Strengthening Statistical and Artificial Intelligence Capabilities for Costa Rica's 911 Service: Application of the “Loan Shark Schemes” Statistical Process for 911 Incidents

Costa Rica's 9-1-1 Emergency System, in collaboration with the Center of Excellence (CoE), continues to promote an innovative initiative in Latin America in the production of official statistics, based on the enormous potential of the data generated daily from emergency calls, using statistical techniques and implementing Artificial Intelligence models.

From February 2 to 6, the Center of Excellence provided training in statistics and data analysis to 9-1-1 personnel to strengthen technical capabilities through the application of a standardized statistical process, aligned with the Generic Statistical Business Process Model (GSBPM), and the use of tools such as Python and natural language processing (NLP) techniques.

The 9-1-1 Emergency System database is one of the most comprehensive and continuous sources of information on incidents reported nationwide. It records the total number of calls handled daily and classifies events according to the institution responsible for responding—including law enforcement, firefighters, the Red Cross, INAMU, PANI, OIJ, and other sectoral entities—which together apply 163 different categories to identify the characteristics of incidents.

In 2025, 9-1-1, with technical support from the CdE, compiled an initial set of data on incidents related to payday loans. Based on the narratives of calls recorded between 2015 and 2025, statistical and advanced analytical techniques were applied to identify temporal and geographic patterns and risk typologies linked to this phenomenon.

This initial result demonstrated the potential of 9-1-1 records to generate empirical evidence on economic activities related to emerging illicit practices, strengthen public safety strategies, guide protection and mental health services, and guide inter-agency action.

Based on this experience, the CdE designed a training program structured around five technical sessions, which guides participants through a complete statistical production cycle. From defining needs and cleaning data to interactive visualization and text analysis with NLP models, the training combines methodological rigor with a practical and replicable approach.

One of the most innovative elements is the incorporation of natural language processing models to analyze unstructured information. Tools such as spaCy and BERTopic allow entities to be extracted, narrative patterns to be identified, and recurring themes in calls to be discovered, broadening understanding of the phenomenon beyond traditional counts.

Upon completion of the training, 9-1-1 personnel have access to a scalable standardized database, a set of statistical indicators, geospatial visualizations, and a reproducible Python notebook, laying the foundation for an institutional statistical process and data analysis.