Why AI Projects Fail: The Missing Business Data Layer Every Company Needs

The same mistake that held back business intelligence is now slowing down AI, and the fix is the same one companies ignored the first time.

AI adoption is stalling for many reasons, but one stands out because it is a rerun. Twenty years ago, BI dashboards failed to deliver not because dashboards were a bad idea, but because the meaning behind the data was never stored in a reusable way. Every new report required someone to rebuild the same business logic from scratch. The result was a "reimplementation tax" that slowed everything down and routed every question through a handful of expert heads.

That same bottleneck is now strangling AI agents. An agent can't wait for the one analyst who knows what "gross margin" really excludes. Pointing an LLM directly at raw databases doesn't fix this: it's slow, expensive, and produces results that are "confidently and inconsistently wrong" because the model makes different judgment calls about your business every time it runs.

The Fix: A Semantic Layer

The solution is a semantic layer, a central place where business data definitions are stored so they can be reused rather than rebuilt for every query. By September 2025, a coalition including Snowflake and Salesforce launched the Open Semantic Interchange to standardize this approach. By early 2026, AWS and Databricks had joined. Even Tableau, historically skeptical, published that "the agentic future demands an open semantic layer."

Key Takeaways

• Without a semantic layer, AI agents fail expensively: every query burns compute re-inventing business logic that already exists somewhere in the organization.
• The people who can fix this already work at your company: BI analysts, finance managers, and operations leads who carry your business definitions in their heads.
• A good semantic layer pays off twice: it unblocks AI agents AND finally delivers the analytical promise BI never kept.

Read the full article on The AI Journal