Why data decay is also an AI problem
One dimension that rarely gets the attention it deserves is the downstream impact on artificial intelligence (AI) and machine learning (ML). AI models for fraud detection, anti-money laundering (AML) screening, liquidity forecasting, and client analytics are only as accurate as the data they train on.
ISO 20022’s structured XML format is a natural fit for AI. Granular, well-tagged data fields produce sharper predictions, lower false positive rates, and more precise client analysis. But if a bank’s payment data pipeline still runs on translation layers, it fails to capture purpose codes, LEIs, or structured remittance details at the source. The models end up training on incomplete data, and the intelligence they produce reflects that limitation.
Banks building native MX processing capabilities are capturing structured data from day one and accumulating a dataset that compounds in value over time. That structural advantage has already started the clock.
What native ISO 20022 data powers:
- AI-driven fraud detection trained on ultimate debtor identifiers and purpose codes can pinpoint the true economic origin of payment flows.
- Intraday liquidity forecasting using purpose-tagged payment data (payroll, FX settlement, supplier payments) and timestamps enables cash flow models that inform capital requirements.
- Automated reconciliation using remittance fields carrying invoice references to match open receivables without manual intervention.
- AML precision through models that reference structured LEIs, purpose codes, and transaction histories in real time to reduce false positive rates.
The next compliance wave is already here
The structured address deadline approaching November 2026 adds urgency. From that date, SWIFT will reject payment messages containing fully unstructured addresses. Many large banks hold years of unstructured address data across client records, enterprise resource planning (ERP) integrations, and onboarding systems. Cleansing, converting, and validating this data is not a quick task
The mandate does not stop there. By November 2027, exception and investigation messages (MT199/MT299) must be replaced with camt.110/camt.111 messages. Full migration of reporting messages from MT9xx to camt is expected to extend through 2028. For banks with native ISO 20022 data architecture, each new requirement becomes an incremental adjustment. For those still reliant on translation layers, every mandate becomes another heavy-lift project.
