Raw data is a record of what happened, stripped of interpretation. A row in an orders table might read order_id: 88213, status: 4, amt: 240.00. Every value is accurate. None of it tells you whether that order counts toward this quarter’s revenue, whether status: 4 means shipped or refunded, or whether amt includes tax. Meaning isn’t in the row. It’s in the rules your business applies to the row — and those rules usually live in people’s heads.

From tables and columns to business terms

Semantic modeling is the bridge from the physical world of tables and columns to the conceptual world of business terms. Instead of asking an analyst to remember that status = 4 means a completed sale, the model records it: this entity is an order, this status means completed, and completed orders count toward net revenue unless later returned.

The warehouse underneath still stores everything the same way (Google Cloud). What changes is that a layer of meaning now sits between the storage and the people asking questions.

The vocabulary of a semantic model

Semantic models are built from a consistent set of concepts. Snowflake’s semantic views organize data into logical tables for business entities, dimensions for categorical context, facts for quantitative detail, metrics for aggregated KPIs, and relationships that define how everything joins (Snowflake Documentation). Translated into everyday terms: the things you track, the ways you slice them, the numbers you calculate, and the connections between them. Each of these carries a definition, and each definition is where business context enters the data.

A hypothetical example: five words, five definitions

Imagine a subscription software company — this scenario is illustrative, not a real customer. Five common terms cause five arguments:

• Customer. Sales counts every signed contract. Finance counts only accounts with an active paid plan. Support counts every login with a ticket.
• Active customer. Marketing means anyone who opened an email this month. Product means anyone who used the app in the last 30 days. Finance means anyone currently billed.
• Revenue. Does it include one-time setup fees? Refunds? Revenue from a legacy billing system that’s being retired?
• Gross margin. Which costs count — hosting only, or hosting plus support headcount?
• Churn. Is it measured by logo (accounts lost) or by dollars (revenue lost)? Counted at contract end or at non-renewal?