AI in the public sector: the challenge lies in transforming the administration

A local authority chatbot can explain how to apply for a grant. If it gets a requirement wrong, however, it may persuade a member of the public to forgo a right. It is at this juncture – the shift from a plausible response to one that a person can rely on – that the complexity of artificial intelligence in public administration begins. Speed matters. So do accuracy, the ability to challenge a decision, and the presence of someone capable of rectifying it. The cost of an error may fall on those least equipped to recognise it. And this is serious because, unlike the private sector which follows market forces, the public administration is bound by the ‘commandment’ not to leave anyone behind.

In Italia, pilot schemes are underway. The latest AgID report has compiled responses from 108 organisations, including central government bodies and national public service providers. Of these, 45 had launched AI initiatives, totalling 120 projects surveyed. 78 per cent of the projects lacked indicators to measure their impact. This makes it difficult to answer the key question: is the service improving?

Artificial intelligence in Italian local authorities is still, for the most part, a prospect for the future. According to the ‘MDigital Municipalities 2026 App”, promoted by the Department for Digital Transformation, ANCI and the Bank of Italia, based on a survey conducted in 2025 involving 4,177 local authorities, the solutions already in operation or planned within twelve months cover less than 10 per cent of local authorities, both in terms of internal administrative activities and services for citizens and businesses. Around a quarter are considering introducing them. The disparity in scale is clear: considering active or planned applications, the main factors holding back adoption are the lack of in-house technical skills, reported by 67 per cent of local authorities, the difficulty in securing funding (52 per cent), regulatory uncertainty (43 per cent) and integration with existing processes (41 per cent). The report therefore highlights a problem of administrative capacity: bringing AI into services requires trained staff, funding and organisation, particularly in smaller local authorities.

During the meeting on 3 November at the SmartCityLab in Milan, organised by POLI.design. Guido Arnone, the City Council’s Chief Technology Officer, outlined three areas of focus: simplifying access to services, reducing repetitive administrative tasks, and making better use of data to manage the local area, including through predictive maintenance. These are areas with different requirements. Summarising a file, predicting a system failure and responding to a family regarding the eligibility criteria for a benefit require different checks and forms of verification. But above all, they require qualified staff.

This is where AI4Gov-X comes in. Coordinated by POLI.design and co-funded by the European Digital Europe programme, it brings together over 40 partners across 20 countries for a 48-month programme. It builds on the experience of AI4Gov, which originated from a European master’s programme on artificial intelligence for public services, and combines advanced training, short courses and experimentation. The aim is to equip civil servants, professionals and decision-makers to integrate technology, law, data management and service design. The European dimension allows for the comparison of recurring issues across different administrative systems. Perhaps this is where it can begin.