Metadata for the AI You Haven't Built
About this session
Metadata quality is as critical as data quality. It is no longer the hidden asset that no one funds. It has a real job now, and here is the progression that got us here.
First we designed metadata for humans to read. Then we used AI to generate that metadata, because the work is tedious and nobody wanted to do it by hand. Now we need metadata good enough for AI to read.
That last step is the hard one. Metadata was built for humans. The reader was always a person. An analyst, a steward, a BI tool running rules someone wrote by hand.
AI changes who is reading. An agent or a model now consumes the data and acts on it. It cannot lean over and ask a colleague what a field means or whether it is safe to use. So the context that used to live in people's heads and half-maintained Confluence pages has to be written into the data itself, in a form a machine can act on.
That’s the silent but significant change we are failing to notice. Metadata stops being the thing you document after the data lands, and becomes the thing that decides whether AI can use your data at all. The right approach is to treat it as the product.
The organizations that get value from AI over the next few years will not be the ones with the best models, because everyone has access to the same models. They will be the ones whose data can actually be read by a machine that cannot ask a follow-up question.
In this session, I will provide practical tips on how I approached this problem
Speaker
Key takeaways
- Model intelligence through metadata
- Struggle to structure metadata
- Why metadata quality is unlike data quality