Percorrer por autor "Bagashvili, Erekle"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Ai for anti-money laundering: the rise of llms and agentic AIPublication . Bagashvili, Erekle; Pires, Luís; Chkareuli, VakhtangThe increasing intricacy of international finance systems has heightened the demand for effective and smart anti-money laundering (AML) solutions. Conventional AML architectures, relying essentially on rule-driven alerts and manual analysis, are under increased strain given growing volumes, non-nativelanguage documents, and complex chains of transactions. Recent developments in AI, especially LLMs and agentic AI systems, have unleashed a new paradigm that could assist-not replace-AML officers' decisions on transaction analysis, onboarding reviews, and document interpretation. This study explores whether AML professionals supported by AI perform better compared to their non-AI counterparts through an experiment conducted within the Reference banking institution in Georgia. Fifty-two employees of AML are randomly assigned to either the AI-enhanced tool group using LLMs combined with secure chatbot user interfaces or the purely traditional method group. Participants perform five common AML tasks: multilingual contract analysis, company due diligence, ownership tracing, person-of-interest profiling and onboarding risk assessment. Each task is assessed by four performance measures: time efficiency, task accuracy, user satisfaction, and confidence level. After completing the tasks, both groups are required to take a survey to assess subjective workload also we are measuring their time and correction of the tasks, perceived effectiveness of the tools at hand, and openness to integrating AI into their work. The central hypothesis is that the AI-supported group will outperform the control group across key dimensions, with striking advantages for both time-sensitive and language-dependent tasks. In addition, the study investigates whether AI use increases AML professionals' confidence in their decisions and reduces cognitive load. This is a hybrid research approach: by empirical data and participant feedback, it offers insights useful in practice for financial institutions considering implementing AI as a decisionsupport layer in compliance departments. It also adds to an emerging literature calling for humancentered design of AI in high-stakes financial environments.