Spanish local body turns to AI to automate 1.6 million annual citizen requests

The Provincial Council of Gipuzkoa in Spain faced a problem that was common to many local governments around the world - a deluge of data with too few workers to process it. 

Every year, the council registry receives more than 1.6 million different citizen requests, inter-administrative communications, and technical documentation. 

Out of these, as many as 100,000 entries required manual classification, a task that was equivalent to 8,300 working hours or the full-time work of five people for a year.  

Even then, the classification lacked consistency, leading to errors and reassignments in approximately eight per cent of the cases. 

The Gipuzkoa council decided to do something about this.  

Working with IZFE, a publicly-owned IT services company for the Gipuzkoa public administration, the council developed an artificial intelligence (AI) system named ERREKA (ERRegistro apunteen sailKatzaile Automatikoa), which automated the classification of registry entries received by the administration. 

The system helped to accelerate processing of these entries, thus improving public service. 

Since ERREKA’s debut in October 2025, its machine learning algorithms and natural language processing helped the government to achieve an overall accuracy of 93.4 per cent in classifying incoming documentation. 

80 per cent automation 

According to the council and IZFE, nearly 80 per cent of entries were classified automatically by the AI, thus drastically reducing the workload that required human review. 

ERREKA automatically categorises the entries into three “confidence levels” of high, medium, and low. 

Documents that were classified with high confidence were automatically routed to the correct departments.  

Documents classified with medium confidence were sent to a human operator for a single-click approval.  

The usual manual process was followed for low confidence documents. 

The ERREKA model was trained on 38,037 real, pre-classified notes and it used an optical character recognition system developed by IZFE, capable of processing highly heterogeneous documentation in multiple languages. 

Before going live, the system underwent a seven-month pilot, where more than 19,000 entries were processed in parallel with the actual workflow without disrupting daily operations.  

The pilot showed an accuracy level of above 92 per cent in diverse administrative units such as human resources, road infrastructure, mobility and tourism, among others. 

This phase validated the AI system’s stability and speed under real conditions. 

Progressively deployed

ERREKA has been progressively deployed in various departments of the Provincial Council.  

The provincial administration emphasised that the AI model was not meant to replace humans and key control and decision-making power would rest on the staff. 

The model aimed to reduce repetitive manual tasks so that employees could have more time to process cases and serve the public. 

The council noted that the project represented a significant step towards intelligent automation in the Basque public administration. 

It showed that AI could integrate safely into public administration thus freeing up time for improving public service. 

Currently, the team which developed ERREKA has been working on semi-automatic retraining systems that would allow for continuous improvement of the model and incorporating new administrative units.