Data Trends>Data Modeling

What Is a Semantic Model? Definition, Examples, and How to Build One

key-takeawayKey Takeaways

  • A semantic model makes analytical concepts such as entities, dimensions, measures, metrics, and relationships reusable across business questions.
  • It sits above source structures, translating technical fields and joins into terms your business uses, while data values can remain in the source.
  • A data model describes structure, while a semantic model describes business meaning. A semantic layer can serve those definitions to downstream queries.
  • Shared definitions offer a common basis for analytics and AI when people and tools use them consistently. They do not guarantee data quality, adoption, or access control.
  • Start with disputed, high-value metrics. Then document grain, logic, ownership, tests, and a process for managing changes.

A semantic model gives you and your analytical tools a shared way to work with business concepts such as revenue, customer, and churn. Rather than interpret raw tables and ambiguous column names each time, users and AI agents can apply documented, reusable meaning.

For a data leader, the practical value is straightforward: business logic becomes easier to find, review, apply, and update. A semantic model does not replace your warehouse, pipelines, catalog, quality program, or access controls; it connects those systems to the questions people ask.

What Is a Semantic Model?

In common analytics usage, a semantic model describes reusable analytical concepts on top of source data. These concepts may include business entities, dimensions, measures, metrics, relationships, synonyms, and calculation rules.

The model creates a business-facing view of technical structures. A field such as `net_amount` becomes net revenue, for example, with documented rules covering refunds, eligible transactions, currency treatment, and reporting periods.

Implementation details vary by platform. In dbt Labs' dbt semantic model documentation, entities, dimensions, and measures make up a project's semantic layer.

ThoughtSpot Models contain structure and business logic rather than the underlying data values. Those values remain at the source, which lets you maintain live connections while managing business meaning.

A Plain-English Semantic Model Example

Suppose three executives ask for net revenue. Your warehouse may hold orders, refunds, subscriptions, currencies, and customer records in separate tables, each with its own grain and join paths.

A semantic model can document how those pieces work together:

  • Entity: Customer, order, subscription, and product.

  • Measure: Gross order amount and refund amount.

  • Metric: Net revenue equals eligible revenue minus recognized refunds.

  • Time rule: Use the recognized revenue date for monthly reporting.

  • Dimensions: Region, product, customer segment, and sales channel.

  • Exclusions: Tests, canceled orders, and other agreed record types.

When connected to the model, a dashboard, analyst, or AI agent can use the same logic. The model still needs valid source data, correct joins, downstream adoption, and review by a business owner.

An analytics and modeling workspace can give analysts a place to examine the data, develop logic, and validate modeled definitions before wider use.

Semantic Model vs. Data Model vs. Semantic Layer

These terms often overlap, and vendors do not always use them the same way. The comparison below is a practical working distinction rather than a universal taxonomy.

ConceptPrimary QuestionTypical ContentsRole in Analytics
Data modelHow is data structured and stored?Tables, columns, keys, schemas, cardinality, and physical relationshipsOrganizes data for storage, processing, and retrieval
Semantic modelWhat does the data mean for the business?Entities, dimensions, measures, metrics, business terms, calculation logic, and analytical relationshipsMakes business meaning reusable across analytical questions
Semantic layerHow are modeled definitions made available?Model management, query interfaces, APIs, integrations, and serving logicCan expose definitions to BI tools, applications, and AI agents

A warehouse data model might show that an order table joins to a refund table through `order_id`. The semantic model adds the business decisions: which refund states count, which date governs recognition, and how the calculation behaves across regions.

A semantic layer may then make the modeled metric available for queries. As one implementation example, dbt Labs documents how users can query metrics and dimensions from its Semantic Layer with the MetricFlow engine.

ThoughtSpot separates modeling semantics from a larger context capability. Its agentic semantic layer covers business logic, terms, synonyms, metrics, and related rules. The ThoughtSpot Context Layer assembles relevant meaning and history for a person or AI agent.

What Belongs in a Semantic Model?

The exact contents depend on the platform and use case, but several elements appear often in analytics implementations. Each one helps answer a different part of the business question.

ComponentWhat It DescribesNet Revenue Example
EntitiesBusiness objects with identityCustomer, order, subscription
DimensionsAttributes used to group or filterRegion, plan, segment, month
MeasuresAggregatable numeric valuesOrder amount, refund amount
Derived metricsBusiness calculations built from measuresNet revenue, refund rate
RelationshipsValid paths between entities or tablesCustomer to order, order to refund
Synonyms and business termsLanguage people use for the same conceptSales, booked revenue, net sales
Calculation logicFilters, exclusions, time rules, and formulasExclude tests and canceled orders

Grain deserves special attention because it states what one row represents. An order may have several refund rows, so a careless join can multiply revenue before the metric calculation begins.

Relationships matter for the same reason. Your model should document valid join paths, cardinality, and when a relationship applies, so reviewers can see how an answer was assembled.

Some platforms may integrate lineage, quality tests, or access controls with semantic modeling. Those capabilities still require explicit configuration and an operating process; adding a metric definition does not create a complete governance or security program.

When AI is part of your plan, names and formulas are only the starting point. Business terminology, relationship context, clear descriptions, and reviewable logic help an AI analyst for enterprise analytics interpret a question against the definitions you have approved.

The Net Revenue Example, From Raw Data to Business Logic

A useful net revenue definition begins with grain, not a formula. Decide whether the metric is calculated per order line, order, invoice, subscription period, or another unit before you connect tables.

An illustrative definition might read:

  • Start with recognized revenue from eligible orders and subscriptions.

  • Subtract refunds that reached an approved status within the reporting policy.

  • Exclude test records, canceled orders, and internal transactions.

  • Convert currencies under the finance team's approved rate and date rule.

  • Group results by the recognized revenue date, not the order creation date.

  • Join customers through the approved customer key and relationship path.

This pseudo-definition is not production code. Your finance owner still needs to confirm eligibility, exclusions, currency treatment, time windows, and late-arriving adjustments before analysts use the metric.

The process also needs a named owner who can answer disputes and approve changes. Those ownership decisions are part of the data leadership priorities surrounding model quality, adoption, and governance.

Why Semantic Models Matter for Analytics and AI

A shared model can give analysts, dashboards, applications, and AI agents a common basis for interpreting a business question. That benefit appears only when those consumers use the same definitions and the organization maintains them.

For traditional BI, reusable definitions reduce the need to rebuild the same calculation inside every report. For AI, the need goes further because a plausible SQL query can still apply the wrong revenue definition, join path, time rule, or customer meaning.

AI agents need business context they can apply and explain, not only a route from natural language to SQL. Reviewable definitions help your data team inspect each metric, relationship, and filter. You can then correct logic that does not match the business rule.

In ThoughtSpot, Spotter converts natural-language questions into search tokens tied to definitions for tables, joins, logic, and metadata in the semantic layer. Those tokens and natural-language explanations give users a visible path for checking how Spotter interpreted a question.

That traceability supports human review; it does not mean every answer from every semantic model is automatically correct. Source quality, model logic, permissions, ownership, and the wording of the question still matter.

The ThoughtSpot Context Layer adds context around when definitions apply and how concepts relate. For developers bringing governed analytics into other AI interfaces, ThoughtSpot's agentic MCP server provides a bridge between external AI agents and ThoughtSpot's analytics capabilities.

How to Build and Operate a Semantic Model

You do not need to model the whole company before delivering value. Start with a small business domain where metric disputes create visible friction, then expand after the definitions hold up under real questions.

A practical five-step path looks like this:

  1. Choose high-friction questions. Identify one or two measures that appear often in executive reviews, operating meetings, or analyst requests.

  2. Map the analytical structure. Document entities, grain, dimensions, relationships, and approved join paths for each question.

  3. Define logic with business owners. Write calculations, filters, exclusions, time rules, and terminology in language both technical and business reviewers understand.

  4. Test against trusted outputs. Compare results with approved reports, investigate differences, and record why the new definition matches or replaces prior logic.

  5. Assign ownership and manage changes. Name technical and business owners, define review steps, version changes, and communicate downstream impact.

Governance here is an operating model. It combines shared definitions with documented ownership and change management, while data-access controls remain separately configured and tested.

Use this first-model checklist before publishing definitions:

  • The business question and intended users are clear.

  • The grain is documented for every source and measure.

  • Entities, dimensions, and relationship paths are named.

  • Filters, exclusions, time rules, and currency rules are explicit.

  • Business terms and synonyms match the language users ask with.

  • A business owner has approved the metric definition.

  • Results match reviewed outputs or explain documented differences.

  • Access controls have been configured and tested separately.

  • Owners, version history, and change notifications are in place.

Start With the Metrics People Argue About

Revenue, active customer, and churn are good starting candidates because disagreements often expose hidden choices. Before modeling one, ask about its grain, included records, time window, currency policy, and relationship path.

Then ask who owns the answer. A metric without a business owner may be technically sound yet remain disputed when finance, sales, and operations use different policies.

ThoughtSpot Model lets teams define business logic, combine data with joins, create calculated metrics, and add synonyms.

Treat modeling work as a draft for review. Your technical owners should inspect joins and grain, while business owners validate calculation choices and terminology before the model reaches dashboards or AI agents.

Put Business Meaning Where Decisions Happen

Semantic modeling is a practical discipline for making important business definitions reusable. It connects stored data, analytical logic, and business language to the questions your users bring to analytics and AI.

Start by testing whether your most important metrics have a clear grain, approved relationships, understood terminology, and named owners. If you want to put that business meaning closer to the people and AI agents making decisions, explore ThoughtSpot.

The Leader in Agentic Analytics

2026 Leader

2026 Leader

2026 G2 Enterprise

2026 G2 Enterprise

2026 G2 SMB

2026 G2 SMB

2026 G2 Momentum

2026 G2 Momentum