Semantic Views: Turning Business Definitions into Reusable Data Platform Objects

In the previous article,  I introduced Governed Context as the missing layer in many modern data platforms.

This time, let’s dive deeper into the topic.

Because shared business meaning or definitions cannot live only in documentation, spreadsheets, wiki pages, or team knowledge. At some point, it needs to become part of the platform itself.

That is where Semantic Views become interesting.

‍

From definitions in documents to definitions in the platform

Every organization has important business concepts: customer, revenue, order, active user, churn, margin, pipeline, retention.

The challenge is not naming them.

The challenge is making sure they are defined, owned, reused, and interpreted consistently across dashboards, reports, applications, and AI workflows.

In Snowflake, Semantic Views provide a business-friendly layer over data by defining logical tables, dimensions, facts, metrics, and relationships on top of physical tables. They are also used by Cortex Analyst to generate SQL against the underlying tables based on that semantic definition.

That matters because most business users do not think in tables, joins, and column names.

They think in business terms.

A semantic layer helps translate between both worlds, tying business language to actual data definitions without technical jargon.

And this is where the first measurable benefit appears:

Time to find the approved definition of a metric should move from hours or days to minutes : ideally close to 0 for the most important KPIs.

If “revenue,” “margin,” or “active customer” is already defined, approved, and exposed through a semantic layer, teams should not need another alignment meeting just to understand what the number means.

‍

What Semantic Views actually contain

‍

A useful Semantic View is not just another technical view with a nicer name.

It describes the business-facing model of the data: logical entities, dimensions, facts, metrics, and relationships. Descriptions, synonyms, verified queries, and custom instructions are elements that can help Cortex Analyst better understand and query the data.

This gives teams a place to answer questions like:

• What is the approved formula for revenue?

• Which join path should be used between customers and orders?

• Which metric should be exposed to business users?

• Which fields are public, restricted, or not relevant for self-service?

• Which examples should guide natural language questions?

‍

‍

‍

In other words I would say:

If a KPI is business-critical, it should not be redefined inside a dashboard.

That does not mean dashboards are not important. They are.

But the logic behind critical metrics should be governed, reusable, and visible beyond one report, one team, or one tool.

The key shift in thinking is simple: :

Business logic moves from being repeatedly rebuilt in many tools to being defined once and reused many times.

A practical target could be:

80% of core business reporting should reuse approved semantic definitions instead of recreating logic inside individual dashboards, SQL queries, or spreadsheets.

Not every calculation needs to be centralized from day one. But the key metrics that drive management reporting, operations, sales, finance, and customer analytics should not exist in five slightly different versions. The critical part is consistency at every stage - from foundational data pipelines through business-ready data structures, all the way up to the tools used for consumption.

‍

Why this matters for Governed Context

Governed Context is the broader idea.

Semantic Views are one of the practical mechanisms that can make it real.

Without this layer, every BI report, analyst notebook, application, and AI assistant may carry its own version of the truth. Even small differences in filters, joins, or aggregation logic can create conflicting numbers. Worse, if those numbers are close enough, they don’t raise any alarms, and discrepancies may go unnoticed until damage has already been done.

With a governed semantic layer, teams can start moving toward more measurable outcomes:

• 1 approved definition for each critical metric

• 0 ambiguity on which metric should be used for official reporting

• Minutes, not days, to identify the correct metric logic

• Fewer duplicate SQL patterns across dashboards and reports

• Higher reuse of semantic objects across BI, applications, and AI workflows

‍

This is especially important for AI-powered workflows, as AI does not only need access to data(and clean data at it) - It needs access to the right business meaning.

Semantic Views are the recommended approach for new Cortex Analyst implementations, while legacy YAML semantic models remain there, supported for backward compatibility.

If an AI assistant can query tables but does not understand which revenue definition is approved, which customer segment is valid, or which metric version is official, the answer may be technically correct but business-wise wrong.

A strong semantic layer helps reduce this risk.

A good target is simple:

For AI-assisted analytics, 100% of business-critical questions should be routed through governed definitions, not raw interpretation of tables.

That does not mean every exploratory question needs a fully governed metric. But when AI supports business decisions, the semantic foundation needs to be trusted. Building on that foundation allows AI to stay grounded in well-defined, known metrics and context - while still giving exploratory questions a better chance of producing meaningful answers.

‍

What to watch out for

Semantic Views will not solve poor governance by themselves.

They still require clear ownership, review processes, naming standards, and agreement between business and data teams. A semantic layer without accountability can simply become another place where definitions drift.

The implementation question is not only:

“How do we create Semantic Views?”

It is also:

• “Who approves definitions?”

• “Who owns metric changes?”

• “How do we avoid duplicate concepts?”

• “How do we measure reuse?”

• “How do we know when a metric is trusted enough for AI or self-service?”

This is where governance should become measurable.

For example:

• Each critical metric should have 1 named business owner

• Each approved metric should have 1 technical owner

• Metric changes should go through a defined review path

• Duplicate metric definitions should be tracked and reduced over time

• Semantic object reuse should become a platform KPI

The goal is not to centralize everything for the sake of control.

The goal is to make trusted definitions easy to find, easy to reuse, and hard to accidentally duplicate.

That is when Semantic Views stop being just a technical capability and become part of how the organization scales analytics.

‍

My Perspective

In my experience, data governance has often been a somewhat orphaned part of data platforms.

At the beginning, teams usually value freedom: build what you need, move fast, experiment, and solve local problems. At a small scale, this works well enough. But as the organization grows, and the business grows with it, the need for governance starts to appear - even if it is not immediately visible.

After enough time passes, many organizations find themselves in a familiar position: the platform is there, the data is available, and the data quality may even be good. But nobody fully controls what is used where, which logic is duplicated, and which definition is the approved one.

The result is tens of siloed data flows. Some are redundant. Some carry ambiguous definitions. Some produce numbers that look correct but mean slightly different things.

And that leads to a complete lack of trust from business users.

With AI emerging as quickly as it is, proper governance at the semantic level becomes a first-order priority. Before organizations can truly benefit from AI workflows, agents, and natural language analytics, they need confidence that the business meaning behind the data is clear, governed, and reusable.

In the next article, I will move from Semantic Views to trusted metrics :how organizations can reduce KPI drift and create more confidence in reporting, self-service analytics, and AI-assisted decision-making.

‍

‍

‍