Name in original language
Processo di gestione automatizzata delle richieste web
Initiative overview
The initiative addresses the inefficiency of the previous manual process used to route citizens’ requests submitted through the INPS Risponde web channel. In the past, all messages were reviewed by contact centre operators and forwarded to back-office staff where necessary, meaning that most requests were handled twice and resolved with delays.
The initiative leverages the opportunity offered by artificial intelligence to automate this routing process and improve service delivery. The main objective is to reduce processing steps and response times by automatically directing each request to the appropriate operational level, either the Contact Centre or back-office units. Additional objectives include improving the accuracy of request classification, reducing unnecessary work for operators, and ensuring compliance with data-protection rules through an internally managed system.
The initiative is expected to evolve through its integration into ordinary operational processes and its progressive extension to additional request categories. Over time, the initiative has evolved from an experimental automation tool into a structured system based on supervised artificial intelligence models. The main technology used is a BERT Transformer.
The models are retrained weekly through a defined workflow that includes data collection, preprocessing, training, testing, benchmarking against the production model, and feedback generation. At the end of each training cycle, a report is produced highlighting any components that require manual updates.
The system is trained exclusively on internal INPS data, organised by thematic area, and operates in compliance with GDPR requirements. The entire solution is built on open-source technologies and hosted on the organisation’s own infrastructure, ensuring full control over both data and operational processes.
Results, outcomes and impacts
The initiative has led to faster and more accurate handling of citizens’ requests and a reduction in unnecessary work for first-level staff. The system automatically analyses all incoming messages and routes them to the appropriate level, so first-level operators handle only requests appropriate to their level.
Results have been measured through system logs and operational statistics. Since 22 April 2024, more than two million requests have been processed by AI, and approximately 30% have been automatically routed directly to second-level staff.
In the future, further improvements are expected in response times, automation rates and resource efficiency as the system is extended to additional request types and services.
