How to calculate growth % vs prior period in dbt

Growth compares a period with an earlier one: (this period − earlier period) / earlier period. As a derived metric it uses the same input metric twice, once as it is and once shifted with offset_window.

Copy and paste

metrics:
  - name: revenue_mom_growth
    label: Revenue growth vs last month
    type: derived
    expr: (revenue - revenue_lm) / revenue_lm
    input_metrics:
      - name: revenue
      - name: revenue
        alias: revenue_lm
        offset_window: 1 month

  - name: revenue_yoy_growth
    label: Revenue growth vs same month last year
    type: derived
    expr: (revenue - revenue_ly) / revenue_ly
    input_metrics:
      - name: revenue
      - name: revenue
        alias: revenue_ly
        offset_window: 1 year

mf query --metrics revenue,revenue_mom_growth,revenue_yoy_growth --group-by metric_time__month --start-time 2025-12-01

Example

The results are ratios: 0.1 is 10%. Format them as percentages in the BI tool rather than multiplying by 100 in expr, so that later calculations keep the precision.

December 2025 has no year-over-year value because the data starts in January 2025. May 2026 is partial in real life; month-over-month and year-over-year growth for the current month compare a few days with a full month, so show them for complete months only, or use YTD growth for the current period.

When the earlier period is empty or zero

• No data in the earlier period. September 2025 has no month-over-month value: August had no orders, so revenue_lm is NULL and the division returns NULL. That is the honest answer: growth from nothing is not a percentage.
• A zero base. If the earlier value can be exactly 0 (for example after fill_nulls_with: 0), some warehouses raise a division-by-zero error and others return NULL or infinity. Guard it in expr:

    expr: (revenue - revenue_lm) / nullif(revenue_lm, 0)

The rules of offsets

• The aliased input is the one that is shifted; expr refers to aliases, so the same metric can appear twice.
• The query must group by metric_time at some grain; without it MetricFlow refuses offset metrics.
• The shift happens on days, then the result is aggregated to the query’s grain. A 1-month offset of daily data grouped by month compares calendar months, whatever their length.

For the absolute value of last year alongside this year’s, see how to compare a metric with the prior year.

Tested with dbt Core 1.12 and MetricFlow 0.213 on DuckDB.