Um sistema baseado em conhecimento para o planejamento de auditorias

Abstract

The annual records of public spending, associated with the need to ensure the efficiency, legality and economy in the application of public resources, have made most of the control institutions face a huge challenge of allocating limited audit resources more effectively. The risk-based audit planning, which is a onerous activity in time and financial resources, aims to ensure that the auditor examine the issues of greatest risk to the achievement of organizational goals. In this work, we propose a methodology of audit planning based on risk assessment with the support of a specialist system that uses fuzzy logic to execute the process of risk assessment. The rules have been defined based on risk factors selected from the Guide to Combating Corruption and Fraud in Development Projects, developed by the International AntiCorruption Resource Center. The membership functions of the linguistic terms of the nebulous rules were initially defined based on the knowledge of the author, who is Auditor of the State of Espírito Santo, and afterwards a questionnaire was applied with 17 (seventeen) other State Auditors, whose result was used to refine the membership functions of the linguistic terms, and to obtain the relevance of each of the risk factors. The prototype was executed using, as input, a sample of one hundred contracts from the Government of the State of Espírito Santo, obtained from the Transparency Portal, and the results were presented to State Auditors, who are experts in the subject, to verify if the system gives an acceptable opinion about the processes that must be audited, in other words, whether it is able to select processes that they consider relevant and reject those they consider to be irrelevant. Regarding the contracts selected by the system and rejected by the specialists, the system obtained an average of 80.5% accuracy. Regarding the contracts rejected by the system and that were selected by the specialists, the system obtained an average of 95.13% accuracy. In addition, the heuristic adopted by the system, different from the heuristics adopted by the specialists, analyzes a greater range of risk factors in a standardized way, confirming the initial hypothesis.