Spain’s Gipuzkoa provincial council faced a problem common to many local governments around the world: a flood of data and too few staff to handle it.
Every year, the City Council register receives over 1.6 million different citizen requests, intergovernmental communications and technical documents.
Of these, there were 100,000 entries that required manual classification, which was the equivalent of 8,300 hours of work or one year of full-time work from five people.
Still, classification was inconsistent, with errors and reassignments occurring in approximately 8% of cases.
The Gipuzkoa City Council decided to do something about this.
The council, in collaboration with IZFE, the public IT services company of the Gipuzkoa administration, has developed an artificial intelligence (AI) system called ERREKA (ERRegistro apunteen SailsKatzaile Automatikoa) that automates the classification of registry entries received by the administration.
This system helped speed up the processing of these entries and improve public services.
Since ERREKA’s debut in October 2025, its machine learning algorithms and natural language processing have enabled the government to achieve 93.4% overall accuracy in classifying incoming documents.
80% automation
According to the council and IZFE, nearly 80 percent of entries were automatically categorized by AI, significantly reducing the workload required for human review.
ERREKA automatically categorizes entries into three “confidence levels”: high, medium, and low.
Highly trusted and classified documents were automatically routed to the appropriate department.
Documents classified with medium confidence were sent to a human operator for one-click approval.
For unreliable documents, we followed our normal manual process.
The ERREKA model was trained on 38,037 pre-classified real notes and used an optical character recognition system developed by IZFE. It can handle highly heterogeneous documents in multiple languages.
Before going live, the system underwent a seven-month pilot run in which more than 19,000 entries were processed in parallel with real-world workflows, without interruption to daily operations.
The pilot showed an accuracy level of over 92% in various government sectors such as human resources, road infrastructure, mobility, and tourism.
In this phase, we verified the stability and speed of the AI system under real-world conditions.
Deploy in stages
ERREKA is being phased in in different sections of the state legislature.
The state government stressed that AI models will not replace humans and that primary control and decision-making authority will remain with officials.
This model was intended to reduce repetitive manual tasks and give employees more time to process cases and serve the public.
The Council noted that this project is an important step towards intelligent automation in Basque public administration.
It shows that AI can be safely integrated into government, freeing up time to improve public services.
The team that developed ERREKA is currently working on a semi-automatic retraining system that will allow for continuous improvement of the model and the incorporation of new management units.
