Five Questions Every CDO Should Ask About AI and Semantics

The metrics powering your AI agents right now, like margin per square foot or total economic cost, probably don’t belong to you. They belong to whichever platform computes them today.

Snowflake, Databricks, the frontier LLM labs, Microsoft, Salesforce, Google, and every metadata, knowledge graph, and semantic layer vendor tries to give your data the context AI needs. Get context wrong in one place, and every AI agent inherits the error. 

Gartner suggests without a clear understanding of the specific relationships and rules inside an organization’s data, AI agents “cannot operate accurately and are far more likely to hallucinate, introduce bias and produce unreliable results.” By 2027, they predict that organizations that prioritize semantics in AI-ready data will see up to 80% higher agentic AI accuracy and 60% lower costs.

For CDOs, here are five questions worth asking before you make a long-term investment in how you manage your context.

Does AI Understand the Business, or Just the Data?

LLMs have gotten remarkably good at language, but they don’t know your business. Most AI initiatives connect models directly to data, retrieve documents, expand context windows, and refine prompts. They don’t specify which SQL table defines customer, revenue, or margin. And without shared meaning, agents, dashboards, and applications will generate different answers to similar questions.

In IBM’s 2025 global study of 1,700 CDOs, only about one-third could clearly explain how their data drives business outcomes. Ninety-two percent say proving business impact is now core to the job. A shared-definition problem is a big reason why. You can’t measure what revenue means if every system defines it differently. Close that gap, and your agents stop guessing. They start giving the same answer as your board deck.

So, the first question to ask is: What business terms, metrics, and meaning does AI need, and where do these definitions live?

If You Switched Platforms Tomorrow, Would Your Context  Survive?

Every enterprise changes or has multiple warehouses, AI co-coding tools, or LLM providers. Yet most organizations tie their business definitions to whatever platform happens to compute them today, and pay for it later when something changes.

Open table formats like Apache Iceberg have brought standards to how organizations store data once in their own object storage and choose whichever engine best fits the workload.

Business semantics need the same portability, and Apache Ossie is emerging as that standard: a vendor-neutral way to define metrics, dimensions, and business logic once so any tool can read and compute them consistently. A metric defined once should compute the same answer whether it’s queried by a dashboard, an application, or an AI agent, regardless of which warehouse or model sits underneath it.

Ask yourself: if I switch from Claude to OpenAI, will my definitions of “Same store sales” be portable, or are they locked in a Claude SKILL file that’s proprietary, unmanaged, untested, and ungoverned?

Is it Right Once, or Right Every Time?

Traditional governance was built around people: who can see what, who approved what. AI governance has to operate continuously, on every query, with no human in the loop to catch a wrong number before it reaches a customer or a regulator.

IBM found that eighty percent of CDOs surveyed are building diverse datasets to train AI agents, yet 79% admit they’re still early in defining how to scale and govern them.

Gartner agrees:

“Context with semantic coherence will become a cost-control and trust strategy, not a nice-to-have.”

The question is whether AI is correct every time, and whether you can prove it after the fact. Get this right, and a rule enforced once at the semantic layer holds everywhere that logic gets used, no re-checking every dashboard and every agent by hand.

Ask: Where is governance actually enforced, and does every data-consuming application abide by it?

Are You Buying an AI Platform, or Someone Else’s Ecosystem?

Most vendors naturally optimize for their own ecosystem, but CDOs should understand the tradeoff before they sign.

Tableau locks semantics inside Tableau Semantics. Snowflake bakes semantics into Semantic Views. Microsoft’s architecture traps semantics in Power BI and Excel.

Leaders should manage semantics as a sovereign asset. That means putting semantics into a neutral, independent, universally accessible platform, like AtScale. Snowflake gets this right in practice: its partnership with AtScale frees Semantic Views through AtScale’s XMLA endpoint, so that semantic layer isn’t locked to any single BI tool.

IBM’s study found that 84% of CDOs say their unique data products have already delivered significant competitive advantage. That advantage is fragile if the logic behind it resides in a single vendor’s proprietary layer rather than your own, governed model. Business logic that can’t outlive the platform it runs on was never really yours to begin with.

Ask: Does my semantic layer and governance approach give me the flexibility to choose new tools and reuse context, or does it just deepen our dependence on this vendor?

Open Standard, or Proprietary Trap?

Every generation starts with proprietary innovation. Eventually, standards emerge, and the fastest movers keep their options open. This is happening now with context and semantics.

The Model Context Protocol standardizes the connection: how an AI agent reaches into a tool or data source to pull out context. Apache Ossie standardizes the content: the business definitions, metrics, and dimensions themselves. One’s the pipe, the other’s what runs through it. Together they ensure the logic outlives whichever vendor computes it today.

Ask: Is our semantic infrastructure built on Ossie? Have we chosen tools that have a clear roadmap to support it? Portable standards?

Context and Semantic Management Is a CDAO Decision

When it comes to context, the organizations that have the clearest and most portable understanding of what their business actually means will be the AI leaders.

Business logic should outlive the platform that computes it today. That’s the case we make, in more depth, in AtScale’s Context Buyer’s Guide, a framework for evaluating context platforms on what they actually do. Download it before your next platform investment.