5 Treasury Management AI Agents, Live in Production.

Treasury runs on a number that is harder to produce than it looks. A bank balance is not a cash position: it includes funds already committed, excludes facilities that are available, and spans currencies converted at whichever rate someone used. Producing the real figure means consolidating accounts manually, which means it is produced daily at best and is stale by the time a decision uses it. Forecasting inherits the problem and adds its own — receipts are assumed on contractual terms rather than on how the customer has actually paid historically, so a forecast can look comfortable while resting on assumptions the payment history does not support. Covenants and mandates suffer from the opposite failure: they are checked periodically, on a schedule that has no relationship to when a breach would occur or when a leaver’s signing authority should have been revoked.

These agents make the position and its constraints continuously known. Cash Position Tracking establishes today’s real position across every account and currency — what is available, what is committed, and what the balance figure is hiding. Liquidity Planning tests a forecast against what actually happens, checking whether assumed receipts are supported by payment history and identifying where the first shortfall lands if they are not. Loan Covenant Monitoring tests every covenant against the actual numbers with the remaining headroom stated. Treasury Controls Check examines the controls around who can move money: mandates, dual authorisation, and the signatories who should have been removed when they left. Capital Expenditure Monitoring tracks projects against their approved case — spend, stage and the benefit that justified them — and names the ones exceeding what the approval allowed.