Imagine if every time you created a document in Microsoft Word you couldn’t read it anywhere else. Not as an email, PDF, or Google Doc. Any time you wanted to collaborate with a new team or adopt a new technology, you’d have to recreate the content from scratch.
That’s the structural problem Carrefour ran into. They had already built a substantial semantic layer: 26 models, 3,000 KPIs, more than 1,000 users. But it came bundled with a proprietary vendor portal, the only surface through which any of that logic could be consumed. It served self-service BI and nothing else. The knowledge existed, but it couldn’t travel. Every new use case required rebuilding from scratch.
Most enterprises don’t confront this issue until an AI project forces it: your semantic layer belongs to the enterprise, not the vendor or platform.
Carrefour operates across 40 countries, governing turnover, margin, transaction data, CRM segmentation, and supply chain metrics across a $100 billion business. The business logic was solid. The architecture was the liability. Nicolas Treanton, Head of Enterprise Analytics, Data Governance & Change, presented Carrefour’s migration at this year’s Semantic Layer Summit. What he described is an industry pattern. The solution Carrefour chose aligns with where enterprise analytics architecture is headed.
The constraint nobody talks about until it’s expensive
Enterprise data teams spend years building a semantic layer, carefully defining metrics, and establishing governance. That work is genuinely hard, and when done well, it’s valuable.
Then, inevitably, a new AI tool enters the stack. When the semantic layer is locked in a single consumption or data platform, business rules have to be rebuilt in the new environment. Definitions drift, and AI assistants return different numbers for the same metric.
Futurum analyst Brad Shimmin calls this “semantic fragmentation,” where different systems or AI models interpret the same metric in fundamentally different ways. It erodes trust and stalls AI projects before they reach production.
Treanton described Carrefour’s version of this:
“The semantic layer’s only goal was to provide self-service BI. And we were not able to reuse all this knowledge and all those rules for other projects.”
The business logic was locked inside the tool that produced it. What looked like a solved problem turned out to be a structural bottleneck.
The Shift from Tool Semantics to Open Semantics
The solution Carrefour chose was to change the architectural premise entirely. Rather than keeping business logic inside a BI tool, Carrefour migrated to a universal semantic layer with AtScale: a shared, governed foundation that exists independently of any single analytics platform.
This distinction matters. TDWI defines a universal semantic layer as
“implemented as a dedicated layer between data sources and all BI tools. Irrespective of the BI tool users choose, the universal semantic layer allows them to work with the same semantics and underlying data layer, leading to insights and reports that are consistent and trusted.”
Carrefour executed the migration under a hard deadline, and rebuilt their full model library before rollout: “We had to tell them you have this tool, now you have this one. You are able to do everything that you were able to do before.” That level of continuity only works if the underlying semantic definitions are portable. The platform changes. The metrics don’t.
Proof: The Consumption Layer Changed, the Metrics Didn’t
After migrating to a universal semantic layer, Carrefour moved its business users from a proprietary self-service portal to Google Sheets as the primary analytics interface. This required co-developing a new connector from scratch in partnership with AtScale. At the start of the rollout, the Google Sheets connector did not yet exist.
Trenton demonstrated the result live at Summit. Blank Google Sheet. Semantic layer connection. Model selection. Governed metrics returned in seconds.
“It’s very easy to manipulate,” Treanton said. “Really no technical skills needed.” Your semantic layer belongs to the enterprise, not the platform. When that’s true, the consumption layer becomes a choice.
Why This Matters for AI
Carrefour’s team is now building conversational analytics that route through the semantic layer to return governed answers. This is where the architectural decision pays its biggest dividend.
An AI agent querying raw data directly with a connection to a warehouse doesn’t guarantee deterministic answers. Without a semantic layer, context is missing. Definitions drift, and business rules are not applied.
With a universal semantic layer sitting between the warehouse and the AI agent, the governed definitions travel. The same business logic that powers Google Sheets also powers AI chat bots.
Josh Klahr, Director of Analytics Product Management at Snowflake, explained the broader industry shift at Summit:
“Customers are realizing, just like [their] data is important, I want my data to be open, interoperable, and accessible. I think customers are saying the same things about semantics.”
The Broader Market Shift
Carrefour’s story reflects the broader market shift to open semantics. The Open Semantic Interchange (OSI), backed by a consortium of data platform providers, was launched to address fragmentation directly: a vendor-neutral specification that allows semantic definitions to be exchanged across tools and systems.
Gartner’s research on the Composite Semantic Layer reaffirms the importance of OSI, and projects that by 2027, organizations that prioritize semantics in AI-ready data will increase agentic AI accuracy by up to 80% and reduce cost by up to 60%.
Analytics and AI tools will come and go. Business context should remain foundational. The cost of rebuilding governance every time a new platform enters the stack is a strategic tax on every AI initiative the business wants to pursue.
The enterprises that have addressed this problem have separated business logic from the tools that consume it. Carrefour’s migration was operationally complex, but the reason it was worth doing is simple: they now have a semantic foundation that does not have to be replaced when the next consumption layer arrives.
The Future Is a Semantic Layer That Sits Above Tools
Open semantics are a prerequisite for any analytics or AI strategy that needs to survive the next platform cycle. From SQL queries to chatbots. From tools that exist today to tools that haven’t been built yet.
Your semantic layer belongs to the enterprise, not the tooling layer. The question is whether yours does.
For organizations serious about AI, the answer has real consequences. If your business definitions are locked inside a single tool, every new use case carries the cost of rebuilding them. If they are portable, every new use case extends what already exists.
SHARE
How to Evaluate Context Platforms