The ISO 20022 series
10 essays on what ISO 20022 makes possible once the migration is done, written between October 2023 and January 2026. In reading order rather than newest first.
-
ISO 20022 — An enabler for AI (and BI)
Most ISO 20022 attention goes to transactional services. This looks at the analytics side: what the data enables for AI and BI once it travels with the money.
-
ISO 20022 — Navigating rich semantics
ISO 20022 is a strategic data asset, but its complex structure and granularity make it hard to navigate. How to work with the semantics rather than around them.
-
ISO 20022 — Address structuration
Benchmarking large language models against specialist ML and NLP packages for structuring postal addresses, and why more real-life bank data is needed before drawing conclusions.
-
ISO 20022 — Road to (data) riches
How much richer is ISO 20022, measured rather than asserted: a metadata analysis of every published version of pacs.008, from SEPA V02 to V09.
-
ISO 20022 — Front-to-back-to-front
Front office and back office run on different data and rarely share it. Told through two data scientists at one bank, this is how ISO 20022 could bridge them.
-
ISO 20022 — A ready-to-use Knowledge Graph?
Turning ISO 20022 XML into a knowledge graph with an LLM, and querying it in plain English: what four models produce, and where Graph RAG helps.
-
ISO 20022 — Visualization
A visualization tool for ISO 20022 data: geolocated entities, message structure and payment chains. The video traces a suspicious cross-border transaction.
-
ISO 20022 — De-risk your migration
Migrating to ISO 20022 unlocks operational efficiency and deeper transaction insights, while paving the way for high-quality AI-powered services.
-
ISO 20022 is not a format change. It’s a data shock.
For years, ISO 20022 was described as a technical migration. An upgrade. A schema change. A regulatory mandate. Now a major milestone has finally been reached.
-
TxAgent: An ISO 20022 AI Agent Industrialized with Massive Parallel Testing
An AI investigation assistant has to be reproducible, not just fast. How TxAgent is tested with evaluation-driven development and parallel regression runs.
Everything else, including company news, is on news and articles.