Semantic Models
Intermediate 13 min read Semantic Models

Semantic Measures and Metrics

A measure is syntax; a metric is a contract

Measure = an aggregation expression; metric = the named, documented, testable business quantity built from measures and dims. Learn the vocabulary that makes a layer rigorous.

measuresmetricsagg exprdimensionmetric definition
Prerequisites: The Semantic Layer Overview
Before we start

What you will learn

  • Distinguish measures from metrics and dims
  • Write a metrics definition with filters, time and format
  • Test a metric definition like code
Introduction

What is this?

In a semantic layer, a measure is a reusable aggregation expression (SUM(amount), COUNT(DISTINCT id)) computed over a model. A metric is a named, typed, documented business quantity — Revenue, Orders, Active Customers — assembled from measures, dims, optional filters and time grains, with owners and tests. Measures are pieces; metrics are the contract consumers sign.

Motivation

Why do you need it?

  • The distinction keeps layers composable and reviewable.
  • Named metrics with owners are governable; ad-hoc expressions aren't.
  • Time grain, filters and defaults are part of a metric's definition—not an afterthought bolt-on.
  • Interviewers probe exactly this vocabulary at the senior level.
Think about it

Mental model

Imagine

Kitchen staples vs menu dishes

Think of

Measures vs metrics

Flour, eggs and stock are staples — reusable pieces (measures). 'Coq au Vin' is a named, plated dish with a recipe (metric): fixed composition, name, portion, cost. You cook staples into dishes; you compose measures into metrics.

Mechanics

How it works

  • Measures: aggregation + target column (+ optional filter) — building blocks.
  • Metrics: name, description, measure, grain/time, optional filters and owners.
  • Dims attach: every metric slices by the layer's conformed dims.
  • Formatting, rounding and defaults are metadata on the metric.
  • Tests: metric-level checks (values in range, totals reconcile, freshness) run in CI.

From measure to governed metric.

Column
Measure
Metric
Wrapped
Consumed
Hands-on

Examples

Measure vs metric in a sessionable format

Measure vs metric in a sessionable format
-- MEASURE (piece):
measure: amount_sum {
  aggregation: SUM
  expr: fact_sales.amount
  labels: ["total_amount"]
}
 
-- METRIC (contract):
metric: revenue {
  description: "Gross revenue of shipped orders"
  type: number
  label: "Revenue"
  name: revenue
  type_params:
    measure: amount_sum
    filters:
      - expr: fact_sales.status = 'shipped'
  owner: finance
  tags: [kpi, certified]
}

The measure defines how to aggregate; the metric defines what it means, how to filter, who owns it and which tags govern it.

Real world

In production

  • A finance 'Revenue' metric filters to shipped orders; every report inherits the filter.
  • An 'Active Customers' metric counts distinct customers per date with a 90-day lookback.
  • Formatting metadata ('Currency USD, 2dp') ships with the metric so every export matches.
Guidelines

Best practices

  • Write metrics as named, owned, documented objects — never bare expressions.
  • Encode business rules (filters, time grain) inside the metric, not in reports.
  • Tag certified vs experiment metrics so consumers know trust level.
  • Test each metric against a source control total in CI.
Watch out

Common mistakes

  • Renaming a raw measure and calling it a metric — no contract.
  • Filters living in reports instead of the metric definition.
  • No owner/tag — a metric nobody owns gets changed by everyone.
  • Typing everything as floats without format, so exports drift.
When things break

Debugging tips

  • A metric differs between consumers → someone's applying extra filters; check both queries' filter provenance.
  • Wrong totals → inspect the measure aggregation and its filter set.
  • Freshness silent → add the metric-level freshness test.
Interviews

Interview checkpoint

Test yourself

Quick quiz

0/2
1

Which is the named, documented, owned business quantity?

2

Business filters belong inside the metric definition, not in every report.

Go deeper

Official documentation

dbt MetricFlow - metrics and measures
docs