Real-time gross settlement is a way of settling payments between banks in which each payment is settled on its own, in full, as soon as it is processed, usually in central bank money. Payments are not netted against each other and settled later in a batch. Most countries run their high-value payment system this way, operated by the central bank.
Why does real-time gross settlement matter once AI agents act?
These systems carry the largest and most final payments a bank makes: interbank transfers, treasury movements, property completions, large corporate payments. Once a payment settles, it cannot be pulled back by the sender. An agent anywhere near these payments is near the most consequential action in the bank.
That makes the timing of checks critical. Fraud screening, sanctions checks and authority checks must all finish before submission, because there is no window after. "Real time" in the system's name describes settlement, not detection; how quickly an agent can spot fraud before a payment leaves is a separate question, covered in can an AI agent detect fraud in real time?
How does it work in practice?
- A participating bank submits a payment instruction to the central bank's system.
- The system checks the sending bank has enough funds in its settlement account.
- If it does, the system debits the sender and credits the receiver at once.
- Settlement is final at that moment.
- If funds are short, the payment waits in a queue until liquidity arrives.
The system runs within set operating hours, and banks manage their liquidity through the day to keep payments flowing.
What does that look like in an example?
A mid-sized bank's payments operations team uses an agent to watch its high-value payment queue. Mid-morning, the agent forecasts that a large outgoing payment scheduled for noon would take the bank's settlement account below the buffer its treasury policy sets, because expected incoming payments have not arrived. It flags this to the treasury desk with the forecast and a list of lower-priority payments that could be delayed. A treasury officer decides which to hold. The agent proposed; a person with authority over liquidity decided, and the agent had no power to release or delay anything itself.
How is it different from deferred net settlement?
In deferred net settlement, payments between banks are added up over a period and only the net amounts are settled at set times. Each bank owes less in total, so less liquidity is needed, but until settlement happens the banks are exposed to each other. Real-time gross settlement removes that exposure by settling each payment fully at once, at the cost of needing more liquidity during the day.
The word "settlement" also means something else in lending. A loan settlement figure is the amount needed to pay a loan off in full, a different subject covered in can an AI agent issue a loan settlement or redemption figure? and how to prepare loan settlement quotes with an AI agent.
Which related terms should you know?
- Settlement finality: the point at which a payment can no longer be unwound.
- Intraday liquidity: the funds a bank needs during the day to keep its payments settling.
The limits on what agents may do across a bank's payments are in AI agents in banking.
Questions people ask
Is real-time gross settlement the same as instant payments?
No. Both settle quickly, but real-time gross settlement systems carry large interbank payments within operating hours, while instant payment schemes carry smaller retail and business payments around the clock.
Can an agent submit high-value payments?
It can prepare and check them. Release should stay with people, usually under dual control, because a mistake at this value is final and large.