Semantic Layer as Metrics Hub
George Xing, co-founder and CEO of Supergrain, on the future of business intelligence
BI is fragmenting into specialized tools, which makes the control point shift from dashboards to the shared metric definitions underneath them. Product teams need exploratory charts, finance needs planning models, and operators need workflows tied to live systems, so no single interface fits all three. What scales is a common semantic layer that lets each tool query the same revenue, conversion, and retention logic without redefining it every time.
-
The old all-in-one stack already broke apart once. Early BI suites bundled storage, modeling, and dashboards, then Snowflake and BigQuery pulled storage out, and tools like Looker and Tableau became separate front ends. This made specialization normal, not exceptional.
-
The real failure mode is not too many tools, it is too many metric definitions. When each dashboard, notebook, or planning model writes its own SQL, revenue can mean one thing in finance and another in product. A semantic layer fixes that by keeping one reusable definition and letting many apps consume it.
-
This is why warehouses do have an incentive to move upward, but not necessarily to become the only BI product. Warehouses make money when more applications run queries on top of them, while workflow tools like dbt and semantic layer vendors focus on making those definitions usable across many downstream surfaces.
The market is heading toward a hub and spoke model. The warehouse remains the system of record, a semantic layer becomes the system of meaning, and many purpose built apps sit on top. The winners will be the companies that become the default place to define metrics once and distribute them everywhere.