Mercedes-Benz Korea is building an AI-ready semantic layer on Databricks Unity Catalog to support its Talk to Data initiative, ensuring consistent and explainable answers across BI and AI tools from a single source of business logic.

The data foundation already includes a gold-layer for reporting, a master KPI catalog, and Lakehouse/Unity Catalog as a single source of truth for over 500 KPIs, with goals to extend semantics to AI experiences.

Industry context shows strong interest in agentic AI, with surveys indicating widespread exploration and concerns about reliability, hallucinations, security, and data privacy.

Semantic layers are becoming a competitive necessity for AI platforms, with evidence that reasoning over an OSI-governed semantic layer yields higher accuracy than raw data parsing.

Genie spaces handle domain-specific questions; Agent Bricks route queries to persona-based agents (e.g., CFO, Sales VP) with Unity Catalog enforcing row- and column-level permissions, all accessible through Databricks Apps.

In collaboration with Databricks, the project scales Talk to Data by establishing a governed semantic layer that unifies BI and AI with trusted KPI definitions and explainable outputs.

The approach aims to scale self-service analytics and agentic AI by providing a trusted foundation of semantic governance, positioning it as a differentiator for enterprise AI platforms.

Early pilot results are promising, with plans to extend the model to additional markets, expand AI-driven semantics, and automate the process via an App solution; a Data + AI Summit 2026 session will explore deeper.

A five-phase governance process ensures AI answers align with Power BI reports, including onboarding KPIs, building semantic layers, organizing domains, incremental testing, and final validation for a reliable, auditable AI experience.

Korea has documented a repeatable eight-step global rollout playbook, covering data onboarding, KPI documentation, DAX-to-metric-view generation, validation, Genie space optimization, persona agent deployment, and workflow integration.

Looking ahead, the plan considers adoption by other markets, potential vendor differentiation or lock-in, governance standards for agentic AI, and whether semantic-layer explainability becomes a regulatory requirement by 2027.

An automated DAX-to-Metric-View transpiler converts Power BI DAX measures into Unity Catalog metric views, enabling AI-ready semantic models with validation and iterative refinement through Genie Code.