Payment exceptions are governance moments, not merely operational incidents. When human operators resolve flagged payments – increasingly with AI-generated analysis shaping their judgement – the resulting decision must be capable of withstanding regulatory examination, client challenge and audit scrutiny. This paper aims to examine why current governance frameworks fail at that task and propose a structured remedy.
Conceptual paper grounded in practice-based analysis. The exception taxonomy is built through normative analysis of Swiss regulatory requirements (FINMA, AMLA and BCBS) and payment operations literature. The exception decision record (EDR) framework is derived deductively from four identified governance failure modes.
Four governance failure modes characterise AI-assisted exception environments: the invisible recommendation, the unexplained override, the undocumented escalation chain and the retrospective rationalisation. Not all of these are equally tractable – the third, involving multi-actor escalation chains, raises implementation difficulties that the other three do not. The EDR – a four-layer instrument comprising exception classification, AI contribution record, human decision record and evidentiary anchor – provides a governance architecture responsive to these failure modes, though its implementation in live institutional environments will require adaptation. Analytical scenarios illustrate its application.
The paper offers two contributions: it theorises payment exception handling as a high-intensity hybrid decision environment with accountability properties distinct from general AI governance contexts; and it extends decision traceability from an abstract governance principle to an operationally implementable instrument grounded in Swiss regulatory and fiduciary obligations.
