This paper aims to introduce a data science approach for guiding auditors to accurately select regions suspected of frauds in welfare programs benefits distribution. The technique relies on Newcomb–Benford’s Law (NBL) for significant digits. It has been analysed Bolsa Familia data from Federal Government Transparency Portal, a tool that aims to increase fiscal transparency of the Brazilian Government through open budget data. The methodology consists in submit four data samples to null hypothesis statistical methods and thereby evaluate the conformity with the law as well as the summation test which looks for excessively large numbers in the dataset. Research results in this paper are that beneficiaries’ cash transfer per se is not a good test variable. Besides, once payment data are grouped by municipalities, they fit NBL, and finally, when submitted to the summation test, the distribution of the Bolsa Familia payments in several municipalities shows some fraud evidence. In this sense, we conclude the NBL can be an appropriate method to fraud investigation of welfare programs’ benefits distribution having beneficiaries’ payment geographically grouped.

本文旨在介绍一种数据科学方法，以指导审计师在福利计划收益分配中准确选择涉嫌欺诈的地区。该技术依赖于纽康-本福德定律（NBL）的有效数字。已对来自联邦政府透明度门户网站的Bolsa Familia数据进行了分析，该工具旨在通过开放预算数据来提高巴西政府的财政透明度。该方法包括将四个数据样本提交给无假设的统计方法，从而评估与法律的一致性以及求和检验，该求和检验在数据集中查找过多的数字。本文的研究结果是，受益人的现金转移本身并不是一个很好的检验变量。此外，将付款数据按城市分组后，它们就适合NBL，最后，在接受总和测试时，Bolsa Familia付款在多个城市的分布显示出一些欺诈证据。从这个意义上讲，我们得出结论认为，将受益人的付款按地理位置分组的NBL可能是对福利计划的利益分配进行欺诈调查的一种适当方法。