Why Generative AI Is Forcing Us to Rethink Data Modeling

Photo by Meral Avdanlı on Unsplash

A few months ago, I watched a non-technical colleague upload an Excel file into a GenAI chatbot. Within seconds, they were getting deep insights, asking follow-up questions, and navigating the data like a seasoned analyst. No SQL. No dashboards. Just natural language.

Read for free here

That moment stuck with me.
It was a glimpse into what’s possible when Large Language Models (LLMs) meet structured data. The promise of semantic layers and data governance suddenly had a new spotlight. Text-to-SQL capabilities are evolving fast, and the idea that anyone — regardless of technical skill — can query data conversationally is no longer science fiction.
But here’s the catch: most enterprise data isn’t ready for this.

The Problem with Traditional Data Models

In large organizations, data is often locked away in highly normalized relational databases. These systems are optimized for performance and storage, not for context or clarity. Field names like cust_id, prod_cd, BKPF or txn_dt are great for machines — but meaningless to LLMs trying to interpret intent.

And that’s just the surface.
What happens when the relationship between a customer and a product isn’t transactional, but…