Many organizations are struggling with inconsistent definitions, fragmented data assets, and siloed business knowledge that can undermine trust in analytics and AI.
Semantic layers and data products are emerging as key approaches for creating shared business context that can be consistently understood and utilized across the enterprise.
DBTA held a roundtable webinar, Trusted Context for AI: Semantic Layers and Data Products, with industry experts who discussed how semantic layers and data products are helping organizations deliver trusted context for analytics, AI applications, and agentic systems.
According to Andreas Blumauer, co-founder and SVP growth at Graphwise, money is already flowing but value isn’t. No single team can maintain a comprehensive overview of all distributed knowledge assets across complex corporate systems. No LLM is able to cope with the full complexity of specialized domains, especially when parts of the answers are scattered across disconnected data and organizational silos.
Looking at AI from a primarily data-driven and technical perspective doesn't get you even halfway through the book, Blumauer said.
The blockers are fragmentation, silos, and missing (business) context, not model quality. AI will only bring about fundamental changes and increase ROI if companies leverage their knowledge advantage in entirely new ways. That's why we need a Semantic Backbone, he urged.
The semantic backbone does and offers the following:
- Makes AI even more accurate
- Saves many more tokens
- Full explainability
- Most domains can be addressed
- Complex reasoning based on linked multimodal corporate data
- Supports full AI sovereignty
Everyone is building their own analytical app and it’s a mess at scale, said Ashwin Warrier, vice president, product at incorta. The problem is not just bespoke self-service. Without a solid data foundation, you are operating without grounded semantics and context. Individual institutional knowledge and decision-making rules are not captured in context catalogs.
Data products, semantics, and feedback loops are critical ingredients for human-in-the-loop to autonomous decision making, Warrier noted.
AI needs access to the right data, not just a portion of it, explained Shawhin Mosadeghzad, director, agentic intelligence at Simba. Without understanding the business, AI will hallucinate. Without organizational context, it will infer joins and meaning you never approved. Without standard definitions, each person gets a different answer.
Susan Laine, chief data technologist at Quest Software, said AI is not lacking intelligence. AI lacks context. Context by Design answers the questions that matter the most.
According to Laine, the data product build consists of the following:
- Create product spec: Automatically generate data model based on domain specific standards.
- Find data assets: Find data in the catalog using semantic search.
- Associate business context: Automatically generate and associate semantic layer.
- Package and share: AI-ready, with trust score, interoperable.
For the full webinar, featuring a more in-depth discussion, Q&A, and more, you can view an archived version of the webinar here.