Details
Why Your AI Needs an Ontology: Getting Started with FIBO
It is still early, but the research is already in: ontologies significantly optimizes AI platform performance, including agents and Gen AI, by providing structured domain knowledge. This allows AI models to connect concepts and fragmented data and validate the logic of outcomes. The result? Improved accuracy and reasoning, lower cost, more explainability, and reduced failures such as hallucinations. The ontology for the financial services sector, FIBO (Financial Industry Business Ontology), is available from the EDM Association. But what is in it? How to find and use it? How to gain the benefits of FIBO for my AI program?
The EDM Association’s webinar on August 25th, 2026 will get you started on FIBO for AI. EDM Association experts and members will talk about the what, why, and how of FIBO, including:
- Why AI needs ontologies
- What are ontologies, and where they fit into the data architecture
- What is FIBO
- Use cases and a report from the field
- FIBO resources: How to get started with FIBO, including the new EDM Association community for FIBO, the FIBO Forum
This is your chance to engage directly from the experts about how – with FIBO – to really supercharge your data program to meet the challenges of AI!
Speakers
Post-event summary
The webinar, “Why Your AI Needs an Ontology: Getting Started with FIBO,” explored how ontologies can address some of the reliability, accuracy, and efficiency challenges organizations face when applying AI to financial data. Expert speakers included:
- Jim Halcomb, Chief Research & Development Officer, EDM Association
- David Mark Blaszkowsky, Senior Program Manager, EDM Association
- Michael Koegler, Senior Program Manager, EDM Association
Jim opened the discussion by positioning AI-ready data as part of a broader data management journey, while David introduced the Financial Industry Business Ontology (FIBO) and explained how it can provide greater context and structure for AI. The speakers moved from the conceptual case for ontologies to practical research demonstrating how FIBO, semantic layers, and knowledge graphs can improve the performance of large language models.
A central theme was that improving AI performance requires organizations to address the data and context provided to their models. Ontologies provide a semantic structure that defines concepts, terminology, relationships, rules, and constraints in a form that machines can interpret. FIBO applies this specifically to financial services, providing defined financial concepts across areas such as securities, entities, derivatives, loans, corporate actions, and agreements. Rather than requiring organizations to replace their existing data environments, FIBO can be used to connect data to a common semantic structure, helping establish the context and relationships needed for AI to work more effectively.
David also discussed how FIBO can sit between an organization’s data and AI technologies to provide additional guardrails. As ontology management and deployment technologies continue to evolve, he noted that organizations have more options for integrating FIBO into their environments, including emerging products with FIBO already built in. The development of data products was also highlighted as a way to enable organizations to use specific FIBO components rather than requiring larger deployments, potentially making adoption easier and more targeted.
Michael then presented early research from EDM Association’s Data Management for Asset Managers (DMAM) forum comparing an LLM-only approach with a pipeline incorporating FIBO, knowledge graphs, and a semantic layer. Using approximately 30–40 bank loan credit agreements and extracting 60–70 critical data elements, the LLM-only approach achieved around 76% accuracy, while the semantic-layer pipeline achieved more than 95%. The research also showed differences in cost: initial LLM costs decreased from approximately $0.39 to $0.14 per document, while subsequent query costs fell from approximately $0.21 to $0.02—a 90% reduction. Michael also emphasized that the semantic-layer approach provided greater explainability and auditability.
Michael captured the potential impact with a simple analogy: “If you’re deploying LLMs across your organization, and you’re not using a semantic layer, you’re playing grade school t-ball. And if you are using it, you’re the New York Yankees.” The comparison reinforced the broader message that semantic layers and knowledge graphs are becoming increasingly important as organizations look to improve the accuracy, efficiency, explainability, and scalability of AI.
The overarching message was that effective AI requires more than the LLM itself: connecting organizational data to shared meaning, relationships, and context can provide a stronger foundation for trusted and effective AI in financial services.
Get involved!
The webinar concluded by encouraging organizations to continue exploring FIBO through the EDM Association’s FIBO Forum on EDMConnect. David outlined resources including tools, introductory videos, research, office hours, and working groups intended to help practitioners understand FIBO and build the case for its use within their organizations.
