Structuring knowledge for AI
Omnichannel panel discussion
What knowledge graphs, ontologies, and semantic layers add to enterprise AI
Many of us are trying to, or even under pressure to, adopt generative AI as fast as possible, at scale. Just as often, we’re getting pushed and pulled in directions we don’t feel are optimal, or maybe even unwise.
How do we navigate the internal communication and change-management challenges for both leadership and teams? And how do we train our AI tools to understand our business context and knowledge?
Gartner recently wrote that “by 2030, universal semantic layers will be treated as critical infrastructure, alongside data platforms and cybersecurity” (Gartner, 2026). Earlier this year, in its Top Predictions for Data and Analytics in 2026, the research company said that “leaders must budget for semantic capabilities as a nonnegotiable foundation.” A semantic layer is commonly built up of knowledge graphs, ontologies, and taxonomies – often scary but increasingly important terms. These relationship technologies are the scaffolding that gives knowledge structure and explainability to AI.
As you talk to leadership and work with vendors to build out your enterprise AI, you need to be familiar with these concepts. Otherwise, you may end up with solutions you don’t need and contracts you regret signing.
In this panel, four practitioners who design and explain knowledge structures for a living get together to compare notes. Using terms that nontechnical people can follow, they’ll cover:
- What the core terms mean
- What many leaders get wrong
- Where AI initiatives stall
- What structured knowledge contributes to an AI system that a language model can’t reliably provide on its own
Who it’s for
Knowledge or content management leaders, data and content strategists, and decision-makers who want to commission or benefit from AI and structured knowledge work without becoming specialists themselves.
What you’ll learn
What are we talking about?
Semantic layers, ontologies, taxonomies, and knowledge graphs. We’ll give you a working vocabulary and the questions to ask when a vendor or consultant offers to “build you a knowledge graph”, “RAG”, or “AI data pipeline”.
Making it work in the enterprise
The panel will discuss who owns shared knowledge infrastructure, why it works better as a programme than a project, and how teams move from a quick AI-generated test to something running and maintained in production.
Where AI fits and where it struggles
Language models are probabilistic. They favour the popular answer. But enterprise users need specific answers and a way to understand where their data, processes, and knowledge repositories are lacking. Sometimes the quick fix creates a long-term problem. We’ll look at the difference between raw data/content and AI-ready data. And we’ll discuss where LLMs can help build ontologies and taxonomies and where humans should do the work.
.
Notes and instructions
Speaker(s):
