Funding Source: Innosuisse

XAI-FinCrime aims to transform financial crime prevention and detection processes, addressing high false positive rates and inefficiencies due to significant manual effort. We leverage machine learning, generative AI, and explainable AI (XAI) to enhance fraud detection and compliance management. We use machine learning models to analyze transactional behaviors and multimodal Large Language Models (LLMs) to process unstructured data. Counterfactual explainers will enhance AI decision understanding, crucial for building trust, ensuring efficiency, and safeguarding compliance.

Recent studies report global increases in Financial Crime Compliance (FCC) costs, with the EMEA region spending $85 billion, 41% of which is labor costs. Swiss financial institutions spend $6.4 billion annually on FCC, with 59% for salaries. This highlights a significant market for efficient, innovative solutions.

Current solutions are barely adequate to tackle the increasing complexity and growing importance of this area. Compared to the current rule-based industry standards, our envisioned AI tools aim to significantly reduce their typical False Positive (FP) rate of 95% while still providing better F1-scores, accuracy, and True Positive (TP) rates. Additionally, our XAI tools will enhance user understanding of the overall AI system and thus reduce investigation effort per alert.

We expect initial deployment and validation through pilot studies, ensuring that our innovations can quickly benefit financial institutions by improving the efficiency and effectiveness of their compliance processes.

Partnering with international collaborators in the context of ITEA provides access to diverse real-world data and expertise, important for empirical validation and refinement of our AI solutions. Moreover, the global reach of the consortium facilitates the international dissemination of project findings, extending the impact beyond the Swiss financial industry.