Issuing a bank guarantee at a large commodity trader is still a desk job. A treasury employee checks how much room is left on a credit facility, pulls the banking details from one system, confirms the limits in another and keys the transaction into a third. At a trader that booked $144 billion in revenue in 2025, that routine repeats for every instrument.
Centalion, the Geneva trader formerly known as Gunvor, said Oct. 14 it has handed that legwork to an artificial intelligence agent. The agent runs on Komgo, the trade finance platform Centalion has used for more than five years, and lets treasury staff ask in plain language whether capacity exists to issue or amend an instrument. It then pulls the data, checks the facility rules and prepares the transaction for a person to approve.
On the first facility covered, issuance processing time fell about 55% and amendment processing time fell about 80%, Centalion said. The figures are the company’s own and have not been independently verified.
Digitized Workflows Still Left Staff Jumping Between Systems
Centalion’s treasury team had already digitized its guarantees and standby letters of credit on Komgo’s platform. The problem, the company said, was that staff still jumped between systems to retrieve data, check it and chase status updates. Digitization was done. Working the tools was the new cost.
The agent lives in Centalion’s own cloud and pairs a large language model with the company’s treasury workflows and facility rules. It connects to Komgo through the platform’s Model Context Protocol (MCP) connector, which lets outside AI tools read from and write to Komgo. That meant Centalion didn’t need a custom integration project. Komgo, founded in 2018 by a group of banks and commodity traders, serves more than 400 clients, according to Centalion’s announcement.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
Olivier Thevenon, Centalion’s treasury manager, described the goal as “removing friction without removing accountability.” Once the agent covers all of Centalion’s trade finance activity, the company projects it will free more than 60 working days of treasury capacity a year for risk oversight and financing strategy.
Treasurers Spend Up to 70% of Their Time on Low-Value Tasks
Centalion’s starting point is common. Treasury functions spend 60% to 70% of their time on manual and low-value activities, EY India found in its “An Agentic AI Adoption Playbook for CFOs and Treasurers.” Sixty-three percent of finance leaders polled at J.P. Morgan’s EMEA Treasurers Forum expect employee productivity to be the area technology will change most in the next 12 months, PYMNTS reported.
AI agents are starting to take on core treasury tasks, from intraday liquidity decisions to foreign exchange forecasting, inside large banks and corporate treasury teams across Europe, according to the same PYMNTS story. PYMNTS CEO Karen Webster has said generative AI and agents bring “a contextual aspect and a reasoning aspect” to the office of the CFO, rather than just faster processing.
Trade finance is a slow corner of that work. Guarantees and standby letters of credit sit between a corporate, its banks and its counterparties, and every change to an amount, a date or a beneficiary goes back through the same checks. An 80% cut in amendment time hits the part of the process that repeats most.
What separates this from a chatbot is where the controls sit. Centalion said checks and approvals are built into the workflow itself rather than left to the model. The agent prepares. A person decides. The company also said cutting manual rekeying lowers operational risk, since the data moves from the platform to the transaction without anyone retyping it.
Centalion calls its first facility “the first production use case, not the end state,” and plans to extend the agent across the rest of its trade finance activity. Thevenon is clear about what the company was trying to avoid. “We didn’t want to just build another chatbot,” he said.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
Add as Preferred Source
