AI agents for cash management in payment systems
Summary
Focus
Payment systems are the lifeblood of modern economies and facilitate the transfer of value between individuals, businesses and governments. We examine how generative artificial intelligence (gen AI), such as ChatGPT, can assist in managing cash and liquidity in real-time gross settlement (RTGS) payment systems. These systems are essential for processing large financial transactions between banks in real time. Managing liquidity within them is a delicate balancing act. Cash managers must ensure there is enough liquidity to process payments without incurring unnecessary costs or delays.
Contribution
We contribute to the growing discussion on the role of AI in financial systems by conducting a series of experiments to test the decision-making abilities of a gen AI agent. We simulate various real-world payment scenarios, with varying levels of complexity, including situations with liquidity constraints, competing payment priorities and uncertainty about future transactions. By doing so, we provide a first-of-its-kind exploration of how AI can mimic human cash managers in high-value payment systems, offering insights into the potential for automating routine tasks in financial operations.
Findings
Our findings suggest that even without specialised training, gen AI agents can replicate key cash-management tasks. The AI agent was able to maintain precautionary liquidity buffers, prioritise urgent payments and balance trade-offs between liquidity costs and settlement delays. The agent's decisions were consistent across various scenarios, demonstrating its ability to adapt to uncertainty and make informed choices. The findings suggest that integrating AI into payment systems could reduce operational costs, improve efficiency and enhance system resilience. However, we also highlight the need for regulatory safeguards, human oversight and further research to ensure safe and responsible adoption of AI in financial market infrastructures.
Abstract
Using prompt-based experiments with ChatGPT's reasoning model, we evaluate whether a generative artificial intelligence (AI) agent can perform high-level intraday liquidity management in a wholesale payment system. We simulate payment scenarios with liquidity shocks and competing priorities to test the agent's ability to maintain precautionary liquidity buffers, dynamically prioritize payments under tight constraints, and optimize the trade-off between settlement speed and liquidity usage. Our results show that even without domain-specific training, the AI agent closely replicates key prudential cash-management practices, issuing calibrated recommendations that preserve liquidity while minimizing delays. These findings suggest that routine cash-management tasks could be automated using general-purpose large language models, potentially reducing operational costs and improving intraday liquidity efficiency. We conclude with a discussion of the regulatory and policy safeguards that central banks and supervisors may need to consider in an era of AI-driven payment operations.
JEL classification: A12, C7, D83, E42, E58
Keywords: generative AI, agentic AI, LLM, payments systems, liquidity management
