Settlement is the foundational function of a payments network and the source of its most consequential risk. The network that cannot settle — that cannot transfer funds between participants as the clearing process requires — has failed at its most basic obligation. The memory of settlement failures in financial infrastructure, and the systemic consequences they generate, explains why settlement risk management at payment networks has historically been conservative: over-collateralise, over-fund, maintain margins of safety that ensure the worst case is survivable.
That conservatism has a direct and measurable cost. Pre-funding held against settlement obligations is capital that is not deployed elsewhere. For network participants, the cost of that capital is real and directly tied to the size of the pre-funding requirement. For the network itself, the efficiency of the clearing and netting process determines both the participants’ capital cost and the network’s settlement risk exposure. A network that runs its clearing at lower netting efficiency than is achievable — where bilateral and multilateral netting opportunities are not fully exploited — is requiring participants to fund more gross settlement exposure than the net position would require.
The economic case for predictive settlement AI is not primarily about risk reduction in the traditional sense. It is about capital efficiency: using better forecasting to narrow the gap between conservative pre-funding and the minimum required, and using better netting optimisation to reduce the gross settlement exposure that drives pre-funding requirements in the first place.
Where settlement inefficiency concentrates
Netting efficiency — the ratio of net settlement obligations to gross transaction volume — is the primary driver of settlement exposure. A multilateral netting arrangement that achieves 95 percent netting efficiency reduces a $1 trillion gross daily settlement obligation to a $50 billion net position. A clearing operation that achieves 90 percent netting efficiency on the same gross volume produces a $100 billion net position. The difference in settlement exposure, and the pre-funding required to support it, is $50 billion. At any reasonable cost of capital, the economic value of that efficiency improvement is substantial.
Netting efficiency is not a fixed property of the clearing arrangement. It varies with the sequencing and batching of transactions through the clearing cycle, with the participation levels and position sizes of individual counterparties, and with the timing of liquidity provision within the intraday settlement window. An optimisation model that sequences clearing batches to maximise multilateral netting — grouping transactions to create offsetting positions that reduce net settlement obligations — systematically improves netting efficiency beyond what static sequencing achieves.
Intraday liquidity management faces a separate but related challenge. Settlement positions build through the trading day as transactions clear, and the peak intraday exposure does not occur at a fixed time. It depends on the transaction mix, participant activity patterns, and the timing of large-value payments. A model that forecasts intraday settlement positions and peak liquidity requirements with precision — rather than assuming a conservative worst-case throughout the day — enables participants and the network to manage liquidity more dynamically and at lower cost.
Stand-in processing and the approval rate argument
Stand-in processing — the network’s ability to approve transactions on behalf of an unavailable issuer — creates a specific precision problem with the same structure as the card fraud false decline problem. Conservative stand-in rules that decline transactions the unavailable issuer would have approved produce unnecessary declines that damage the cardholder experience and the merchant relationship. Liberal stand-in rules that approve transactions the issuer would have declined generate losses that the network must absorb or recover.
A stand-in model that has learned the approval behaviour of each issuer — their tolerance for spend amounts, geographic patterns, merchant categories, and cardholder risk profiles — can replicate that behaviour during unavailability with greater precision than static conservative rules. The improvement in stand-in approval rates directly represents transactions the network successfully facilitated on the issuer’s behalf that conservative rules would have declined. At 500 million stand-in transactions annually, a 1 percent improvement in stand-in approval rate is 5 million additional approved transactions.
The early warning dimension
Settlement risk does not become a crisis instantaneously. It builds through position accumulation by specific participants, often in patterns that are visible in intraday settlement data before they reach levels that require emergency intervention. A participant whose intraday settlement position is growing faster than historical norms, or whose funding behaviour has changed in ways that are inconsistent with their normal settlement pattern, is exhibiting early signals of potential settlement stress.
A dynamic position monitoring model that tracks each participant’s intraday settlement trajectory and generates early warning signals when positions diverge from expected behaviour enables the network to engage with the participant while options are still available — requesting additional collateral, adjusting exposure limits, or facilitating liquidity support through established mechanisms. The same early warning that enables proactive management is the signal that, if missed, becomes a settlement failure that could not have been prevented reactively.
The technology dimension
Settlement and clearing infrastructure at major payment networks runs on IBM Z, processing the clearing and settlement flows that determine daily net positions across thousands of participants. Deploying predictive settlement AI via IBM Machine Learning for z/OS keeps the inference within the clearing processing environment, with direct access to position data, participant history, and transaction flows on the same platform. The netting optimisation and intraday liquidity forecasting models that improve settlement efficiency can run continuously against live settlement data without the data movement overhead that off-platform analytics would require.
What success looks like
The primary metrics are netting efficiency rate, pre-funding excess versus minimum required, stand-in approval rate compared to the issuer’s normal approval rate, intraday liquidity cost, and settlement shortfall frequency. The programme should establish baselines for each before deployment. The bankable savings from improved netting efficiency — measured as the capital cost reduction from lower net settlement exposure — and from reduced pre-funding excess are the CFO-level metrics that justify the investment. Both are directly calculable from the settlement data that the network already captures.
