Brazil's National Treasury (STN) is driving the compilation of government finance statistics. It is using artificial intelligence (AI) to classify subnational governments' expenditures according to the international standard COFOG (Classification of Government Functions). This initiative is part of a broader goal of using AI to strengthen transparency and improve efficiency in the management of public resources.
Historically, there was a manual process of converting data on public expenditure to the COFOG standard. This process was especially slow, labor-intensive, and prone to errors when it came to subnational government spending. The manual classification of approximately 100,000 budget records each year, following the COFOG methodology, required meticulous analysis of the databases, requiring significant time and human resources.
Artificial Intelligence (AI) entered the scene from 2021 when STN adopted a textual classification method for this process. A probabilistic text classification model based on machine learning techniques was specifically developed to categorize government expenditures according to COFOG codes. This revolutionary machine learning model utilizes Convolutional and Recurrent Neural Networks.
The results are impressive: the classification time for data from subnational governments was reduced from 1,000 hours to just 8 hours, an efficiency increase of 12,400%. The accuracy of the AI model exceeds 97%, ensuring that Brazil's fiscal data is not only timely but also reliable.
In 2024, the publication of the report Despesa por Função do Governo Geral (Expenditure by Function of the General Government), made possible through AI, became a milestone in fiscal statistics and, consequently, in fiscal transparency in Brazil. These results were widely recognized both nationally and internationally, culminating in invitations for the STN team to join global working groups and present the Brazilian experience in forums of great relevance.[1]
But the story doesn't end there. STN is expanding the use of AI to other critical areas, such as the classification of climate-related expenditures. In collaboration with the Inter-American Development Bank (IDB), Brazil is intensifying its efforts to understand and more effectively manage the fiscal implications of climate change.
This initiative positions Brazil at the forefront of a global movement to leverage AI in public financial management. As countries around the world face the complexities of fiscal management and environmental challenges, Brazil's experience stands as an example for other nations that want to modernize their public finance statistics systems and respond to the demands of a constantly changing world. The potential for AI to transform public finance management and fiscal transparency is immense, and Brazil's successful application of this technology demonstrates that innovation can lead to significant gains in efficiency and transparency.
[1] The methodology was presented at a meeting of the "Data Gaps Initiative", a G-20 statistical forum led by the IMF, which fosters the production and dissemination of data defined as strategic at the G-20 Ministers' Meeting. The project was presented at the event and considered a model of international best practices for data compilation. The presentation is available on the IMF website at: Global Conferences on DGI (imf.org), Global Conference in 2024, Presentations, IV Session, P_13_Session IV Challenges of tagging climate change related subsidies on existing databases_Brazil.pdf.
