Why Your Semantic Views Need Process Context (And How to Build Them)

Building accurate AI agents on ERP data requires embedding business process context directly into semantic views. Without it, agents see raw column names like SAP's 'F0911' or 'A FIH' with no understanding of what they represent. A business-first approach inverts the typical workflow: define domain, process, and governance before connecting source data, so naming, metrics, and relationships reflect how the business actually operates. The resulting semantic view DDL includes structured table comments with grain and process metadata, explicit join relationships, governed metrics with synonyms, verified queries, and custom instructions (e.g., fiscal calendar rules). This context layer — rather than the model itself — determines answer accuracy. A metrics library ensures consistent calculations across multiple semantic views, eliminating competing definitions. The approach is illustrated with a SAP-sourced 'procure-to-pay' example and references a commercial product called Data Product Studio.