- Funding Programme
- Year
- 2022
Use of artificial intelligence on audit of the EU Funds (AI4Audit)
AI4Audit, funded by the Technical Support Instrument, envisages the development of AI predictive models to detect and predict irregularities in EU Funds in Portugal, enhancing audit efficiency, effectiveness and proportionality, thus reducing burdens on beneficiaries. The project is developed by NOVA Information Management School and has as beneficiary institution the Inspeção Geral de Finanças – Audit Authority in Portugal.
Context
The EU has evidenced its priority in digital transition, especially through country-specific recommendations. In the context of EU Funds’ auditing, the highly time-consuming audit work and substantial workload undertaken by the auditors led to the need for a change in the current auditing framework. These problems are aggravated by the sampling design, which favours the selection of larger and more laborious operations; the rising number of operations declared to the European Commission; and the occurrence of inconclusive results that may require additional audit work nearing regulatory deadlines. The project aims to address these challenges, using data science approaches to streamline novel audit strategies.
Support delivered
Project activities were divided into three work packages (WP).
- WP1: Developed during the whole range of the project and aimed to ensure its monitoring, including progress, financial management, data management, and adequate relationship between team members and other stakeholders.
- WP2: Central WP for the development of innovative methodologies to replace or complement the traditional audit of operations approach. The new methodologies were developed aiming to guarantee the accuracy of the estimates produced, simplifying the effort spent on their production, namely in human resources and costs; to induce a better use of the data already available in public administration; and ultimately eliminate or significantly reduce the effort currently made by human resources, while increasing the confidence of conclusions about the existence of material errors and other irregularities in financed operations. This WP also included the design a new data governance, audit framework and management strategy considering/implementing the adopted techniques, envisaging the application of the results by the IGF and other stakeholders.
- WP3: Devoted to the dissemination of results, namely through scientific publications, participation in dissemination actions, and workshops with national entities and to the European Commission. A set of training and knowledge transfer actions were carried out, allowing the most relevant entities involved in the errors’ evaluation and fraud detection to use the results and recommendations made throughout the project.
Results achieved
Model predictions match traditional auditing error estimates, improving precision and reducing sample size. Considering the same sample size, it results in a 35% improvement in precision, reducing occurrence of inconclusive results. Alternatively, it is possible to reduce sample size up to 60% without losing precision. It showed to be possible to fully replace audit activities in the last period of each accounting year by model predictions. The new predictive approach also showed the ability to ungroup operational programmes, thus supporting separate audit opinions. A new sampling design based on the monetary risk was developed, showing 21%-28% precision gains and 38%-52% sample size reductions compared to the current strategy. A new complete auditing framework, incorporating risk prediction, and with sample sizes as low as twenty items, was developed as well.
More about the project
You can read the documents related to the project here: