Skip to main content

Mart Layer

The mart layer transforms canonical models into specific use cases, such as dashboards, reports, or business unit requirements. These models are designed to serve particular business needs while maintaining consistency with the canonical layer.

Example Folder Structure​

models
└── marts
    ├── finance
    │   ├── revenue_dashboard/
    │   │   ├── daily_revenue.sql
    │   │   └── customer_ltv.sql
    │   └── reporting/
    │       ├── monthly_p_and_l.sql
    │       └── cash_flow.sql
    ├── marketing
    │   ├── campaign_performance/
    │   │   ├── campaign_roi.sql
    │   │   └── channel_metrics.sql
    │   └── customer_analytics/
    │       ├── cohort_analysis.sql
    │       └── customer_segments.sql
    └── sales
        └── sales_reporting/
            ├── sales_targets.sql
            └── team_performance.sql

Structure Best Practices​

  • ✅ Business Unit Grouping ~ Organize by department or function
  • ✅ Use Case Subfolders ~ Group related models by their purpose
  • ✅ Clear Naming ~ Names should reflect the business purpose

Example Model​

-- models/marts/finance/revenue_dashboard/daily_revenue.sql
WITH customer_orders AS (
    SELECT * FROM {{ ref('customers') }}  -- from canonical layer
),

order_data AS (
    SELECT * FROM {{ ref('orders') }}     -- from canonical layer
),

final AS (
    SELECT 
        date_trunc('day', order_date) as date,
        customer_segment,
        count(distinct customer_id) as unique_customers,
        sum(order_value) as daily_revenue,
        sum(order_value) / count(distinct customer_id) as average_order_value
    FROM order_data o
    JOIN customer_orders c 
        ON o.customer_id = c.customer_id
    GROUP BY 1, 2
)

SELECT * FROM final

Materialization​

Marts should typically be materialized as tables:

# dbt_project.yml
models:
  your_project:
    marts:
      +materialized: table
Why Tables?
  • Optimized for end-user query performance
  • Refreshed on a schedule matching business needs
  • Complex calculations already processed
  • Ready for BI tool consumption