Gipuzkoa Provincial Council Puts ERREKA AI Assistant to Work Classifying Registry Entries

Since October 2025, ERREKA has classified incoming registry entries at the Provincial Council of Gipuzkoa (Diputación Foral de Gipuzkoa) with an overall accuracy of 93.4%. Nearly 80% of entries are now classified fully automatically, sharply reducing the volume left for human review. The system was developed by Izfe, the Council's IT services company, and its name stands for ERRegistro apunteen sailKatzaile Automatikoa. The Council's Registry receives more than 1.6 million entries a year from citizen requests, communications between administrations and technical documentation. More than 100,000 of them had to be classified manually, a repetitive task equivalent to 8,300 working hours a year, or about five people working full time. Manual classification was also not always consistent and led to reassignments in around 8% of cases. ERREKA is based on machine learning and natural language processing. It works in tiers. When its confidence is high, it classifies the entry automatically. When confidence is medium, it proposes a classification that staff validate with one click. When confidence is low, the entry follows the usual manual route. Izfe stresses that the system does not replace people: final control stays human. Development followed the CRISP-DM methodology. The model was trained on 38,037 real entries that had already been classified, and relies on zOCR, Izfe's optical character recognition system, which can process very heterogeneous documents in several languages. Before going live, ERREKA ran for seven months as a silent pilot, processing more than 19,000 entries in parallel with the real workflow without interfering with daily operations. Accuracy exceeded 92% in administrative units such as Personnel Management, Road Infrastructure, Mobility and Tourism, and Green Territorial Balance. ERREKA has been rolled out progressively across several departments, easing repetitive work for the Generic Units that classify registry entries. Izfe is now working on semi-automatic retraining so the model can improve continuously and take in new administrative units, and is exploring generative AI models for areas with less historical data.