Dan Bennett, S&P Global, Head of Technology, Enterprise Data Organization

We have made good progress with formatting our unstructured, textual, data for AI consumption with the RAG pattern. Yet, here in early 2025, the industry has not centered on a definition of what AI ready data means for structured, tabular data. Everyone is talking about it; few can define it in a meaningful way.

The industry has long aligned on SQL as the preferred language for analysis of data in such warehouses. SQL implementations are typically self-describing providing a data dictionary that describes base column types, relations between tables etc.

However, this base metadata has proven unsatisfactory for enterprise use and an entire industry of incompatible semantic tooling has emerged over time to permit enterprises to add more definition, more metadata, to data. Be it the semantic model in Power BI, data catalogs, data science notebooks, or, most recently, LLM based SQL generation tooling.

Not only is the end user tasked with curating that metadata for their own data, but they must do it for any data they purchase, since – unlike tables and SQL – no standard exists for sharing this data between producer and consumer. Instead, the pertinent information is typically buried in PDF documentation.

We aim to simplify this curation process for our customers by publishing vendor neutral metadata describing the data we distribute and at this talk we will explain how and why we believe this to be a critical building block to being AI ready.