Enterprise OLAP engines, BI tools, and metric systems have been in place for years. But once organizations move into natural language data Q&A, cross-engine analytics, and AI agents consuming data automatically, hidden problems surface quickly. The same metric may have different definitions across systems, tables and columns may lack consistent business meaning, and governance rules often do not travel with the analysis flow. As a result, analytics becomes hard to reuse, AI-generated answers become hard to trust, and many promising AI use cases fail to reach production.

This talk shares practical observations from real enterprise conversations and solution design work. It focuses on the most common gaps that appear when OLAP systems evolve toward AI-driven analytics: fragmented metadata, missing semantic layers, metric definitions that cannot be reused consistently, governance that stops at system boundaries, and unstable or uncontrolled data access context for AI applications.

I will discuss an integrated approach for open data stacks. The idea is to use a unified metadata layer to organize data objects across multiple engines and catalogs, and a semantic layer to define reusable metrics and analytical meaning, then expose these capabilities consistently to BI tools, natural language data Q&A, and agent-based applications.

The talk will cover three questions. First, why do OLAP systems face new consistency and governance challenges after AI is introduced? Second, how should metadata, semantic layers, query engines, and AI agents work together, with clear roles in discovery, definition, access control, and execution? Third, how can architects balance ecosystem openness, implementation complexity, and user experience, so analytics can not only run, but also be reusable, explainable, and governed?

Rather than focusing on a single query engine, this session looks at the problem from an Apache data ecosystem perspective and explores how OLAP systems can add the missing semantic and governance layer needed for more trusted analytics, more natural data interaction, and more controlled AI-driven data consumption.

Speakers:

Binhui Liu: Solutions Engineer at Datastrato, focusing on metadata, data governance, and AI-ready data infrastructure

Levent Liu is a Solutions Engineer at Datastrato, focusing on metadata, data governance, and AI-ready data infrastructure. He works with enterprise users on open data architecture adoption, especially around metadata management, lakehouse interoperability, and AI access to governed data. His recent work centers on how metadata, semantics, and governance can support more reliable AI use cases such as natural language BI, RAG, and agent-based workflows.