VLAIO (Flanders Innovation & Entrepreneurship) supports entrepreneurs in Flanders through different grants such as the SME e-wallet (KMOP), Ecology Premium Plus (EP-Plus) or COVID-19-related subsidies. To aid inspectors in targeting high-risk subsidiaries, Devoteam developed a fraud detection tool for each of the three aid measures that provides a score on the calculated risk. The tool doesn’t only pick out potential frauds, but also saves the inspectors lots of time, a win-win situation!
VLAIO is the point of contact for entrepreneurs in Flanders. The institute encourages and supports innovation and entrepreneurship and contributes to a favourable business climate. Among other objectives, VLAIO support businesses through grants in order to improve their growth, transformation, and innovation.
The SME e-wallet aims to subsidise the missing strategic knowledge of SMEs that is necessary to establish new growth strategies in the case of a turning moment via internationalisation or innovation. The Ecology Premium Plus on the other hand, is granted to enterprises that make certain ecological professional investments in Flanders. The government also took some jurisdictional measures and reliefs because of the pandemic. Companies that faced financial difficulties because of COVID-19 could apply for several social and fiscal support measures.
The challenge
The SME e-wallet (KMOP), Ecology Premium Plus (EPPLUS) and COVID-19-related subsidy require an automated fraud detection system to aid inspectors in targeting high-risk subsidiaries specifically, since a manual case-by-case approach is not possible given the high number of requested grants.
So, how does it work? The fraud detection system provides a score on the calculated risk for each submitted request. These scores are composed by the combination of different underlying methods, which each consider risk in a different way. The resulting data is visualised in a custom-built Power BI dashboard that inspectors can use to identify high-risk applicants and perform targeted inspections accordingly.
One important constraint of the system is the sensitivity of the data that is used to determine risk. For example, it includes data on sole proprietorships and thus data on individual citizens. Therefore, keeping data secure is extremely important. All steps in the process, from ingesting the data to displaying the data to inspectors, must be highly secure.
The solution
The fraud detection system takes in data from multiple sources, including VLAIO’s own data and data from the Central Database for Enterprises (KBO). This data is processed, and multiple methods are used to define risk. One of these underlying methods is community detection (the act of identifying groups of companies with close ties to each other). Other methods include rule-based methods, supervised machine learning and anomaly detection.
Combining the different methods, an overall risk score is defined for each submitted request for subsidies. These risk scores can aid the VLAIO inspectors in targeting high-risk instances first. More low-level information is also provided to the inspectors to aid them in understanding the fraud detection system’s decision process. This low-level information also helps inspectors to gain more insight into subsidy requests and the companies behind them, in an interactive, visual and user-friendly fashion.
The fraud detection system is built within the Azure environment of VLAIO and its code is written in Python, used in combination with other Azure components, like an Azure Data Factory (ADF) used to automate the process. The results are automatically gathered in a Power BI dashboard.
Main benefits
Since the deployment of the fraud detection system and the Power BI dashboard, inspectors can quickly identify high-risk subsidy requests and companies. They also have graphs at their disposal that allow them to visually represent companies and their relations to others. While the previous solutions were limited to searching for values in Excel sheets, the new dashboard improves readability and reduces time spent on each case. Furthermore, the custom-built Power BI dashboard acts as an information hub, saving inspectors from having to access multiple systems to get to all the needed information.
