Poor-quality data and ageing banking systems are limiting the benefits of artificial intelligence for treasury teams, according to a webinar reported by Trade Finance Global. The discussion, convened by Tieto and the publication, examined how payment data standards could improve automation while leaving significant challenges around trade finance documents, infrastructure and human oversight unresolved.
Many core banking systems were designed for batch processing rather than round-the-clock payments, the report said. That means an apparently instant transaction at the customer interface can still depend on overnight reconciliation behind the scenes. Trade Finance Global cited RedCompass Labs figures showing 118 billion real-time transactions in 2021, with volumes forecast to reach 428 billion by the end of 2026. Some banks have resorted to temporary parallel ledgers to respond to balance enquiries during core system maintenance.
ISO 20022 offers a more consistent framework for payment information, supporting automated reconciliation, forecasting and monitoring. However, the additional detail also increases the burden on banks: according to the report, a cross-border payment can contain more than 2,000 data points. The expense of storing, cleaning and governing that information is prompting many institutions to purchase foundational technology rather than develop it themselves.
Standardised payment messages do not resolve every data problem. Panellists highlighted trade finance’s dependence on physical paperwork, alongside tax and policy documents that remain outside structured messaging. Inconsistent bank formats also impose costs on corporate treasury teams, whose treasury management and enterprise resource planning systems need dependable inputs. Data preparation, the discussion suggested, is frequently an overlooked expense in AI investment proposals.
Reliable information could nevertheless allow AI to forecast liquidity requirements more accurately, helping treasurers reduce cash buffers and deploy surplus funds. The panel also identified potential improvements in foreign exchange monitoring and earlier fraud detection. Those gains require clear limits on autonomous decisions, with treasury leaders defining when systems may act and when people must intervene. The report contrasted that work with narrowly focused pilots and overlapping automation projects that create visible activity without fixing underlying data weaknesses.