Improving the quality of citizen service through generative AI: Automated call auditing on line 148 of the Province of Buenos Aires

Abstract

In public citizen services, telephone helplines are a fundamental channel to guarantee rights, inform about public policies, disseminate campaigns of public interest, and support a comprehensive approach to social issues. In particular, the 148 helpline of the Province of Buenos Aires currently centralizes, at the time of writing this study, nineteen telephone assistance services from various state agencies, including areas such as health, labor, social security, justice, social development, sexual and reproductive health, consumer protection, among others. Given the high volume of calls and considering the estimated growth in call volume for the year 2025 for this helpline, manual supervision, monitoring, and auditing processes are costly in terms of time and human resources, difficult to scale, and challenging in terms of traceability. This limits the capacity to detect irregularities, improve service quality, and ensure staff compliance with established protocols. This paper presents a public innovation initiative based on generative artificial intelligence, developed by the Directorate of Digitalization and Artificial Intelligence, Provincial Directorate of Digital Innovation, Undersecretariat of Digital Government, Ministry of Government, aimed at automating the call auditing process carried out by the Directorate of Monitoring and Evaluation, Provincial Directorate of Online Government, Undersecretariat of Digital Government, Ministry of Government. The solution combines open-source technologies and processing on inhouse infrastructure to ensure technological sovereignty and the protection of sensitive data. It consists of two main stages for audit automation: automatic transcription of calls using speech recognition models, and content classification/auditing through large language models (LLMs) trained using few-shot prompting strategies and finetuning with specific instructions and templates. Additionally, the initiative works on the automated generation of databases, reports, and key indicators through interactive dashboards. This tool enables the scaling of monitoring processes, reduces time and costs, identifies critical patterns (such as inappropriate language or aggressive behavior), and generates evidence for the continuous improvement of service quality and public policy design. It not only enhances the quality of citizen care but also introduces a paradigm shift in the monitoring and decision-making processes of public services, providing daily processed data for more efficient, preventive, and evidence-based management. The initiative aims to be replicable in other citizen service areas within the public sector.