Treasury agents in production: one protocol connection cut syndicated credit processing time in half

In a joint disclosure on 8 October, commodity trader Centalion and trade-finance platform Komgo gave transaction-level figures: in the off-balance-sheet instrument (OBSI) syndicated credit business, processing time for instrument issuance fell by about 55% and for amendments by about 80%. Note that these are company-disclosed numbers; the baseline period and methodology are not public.

What they did

They did not overturn the existing system. On a trade-finance platform that was already digitalised they added a layer of agents: the model handles understanding and orchestration, while the structured process and business rules stay on the enterprise side, and the two connect over a standard protocol instead of point-to-point API development for every action. For both parties that is an order-of-magnitude difference in integration cost.

Why this scenario first

The data for syndicated credit was already in the system; the drudgery was concentrated in interaction — reading, comparing, filling in, amending. Agents can take that over at once, without first paying off the data-governance debt. If the data were still scattered across spreadsheets and email, no agent would save it.

How to read the numbers

The first two are realised, transaction-level latency improvements. The third — “more than 60 working days released per year once extended to all trade-finance business” — is a projection, and those working days are a capacity equivalent, not headcount. Against industry norms, the redeployment ratio in treasury scenarios, where analyst time moves from manual reconciliation to exception handling, generally falls between 20% and 40% — so these numbers sit in a reasonable range.

Six repeatable moves

• Choose the system: pick a business system that is already digitalised, has a structured data foundation, and where interaction is still the bottleneck.
• Build the agent yourself and deploy it in your own environment, keeping orchestration logic on your side — that is where control over future upgrades lives.
• Connect over a standard protocol; do not do point-to-point integration.
• Run one small scenario at transaction level first; do not roll out across the whole business on day one.
• Set metrics in business terms such as processing time, not in number of calls.
• Once it works, copy it sideways to similar business — copy the framework, not the code.