Applied Machine Learning to Anomaly Detection in Enterprise Purchase Processes

The Internal Audit department of a company (normally multinationals groups and/or big-sized entities) is aimed to ensure the correctness and effectiveness of the entities' processes, its compliance to the approved internal policies and to reduce risks in any form that could be presented [1]. In order to achieve this goal, the companies' internal teams conduct audits through on a regular basis defined audit engagements. During their missions, the auditors identify, evaluate and document adequate information to achieve the objectives of the engagement [2], carrying out interviews with the auditees and performing a rigorous tracking of evidences supporting the audit findings. Currently, auditing still mainly relies on sampling the information (registers, transactions, etc.) to assess the processes' compliance during the audit engagements [3]. Consequently, the so-called sampling-risk makes that relevant information in the registers/transactions could remain out of the sampling selection to be reviewed. Additionally, with the growing amount of data, this traditional approach becomes obsolete, and the sampling risk is aggravated [4]. Among the business processes, a special interest resides in searching for anomalies or misbehaviours on purchases. Internal audit and purchase managers need to prospect, evaluate, and select the methodologies and IT tools capable of monitoring expenses and discovering relevant information that can highlight an out-of-policy act or, even, fraud [5, 6]. The goal is to automate processes within the company that help to prioritize the investigation activities according to the level of suspicion of any fact.