Malha Fina de Convenios

The Brazilian federal voluntary transfers process handled more than R$100 billion between 2008 and 2018 by means of over 140,000 instruments among the entities of the Federation. However, the number of transfers made required an analysis effort much higher than the available analysis capacity of the transferring agencies. Thus, the problem of continued growth of presented accounts pending analysis emerged. The project "Malha Fina de Convenios" is a tool to solve this bottleneck.

Innovation Summary

Innovation Overview

The "Malha Fina de Convenios" was designed to solve the stock problem of presented accounts pending analysis in the process of federal voluntary transfers among subnational entities of the Brazilian federation. The innovation is the use of a machine learning algorithm based on the characteristics of the agreements whose accounts have already been analysed to allow a fast and efficient analysis of covenant accounts sent to the federal agencies in Brazil. Between September 2008 and December 2017, more than 61,000 agreements had their accounts analysed by the grantors, providing a satisfactory amount of data for the learning of the algorithm to provide accurate results.

The application of the algorithm results in the constitution of an individual note for each agreement, varying between 0 and 1. The closer to 0 the note is, the greater the chance that the agreement will have its accounts disapproved. Alternatively, the closer to 1, the greater the chances that the agreement will have its accounts rejected. Consequently, the rejection of the accounts of an agreement entitles the grantor to take the appropriate measures to recover the damage to the Treasury. The value calculated for each agreement is compared to the “cut-off value” established by the federal manager, which also varies from 0 to 1. Thus, all agreements whose score calculated by the algorithm was above this limit would be considered “objectionable”, requiring a conventional analysis.

Thus, for the operation of the algorithm, it is enough that the agency stipulates a minimum score before which all agreements classified below it are approved. As an example, if a specific agency stipulates a score of 0.8 as its threshold, it means that 79.4% of its agreements may be subject to tacit approval, of which 4.62% would be inadvertently approved. It should be noted that the decision on the passing score by the granting body reflects the risk appetite of the federal manager who is transferring the money to the subnational entities.

The life cycle of the transfer of discretionary resources ends with its rendering of accounts and consequent analysis by the transferring body, which opines for the approval or rejection of the accounts. Accountability analysis is a lengthy process and calls for the use of resources for its realisation, in addition to trained public servants. In turn, the “Malha Fina de Convênios” system presents a quick, rational and innovative alternative for the analysis of accountability.

Consequently, the validation of the automated accountability method is fundamental to the continuity of this innovative approach. It allows a disruptive away to analyse 15,300 accounts that represents a liability of almost R$ 17 billion (approximately U$ 3,95 billion). Hence, all the efforts and the bureaucratic body of the transferring agencies installed to analyse pending accounts can be rationalised.

The greatest inherent risk in the process of the “Malha fina de convênios” system is the inadvertent classification of covenants whose accounts were rejected with a score close to zero. Indeed, the machine learning is not infallible, sometimes assigning good scores to bad covenants. The error rate increases as the number of eligible convents submitted for an automatic analyses of accounts, based on the score given by the system, raises. Thus, the heart of the problem is to determine a passing score that will allow a great number of automatic analysis of covenants and a low error of misclassification of bad convents.

The determination of this threshold score considers the cost of the number of public servants and the time that would take to analyse all the accounts pending analyses, one by one, compared to the possibility of approving all of them by simply pushing one simple button. This is the risk appetite.

As one of the ways to add quality to the process - since the predictive model seeks to reproduce a manager action in the analysis of account presentations, with lower cost and time optimization - the internal audit activity performed by Controladoria-Geral da União adopts the concept of continuous audits. Hence, continuous audit was also aggregated to the system "Malha Fina de Convenios" using the “agreement audit trails” methodology, which also contributed to the mitigation of residual risks. The audit trails refer to a comparison of databases in the search for pre-defined patterns that point to signs of improprieties or irregularities. So, besides the score assigned by the machine learning algorithm, the federal manager can also use the alerts given by the audit trails.

In short, this is a disruptive solution to the huge problem of covenants liability in Brazil.

Innovation Description

Innovation Development

Innovation Reflections

Leave a Reply

Your email address will not be published. Required fields are marked *

Status:

  • Implementation - making the innovation happen
  • Diffusing Lessons - using what was learnt to inform other projects and understanding how the innovation can be applied in other ways

Innovation provided by:

Join our community:

It only takes a few minutes to complete the form and share your project.