Most AI pilots succeed in controlled conditions and stall when organizations try to scale them. The problem is almost never the model. This session brings the full seven-part AI Readiness series together in a practical, phase-by-phase walkthrough of a 12-week architecture pilot designed to deliver a working RAG system and a blueprint for enterprise rollout.
- Use Cases Drive Everything: Good is defined by the user and the use case, not the model. A well-formed use case has a clear role, action, and outcome. It resolves to a pass or fail, and that testability is what makes a pilot provable rather than just demoable.
- The Semantic Layer is the Foundation: Phase 2 produces three concrete deliverables: a domain glossary, a metadata schema, and an information architecture. Each one is determined by the use cases, not the other way around. You build what the use cases require, nothing more.
- IAD-RAG Directs Retrieval with Structure: Information Architecture-Directed RAG retrieves within the boundaries defined by the information architecture. Instead of letting the AI wander the warehouse and grab whatever looks similar, you hand it the exact aisle and the shelf.
- Componentization Makes Content Machine-Ready: Transforming prose into typed, semantically chunked components is what allows AI to retrieve complete, accurate answers rather than fragments. Naive chunking splits procedures down the middle. Semantic chunking keeps complete thoughts together.
- If You Cannot Benchmark It, It Is a Demo: Measuring the current state before the intervention and comparing it to the outcome after is what makes value provable. Benchmarks connect the data-level metric to the process, the outcome, and the strategy.
Speakers