Italian startup develops AI to predict corporate crimes and attracts the attention of institutions

Through the analysis of company financial statements on open sources, intervene in advance with AI to identify companies with suspicious behavior that are at risk of committing crimes or find those that are already committing them. An Italian startup has achieved its objective of 'predictive intelligence' which now also intrigues the institutions: for 30 September the authors of this system have been summoned by the parliamentary commission of inquiry into gangmastering, so that it can tell the institutions about the 'indicators' it has developed. 'Rozes', born in 2020, has developed to the point of evolving and obtaining the status of a university spin-off in addition to being engaged by various companies attentive to the prevention front, including the so-called 'investee companies' that aim to guarantee transparency. The project, developed on the basis of research by the University of Padua, made use of the support of criminal economics experts such as university professor Antonio Parbonetti, was scientifically published and patented.

"This is the first artificial intelligence model that is able, through the algorithm, based on the analysis of the financial statements of the various companies, to identify the degree of similarity with those that commit crimes", explains the co-founder and CEO of the startup, Jacopo Berti, clarifying that the results of the AI ​​models are a statistical indication, not an assessment, remembering that the final decision remains with humans. The algorithm extracts information from public databases and works on Anac data or various financial statements filed with the Chambers of Commerce, from which it is possible to obtain raw accounting entries and financial indices. Once the information has been compiled, it is compared with the degree of similarity to that of a reference sample of companies for which certain crimes have already been definitively ascertained by law. With this system - according to what the authors of the startup report - it would be possible to identify companies exposed to the risk of mafia collusion, bid rigging, gangmastering, but also to the probability of dissolution of a municipality due to mafia infiltration or accounting anomalies associated with fraud, money laundering, false invoicing and fraudulent bankruptcy, debt or credit scams, frontman activities and cartel activities: in total 15 macro categories of crime.

"We have been working on these issues for ten years and by studying the phenomenon we have verified how much the Pnrr funds and large capital injections in general are at risk", says Berti, who by cross-referencing millions of data is trying to reconstruct the economic world in a digital way, according to a scheme that already exists in medicine and concerns the assessment of risks. "When AI is applied in serious and concrete terms - he adds - it can have a positive impact on citizens and prevent tens of billions from ending up in mafias or laundered for terrorist or criminal activities. We calculated that if we had applied these models every year to the world of public administration procurement, the State could have recovered between 10 and 15 billion euros per year". Therefore one of the future tasks could be to extract information to facilitate investigations: "ours is a tool which, if granted, will be able to support investigators, because the method used is scientific". According to the startup's statistics, the accuracy of the algorithm is 90%: that is, out of ten legal entities indicated as risky, for nine of these the alarm analysis hypothesized by the machine is confirmed. But there is still room to implement the system: analyzing 10 risky legal entities, 4 still escape Rozes.