return information, the prototype has demonstrated broader applicability when shared within the Directorate. It has proven useful for additional analytical use cases, such as territorial and municipal analysis, by integrating tax return data with other datasets at municipal level. As adoption expands, we expect new applications to emerge, supporting wider use of AI based analysis and improving analytical capacity across multiple policy areas.
Challenges and lessons learned: The main challenges encountered concerned data integration, ensuring the reliability of outputs, and managing expectations related to an early stage prototype. Initial risks included limited harmonisation across datasets and the need to ensure appropriate governance and responsible use of AI. These challenges were addressed through an iterative development process involving close collaboration between the Directorate and the IT team. A key lesson learned is the value of designing flexible and reusable tools, as the chatbot rapidly proved useful beyond its original scope. The initiative can be successful if the tool consistently delivers high quality, accurate analyses and visualisations, is user friendly for a broad range of civil servants, remains flexible, and is actively promoted so that other units can identify new ways to improve their work using it.
