COAF predictive model speeds triage of Brazil's suspicious reports

FATF and GAFILAT assessors found that COAF's AI predictive model allows rapid analysis of the 4.8 million suspicious transaction reports received in 2018–2022, but that only 5% go to analysts.

  • 1 for an organization

    The assessors found that the model allows rapid analysis of an average of 971 383 reports a year but gave no measure of the time saved.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Partly reported benefit.
Evidence

foundstudiedreportedclaimed

An independent assessment report supports the benefit.

The system’s part
Not isolated. The assessors relay that the model rests on artificial intelligence, data mining and automated processes and that stored reports continuously inform it. They did not test its accuracy or compare triage with and without it. They linked the low share of medium- and high-risk scores to report quality or the calibration of the risk matrix and tested neither.
Not yet checked
LimitsCosts and harmsLearns from dataLater outcome

Download data Cite Suggest a correction

Benefit type
Repetitive Task Automation
System
COAF predictive model for scoring suspicious transaction reports
Made by
Conselho de Controle de Atividades Financeiras (COAF) (the report names no other developer)
Used by
Conselho de Controle de Atividades Financeiras (COAF), Brazil's financial intelligence unit
Where
Brazil (national)
Stage
In routine use
Date
, when the source was published
Kind
Aggregate result
Broad benefit
Analyzing Data or Information, Organizing, Planning, and Prioritizing Work
Collections
Scrutiny and findings
Last revised
1 November 2026

Sources

  1. Assessment report, by authors with no tie to the system.

    FATF and GAFILAT. Anti-money laundering and counter-terrorist financing measures – Brazil. Fourth Round Mutual Evaluation Report, December 2023, adopted at the FATF Plenary of October 2023. Chapter 3, Immediate Outcome 6, paras 180-184 and 196-198.

    while the predictive model is an important IT tool that allows for a rapid analysis of the STRs, only the 5% of the STRs are sent for individual analysis by analysts

Details

AI system

Workflow

Reporting entities filed suspicious transaction reports through SISCOAF, which loaded them straight into COAF's systems. The model scored each report after crossing it with COAF's databases, including its cash transaction reports. Reports with a 30% likelihood of becoming a financial intelligence report went to individual analysis, with an automatic initial report.

Follow-up

Conditions for replication

Reporting entities must file reports of good quality, and the assessors found that report quality was an issue. COAF needs enough analysts to examine the reports that the model selects, and the assessors found that the unit was understaffed.

Entry details

Sector

Anti-money laundering: financial intelligence.

Purpose

COAF built its predictive model to score every incoming suspicious transaction report and send the riskiest to analysts for individual analysis.

31 more questions 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). COAF predictive model speeds triage of Brazil's suspicious reports (AIBD-1038). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1038.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)

BibTeX Data Field definitions