The Provincial Council of Gipuzkoa in Spain has implemented an artificial intelligence system named ERREKA to automate the classification of over 1.6 million annual citizen requests, technical documentation, and inter-administrative communications. Previously, the council faced significant administrative burdens, with approximately 100,000 entries requiring manual sorting. This legacy process demanded 8,300 working hours annually, equivalent to five full-time employees, and suffered from an 8 per cent error rate due to inconsistent classification.

Developed in partnership with the public IT services provider IZFE, ERREKA debuted in October 2025. The system utilizes machine learning and natural language processing to achieve a 93.4 per cent classification accuracy. Since its introduction, nearly 80 per cent of incoming registry entries have been classified automatically. The model employs three confidence levels to manage documentation: high-confidence entries are routed automatically, medium-confidence items go to human operators for single-click approval, and low-confidence documents follow traditional manual review workflows.

The model was trained using 38,037 pre-classified notes and incorporates an optical character recognition system, allowing it to process heterogeneous documentation in multiple languages. Prior to the full rollout, the administration conducted a seven-month pilot program involving 19,000 entries, where the system maintained an accuracy level above 92 per cent across diverse units including human resources, road infrastructure, mobility, and tourism. The administration emphasizes that ERREKA is intended to reduce repetitive manual tasks rather than replace human workers, ensuring that final decision-making power remains with the staff. Development teams are currently working on semi-automatic retraining systems to facilitate continuous model improvement and the integration of additional administrative units.

The Provincial Council of Gipuzkoa in Spain has implemented an artificial intelligence system named ERREKA to automate the classification of over 1.6 million annual citizen requests, technical documentation, and inter-administrative communications. Previously, the council faced significant administrative burdens, with approximately 100,000 entries requiring manual sorting. This legacy process demanded 8,300 working hours annually, equivalent to five full-time employees, and suffered from an 8 per cent error rate due to inconsistent classification.

Developed in partnership with the public IT services provider IZFE, ERREKA debuted in October 2025. The system utilizes machine learning and natural language processing to achieve a 93.4 per cent classification accuracy. Since its introduction, nearly 80 per cent of incoming registry entries have been classified automatically. The model employs three confidence levels to manage documentation: high-confidence entries are routed automatically, medium-confidence items go to human operators for single-click approval, and low-confidence documents follow traditional manual review workflows.

The model was trained using 38,037 pre-classified notes and incorporates an optical character recognition system, allowing it to process heterogeneous documentation in multiple languages. Prior to the full rollout, the administration conducted a seven-month pilot program involving 19,000 entries, where the system maintained an accuracy level above 92 per cent across diverse units including human resources, road infrastructure, mobility, and tourism. The administration emphasizes that ERREKA is intended to reduce repetitive manual tasks rather than replace human workers, ensuring that final decision-making power remains with the staff. Development teams are currently working on semi-automatic retraining systems to facilitate continuous model improvement and the integration of additional administrative units.