What Are Metrics?
Numbers with owners, definitions and decisions attached
A metric is a quantified measure of business behavior — like revenue or retention — with a precise definition someone can audit. Metrics become KPIs when they are tied to a target that drives a decision.
What you will learn
- Distinguish a metric from a KPI and from raw data
- Write a precise, auditable metric definition
- Explain why the same name can mean different numbers
What is this?
A metric is a quantity computed from data that measures some business behavior — revenue, churn, order-to-delivery time. It is only trustworthy when its definition is exact: the numerator, denominator, filters and grain. A KPI is a metric promoted to strategic importance by tying it to a target and a decision owner.
Why do you need it?
- Undefined metrics silently diverge — two teams compute 'active users' differently and fight about it.
- Written definitions let anyone reproduce the number and catch definition drift.
- Metrics are the vocabulary of the business; consistent vocabulary means productive meetings.
- Only with solid metrics can dashboards, forecasts and incentives be trusted.
Mental model
A recipe for a signature dish
A metric definition
The name 'grandma's gravy' means nothing without the recipe. Everyone who follows the recipe gets the same gravy; change an ingredient and you've changed the dish. Metrics are recipes, and the BI layer is the cookbook.
How it works
- A metric is computed from facts (measurements) filtered by a defined scope.
- The definition specifies measure, dimensions, filters, grain and as-of time.
- It lives once in the semantic layer so it is computed identically everywhere.
- A KPI adds comparison to a target or threshold over a cadence, and an owner.
- Trending and variance views turn the metric into a decision input.
From raw data to a decision-ready KPI.
In production
- 'Monthly Recurring Revenue' is defined as the sum of active subscription amounts as of month-end, excluding trials; finance, sales and exec dashboards read the same measure.
- 'Churn' is defined as customers who did not renew in the month divided by customers due for renewal that month — not a metric anyone can re-derive differently.
- An ops KPI 'On-time delivery' pairs a metric (deliveries within SLA / total deliveries) with a 95% target reviewed weekly by the ops director.
Best practices
- Write every metric as formula + named scope + grain + as-of rule.
- Give each metric one owner in the catalog with a documented definition.
- Keep the definition in the semantic/model layer, never in the report.
- Validate new metrics against a hand-computed sample before publishing.
Common mistakes
- Naming a metric differently per team and assuming the numbers align.
- Omitting the filter ('includes refunds?') and letting consumers guess.
- Creating the metric in the dashboard so the next dashboard forks it.
- Calling every number on a report a KPI — diluting the strategic ones.
Common issues
Finance and marketing report different 'revenue'Each computed the metric with different filters or timing
Define revenue once (scope and as-of rule) in the semantic layer and retire the others.
A metric changed overnight with no deployDefinition drift — someone edited the query or a filter changed
Version metric definitions and diff them on changes; alert on definition edits.
Debugging tips
- If two dashboards disagree, diff the metric definitions first — filters are the usual culprit.
- Ask 'canceled orders included?' and 'partial periods included?' to smoke out a fuzzy definition.
- Check grain: daily vs monthly vs as-of-date explains most 'mystery' differences.
Interview checkpoint
Hands-on lab
- Turn raw events into a precisely defined metric
- Show how a filter choice changes the value
- Prove the definition is reproducible by anyone
Write an Auditable Metric Definition
30 min · beginner
Quick quiz
Which component is the most common source of two teams disagreeing on a metric?
A KPI is exactly the same thing as any other dashboard number.