Skip to content

Glossary

Analytics & BI

Semantic Layer

A semantic layer is a shared model between raw data and the tools people use, defining business metrics, dimensions and joins once so that every dashboard, spreadsheet and AI assistant calculates a term like revenue the same way.

Updated · By the VIZIO AI team

What is a semantic layer?

A semantic layer turns table and column names into business terms, and fixes the math behind each metric in one place. Warehouse tables hold fields like ord_amt and status_cd. The semantic layer says that net revenue is the sum of order amounts minus refunds, for completed orders.

It also defines dimensions, such as region, product and month, and how tables join. BI tools, notebooks and apps then ask for net revenue by region instead of writing their own SQL.

This is how a team keeps a single source of truth for its metrics when many tools read the same data.

What goes into a semantic layer, and which tools provide one?

A semantic layer has four parts: entities and joins, dimensions, measures, and access rules.

  • Entities and joins: orders link to customers through customer_id.

  • Dimensions: fields you group and filter by, such as date or channel.

  • Measures and metrics: sums, counts and ratios, with their filters built in.

  • Access rules: who may see which rows or fields, often with row-level security.

The dbt Semantic Layer, powered by MetricFlow, defines metrics inside a dbt project; dbt lists it for its Starter, Enterprise and Enterprise+ plans. Cube offers an open-source semantic layer, Cube Core, with SQL, REST and GraphQL APIs. LookML is Looker's language for defining fields, calculations and joins. Power BI semantic models do the same job inside Microsoft's stack.

What does a semantic layer look like in practice?

A semantic layer project usually starts with the five metrics people argue about most. Picture a meal-kit subscription company where finance, marketing and the CEO each report a different monthly revenue.

  1. The analytics lead lists every definition in use. Finance excludes refunds, marketing counts gross order value, and the CEO dashboard includes tax.

  2. The leaders agree on one net revenue and one gross revenue definition, each with a short description.

  3. The team writes both as metrics in the dbt Semantic Layer, with order date, channel and plan as dimensions.

  4. Dashboards and the weekly spreadsheet switch to the shared metrics. Old calculated fields are deleted.

Say the Monday meeting now spends its time on why revenue moved, not on whose number is right. New metrics go through the same review.

What are the limits of a semantic layer?

A semantic layer fixes inconsistent math, not bad source data or unclear ownership.

  • Bad inputs: if order status is wrong in the source system, every tool is now wrong in the same way. Pair the layer with data quality checks.

  • Ownership: each metric needs a named owner who approves changes.

  • Tool support: not every BI tool connects to every semantic layer. Check the integration list before you choose.

  • Lock-in: definitions written in one vendor's language take work to move later.

  • Over-modeling: trying to model every column on day one delays launch for months.

How is a semantic layer different from a data warehouse?

A data warehouse stores the data; a semantic layer describes what it means and how to calculate from it. A data warehouse holds cleaned tables, and a data mart is a subset of them for one team. Neither says how revenue is calculated.

Calculated fields inside one BI tool only work in that tool. A shared semantic layer serves many tools from one definition, which is what makes self-service analytics consistent.

The layer also matters for AI. When an assistant answers "what was revenue last month?", governed metric definitions give a large language model tested logic to use instead of guessing from raw tables. Cube, for example, documents a Meta API that lets AI agents discover what can be queried.

Semantic layer vs metrics defined in each BI tool

Semantic layer

Metrics in each BI tool

Where metrics live

One shared model

Inside each report or tool

Consistency across tools

Same result everywhere

Varies by tool and author

Changing a definition

Edit once, review, deploy

Edit every report that uses it

Setup effort

Higher at the start

Low at the start

Examples

dbt Semantic Layer, Cube, LookML

Calculated fields in a single dashboard

How does Vizio AI approach semantic layers?

Vizio AI's published work includes a shared KPI model that does the job of a semantic layer for one client's dashboards. For a US-based women's fashion e-commerce brand, DX Studio built a real-time fashion analytics dashboard. Marketing, sales and pricing data were centralized through Precog, and inconsistent formats were standardized into a shared KPI model so that KPIs stayed comparable across marketing, finance and merchandising.

The case study lists "Improved data accuracy by 28%" and "Enabled 32% faster operational decision-making". Where to start: write down your five most disputed metrics and the formula each team uses today.

logo of Facebook
logo of X
logo of LinkedIn
logo of Reddit
logo of Telegram

Think AI-first.
Move future-fast.

Let’s build what’s next! From product ideas and intelligent workflows to data-driven systems and scalable growth operations.

Book a Strategy Call