Enterprise AI adoption is stalling — not because of model quality, but because companies cannot get their internal data into shape to train and fine-tune agents that actually work.
Most enterprises already have enough data to power intelligent AI agents. The challenge is making that data AI-ready.
Why Consolidation Isn’t Enough
Years of growth and acquisitions leave large enterprises with data silos: isolated platforms inherited from other systems, often dressed up in the language of “federated data lakes,” but still impossible to query as one. The data exists. It just can’t answer a single business question across the organization.
Moving everything onto Snowflake or Databricks solves the physical problem—your data finally sits in one place. But AI still can’t make sense of it: every domain speaks its own language, and there’s no layer that translates raw tables into business meaning.
Consolidation addresses the plumbing. It doesn’t address meaning.
This isn’t just our observation. Different studies use different methodologies, but they all point to the same conclusion: Enterprise AI adoption isn’t failing on the model. It’s failing on the data behind it:
● Cloudera and Harvard Business Review found that only 7% of enterprises consider their data completely AI-ready, while 27% say it isn’t ready at all.
● Gartner estimates that organizations will abandon 60% of AI projects that lack AI-ready data by 2026.
● S&P Global reports that 42% of companies abandoned most of their AI initiatives in 2025, with an average sunk cost of $7.2 million each.
The Semantic Layer
To make consolidated data genuinely AI-ready, you need another layer on top of the physical data model: the semantic layer. It explicitly defines what the data represents—facts, dimensions, metrics, and the relationships between them.
Once that layer covers the physical model, AI can read it, map it correctly to the underlying data model, build the right query, run it, and return the answer in the same business language the question was asked in.
The industry is starting to recognize this shift. Snowflake is introducing SQL entities to describe data semantics, alongside agentic capabilities built to act on them. Databricks is moving in the same direction.
To me, that’s an important signal. The conversation has moved beyond getting data into one place. Semantic layers—and the best practices around them—are being defined in the market as we speak
From AI-ready data to production
Many organizations already have the data needed to support AI agents — across CRMs, ERPs, support systems, and internal documents. The harder part is making that data usable: connecting fragmented sources, handling sensitive information, and creating the structure and controls required for AI to work reliably.
Moving from experimentation to production often means addressing several layers at once: data preparation, a semantic layer that gives AI adoption systems the right context, governance and security, and the underlying technology platform. The approach can vary depending on the organization and its existing data environment, whether that means Snowflake, Databricks, or another platform.
One way to assess where an organization stands is the Sigma Software AI Compass, a framework for evaluating AI readiness, identifying gaps, and prioritizing the steps needed to move toward production.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Intelligent Tech sections to stay informed and up-to-date with our daily articles.
