What happened
Brazil's education fund, the FNDE, receives about 17,000 accountability reports a year on transfers under three school programs. Its staff cannot analyze them all. In 2022 it first applied Malha Fina FNDE to the reports for 2011 to 2018. The triage combines value limits and rule-based audit trails with a predictive model. That model uses supervised machine learning to pick out reports that manual review would very likely approve. The first round approved 60,471 reports, about 20% of a backlog of some 300,000. In a ruling of 2 April 2025, the Federal Court of Accounts (TCU) called the model a good practice. From the FNDE's records of March 2023, its auditors counted 11,931 school-meal reports approved, 27% of the 44,386 filed. Staff had concluded the analysis of 1,369. The approved reports held only 3.6% of the money transferred. Of the school-meal reports, 66% still awaited analysis. The TCU concluded that the backlog problem was not resolved as hoped.
Sources
-
Audit report, by authors with no tie to the system.
Tribunal de Contas da União. Acórdão 743/2025 – TCU – Plenário (TC 020.760/2022-1), relator Ministro Walton Alencar Rodrigues, session of 2 April 2025. Operational audit of the FNDE's internal controls on the financial management of the Programa Nacional de Alimentação Escolar (PNAE), Relatório paragraphs 128-140 and Tabela 1.
Proporcionalmente à quantidade total de prestações de contas homologas (60.471), o PNAE foi contemplado com 11.931 contas homologadas, o que representa 27% do total de prestações de contas desse programa que deram entrada no FNDE, relativas aos repasses realizados no período 2011 a 2018
Details
AI system
- Workflow
-
An ABC value curve set minimum and maximum values of R$ 80,000 and R$ 1,000,000 for the analysis. Audit trails ran objective tests for signs of irregularity. The model picked out reports like those that manual review had approved.
Follow-up
- Conditions for replication
-
Reports that the triage does not approve still need manual analysis before claims become time-barred. The FNDE must review the model's criteria and its risk tolerance, set at seven tenths or more on a scale of 0 to 1.
Entry details
- Sector
-
Public administration: education grant accountability review.
- Purpose
-
The FNDE built Malha Fina FNDE to approve low-risk accountability reports automatically and leave the rest for its staff to analyze.
Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). Malha Fina clears 27% of school-meal accounts at Brazil's FNDE (AIBD-0918). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0918.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)