Building Dbt Semantic Layer

Building the dbt Semantic Layer

This skill guides the creation and modification of dbt Semantic Layer components: semantic models, entities, dimensions, and metrics.

• Semantic models - Metadata configurations that define how dbt models map to business concepts
• Entities - Keys that identify the grain of your data and enable joins between semantic models
• Dimensions - Attributes used to filter or group metrics (categorical or time-based)
• Metrics - Business calculations defined on top of semantic models (e.g., revenue, order count)

Additional Resources

• Time Spine Setup - Required for time-based metrics and aggregations
• Best Practices - Design patterns and recommendations for semantic models and metrics
• Latest Spec Authoring Guide - Full YAML reference for dbt Core 1.12+ and Fusion
• Legacy Spec Authoring Guide - Full YAML reference for dbt Core 1.6-1.11

Determine Which Spec to Use

There are two versions of the Semantic Layer YAML spec:

• Latest spec - Semantic models are configured as metadata on dbt models. Simpler authoring. Supported by dbt Core 1.12+ and Fusion.
• Legacy spec - Semantic models are defined as separate top-level resources. Uses measures as building blocks for metrics. Supported by dbt Core 1.6 through 1.11. Also supported by Core 1.12+ for backwards compatibility.

Step 1: Check for Existing Semantic Layer Config

Look for existing semantic layer configuration in the project:

• Top-level semantic_models: key in YAML files → legacy spec
• semantic_model: block nested under a model → latest spec

Step 2: Route Based on What You Found

If semantic layer already exists:

1. Determine which spec is currently in use (legacy or latest)
2. Check dbt version for compatibility:

• Legacy spec + Core 1.6-1.11 → Compatible. Use legacy spec guide.
• Legacy spec + Core 1.12+ or Fusion → Compatible, but offer to upgrade first using uvx dbt-autofix deprecations --semantic-layer or the migration guide. They don't have to upgrade; continuing with legacy is fine.
• Latest spec + Core 1.12+ or Fusion → Compatible. Use latest spec guide.
• Latest spec + Core <1.12 → Incompatible. Help them upgrade to dbt Core 1.12+.

If no semantic layer exists:

1. Core 1.12+ or Fusion → Use latest spec guide (no need to ask).
2. Core 1.6-1.11 → Ask if they want to upgrade to Core 1.12+ for the easier authoring experience. If yes, help upgrade. If no, use legacy spec guide.

Step 3: Follow the Spec-Specific Guide

Once you know which spec to use, follow the corresponding guide's implementation workflow (Steps 1-4) for all YAML authoring. The guides are self-contained with full examples.

Minimal latest spec example (dbt Core 1.12+ / Fusion) — use this as your starting point to avoid guessing the structure:

# models/fct_orders.yml
models:
  - name: fct_orders
    semantic_model:
      enabled: true
    agg_time_dimension: order_date
    columns:
      - name: order_id
        entity:
          type: primary
          name: order
      - name: customer_id
        entity:
          type: foreign
          name: customer
      - name: order_date
        granularity: day
        dimension:
          type: time
      - name: status
        dimension:
          type: categorical
    metrics:
      - name: total_revenue
        type: simple
        label: Total Revenue
        agg: sum
        expr: amount

Minimal legacy spec example (dbt Core 1.6–1.11) — use this if the project is on an older version:

# models/sem_orders.yml
semantic_models:
  - name: orders
    model: ref('fct_orders')
    defaults:
      agg_time_dimension: order_date
    entities:
      - name: order
        type: primary
        expr: order_id
    dimensions:
      - name: order_date
        type: time
        type_params:
          time_granularity: day
    measures:
      - name: revenue
        agg: sum
        expr: amount

metrics:
  - name: total_revenue
    type: simple
    label: Total Revenue
    type_params:
      measure: revenue

Entry Points

Users may ask questions related to building metrics with the semantic layer in a few different ways. Here are the common entry points to look out for:

Business Question First

When the user describes a metric or analysis need (e.g., "I need to track customer lifetime value by segment"):

1. Search project models or existing semantic models by name, description, and column names for relevant candidates
2. Present top matches with brief context (model name, description, key columns)
3. User confirms which model(s) / semantic models to build on / extend / update
4. Work backwards from users need to define entities, dimensions, and metrics

Model First

When the user specifies a model to expose (e.g., "Add semantic layer to customers model"):

1. Read the model SQL and existing YAML config
2. Identify the grain (primary key / entity)
3. Suggest dimensions based on column types and names
4. Ask what metrics the user wants to define

Both paths converge on the same implementation workflow.

Open Ended

User asks to build the semantic layer for a project or models that are not specified. ("Build the semantic layer for my project")

1. Identify high importance models in the project
2. Suggest some metrics and dimensions for those models
3. Ask the user if they want to create more metrics and dimensions or if there are any other models they want to build the semantic layer on

Metric Types

Both specs support these metric types. For YAML syntax, see the spec-specific guides.

Simple Metrics

Directly aggregate a single column expression. The most common metric type and the building block for all others.

• Latest spec: Defined under metrics: on the model with type: simple, agg, and expr
• Legacy spec: Defined as top-level metrics: referencing a measure via type_params.measure

Derived Metrics

Combine multiple metrics using a mathematical expression. Use for calculations like profit (revenue - cost) or growth rates (period-over-period with offset_window).

Cumulative Metrics

Aggregate a metric over a running window or grain-to-date period. Requires a time spine. Use for running totals, trailing windows (e.g., 7-day rolling average), or period-to-date (MTD, YTD).

Note: window and grain_to_date cannot be used together on the same cumulative metric.

Ratio Metrics

Create a ratio between two metrics (numerator / denominator). Use for conversion rates, percentages, and proportions. Both numerator and denominator can have optional filters.

Conversion Metrics

Measure how often one event leads to another for a specific entity within a time window. Use for funnel analysis (e.g., visit-to-purchase conversion rate). Supports constant_properties to ensure the same dimension value across both events.

Filtering Metrics

Filters can be added to simple metrics or metric inputs to advanced metrics. Use Jinja template syntax:

filter: |
  {{ Entity('entity_name') }} = 'value'

filter: |
  {{ Dimension('primary_entity__dimension_name') }} > 100

filter: |
  {{ TimeDimension('time_dimension', 'granularity') }} > '2026-01-01'

filter: |
  {{ Metric('metric_name', group_by=['entity_name']) }} > 100

Important: Filter expressions can only reference columns that are declared as dimensions or entities in the semantic model. Raw table columns that aren't defined as dimensions cannot be used in filters — even if they appear in a measure's expr.

External Tools