The innovation involves the automatic classification and sorting of Certified Electronic Mail (PEC) within the National Institute of Social Security (INPS) in Italy.
Innovation Summary
Innovation Overview
PEC is a crucial channel through which citizens communicate with INPS, and its usage has notably increased in recent years. The manual process of receiving PEC, reading its content and attachments, understanding the subject matter, and then assigning it to the relevant officer has become highly repetitive and resource-intensive. With a workload increase of over 50% in a few years, reaching over 6 million incoming PECs annually, there was a need for an efficient solution.
To address this challenge, INPS explored the application of modern Artificial Intelligence (AI) technologies. The organization experimented with early versions of open-source models like GPT-2 and later transitioned to BERT, a machine learning model developed by Google. These models, based on Transformers architecture, were fine-tuned using PEC data and feedback from operators. Notably, the entire system, including the AI models and processing, was developed and executed internally in the INPS Data Centers to ensure compliance with GDPR and data privacy regulations.
Innovation Description
What Makes Your Project Innovative?
Building on the experience gained from the Automated Classification and Sorting of Certified Electronic Mail (PEC) project, INPS explored the application of Artificial Intelligence (AI) technologies to optimize the overall process of handling web requests. An ongoing experimental project involves an AI decision-making engine that evaluates the complexity of citizen requests. The AI system aims to directly and accurately route 34% of requests to the appropriate second-level operator without the intervention of the first-level operator. This approach streamlines the processing of 850,000 requests, alleviating the workload of first-level operators and ensuring a more immediate resolution of citizen requests.
Innovation Development
Collaborations & Partnerships
The project was developed and implemented internally in INPS by the team led by Vincenzo Di Nicola, the Responsible for Technological Innovation and Digital Transformation.
Users, Stakeholders & Beneficiaries
The target audience for this innovation includes citizens utilising the web service INPS Risponde to communicate with INPS. The focus on optimising the handling of citizen requests reflects a people-centric approach. The utilisation of AI is aimed at improving citizen experience by expediting request processing. Further, to ensure safety, the entire AI system for web request management is developed and executed internally within INPS Data Centres.
Innovation Reflections
Results, Outcomes & Impacts
The AI-driven system has significantly improved the processing of citizen-INPS communications. It is active in 10 Italian cities, including major ones like Rome, Milan, and Naples, processing over 1 million PECs to date. The innovation is estimated to save between 30,000 and 40,000 days of work annually when considering all territorial offices. Operators, previously engaged in manual sorting, can now focus on more value-added activities, benefiting citizens. The project has also created internal technological know-how that can be leveraged to enhance other services.
Replication
This innovative approach has already received recognition from Osservatori Digital Innovation of the Politecnico di Milano, earning accolades in the "Implementation of the Digital Agenda" category at the 2022 Digital Agenda Awards. The ongoing experimentation and the recognition received from the Digital Innovation Observatory indicate the sustainability and potential replicability of the innovation in optimising web request management for citizens.
Status:
Date Published:
5 November 2024