Skip to content
Concept Draft

Measure, metric and KPI

Two dashboards, one question

The monthly review opens with two dashboards that disagree about revenue. Both are, in their own terms, correct. One excludes cancelled orders. One converts currency at month end. One has the growth target built into the thresholds that turn a tile red.

Each report has fused four different things into a single formula: what the number means, how it is calculated, what it is judged against and how it is shown. Once they are fused, nobody can say which of the four is the difference.

Arguments about numbers become arguments about reports

  • Every conversation about the figure turns into forensic comparison of report logic.
  • Targets leak into definitions. Moving a threshold quietly changes what the number appears to mean.
  • Each new consumption tool (another BI tool, a query engine, an AI assistant) becomes one more independent place where the number is defined.
  • Replacing a tool means rediscovering business logic that only ever lived in its formulas.

Four concerns, four authorities

Separate meaning (metric), deterministic evaluation (measure and the analytical model), performance intent (KPI) and representation (presentation). Each has its own authority.

  • Measure: a quantitative calculation over business facts at a known grain, with an explicit unit, valid dimensions, time behaviour and aggregation rules (additive, semi-additive, non-additive, distinct count). It is the evaluation primitive.
  • Metric: the governed meaning of a business quantity: its identity, scope, owner and version. It may be realised by one measure, several measures, other metrics or an expression over facts.
  • KPI: a metric, or governed composition of metrics, interpreted against an explicit objective, target or threshold. The defining addition is performance intent, not importance or mathematical complexity.
  • Presentation: how a particular consumer shows the result. It does not become part of the definition because it happens to be implemented in a report.

The picture

Four concerns, four authorities Four stacked bands over business facts. Measure: how it is evaluated, decided by the analytical model, for example the sum of order value at order-line grain. Metric: what it means, decided by the semantic owner, for example gross revenue. KPI: what it is judged against, decided by the performance owner, for example growth against an 8 per cent target. Presentation: how it is shown, decided by each consumer, for example a red, amber or green tile. Concern Decided by Example: gross revenue Presentation how it is shown Each consumer Red / amber / green tile KPI what it is judged against Performance owner Growth against an 8% target Metric what it means Semantic owner Gross revenue Measure how it is evaluated Analytical model Sum of order value, order-line grain Business facts from source-aligned data products
Logical distinctions of authority, not mandatory physical layers. A single query may evaluate several of them together.

Gross revenue and its target

Illustrative example

A finance function owns what gross revenue means: which order lines count, when revenue is recognised and how currency is handled. A commercial function sets this year's growth target against it.

The target lives in the KPI definition, not in the metric. When the commercial function raises the target mid-year from 6% to 8%, the meaning of gross revenue does not change, finance approves nothing, and every historical report of gross revenue still reconciles. Each dashboard remains free to choose its own colours.

In detail

Measure and metric are not stages

A measure and a metric can exist for the same quantity at the same time: a measure inside an analytical model and, where the meaning must be authoritative and could be disputed, a governed metric identity over it. Nothing about being "just" a measure prevents it from also needing governed meaning.

A KPI introduces a second authority

Semantic authority governs what a metric means. Performance authority governs the objective set against it. They may be the same party or different ones, but the target must never be placed inside the metric's definition.

Authority is not materialisation

Making a metric's definition authoritative does not require materialising it. Caching, a query language, a BI tool's formula language or a particular engine are implementation choices. The data product that supplies the underlying facts does not, by supplying them, become the authority for the metric's meaning.

When a number needs this treatment

Apply the shared-understanding test: if two people, systems or purposes could read the term differently in a way that matters, it needs a governed identity, scope or explicit variants. Revenue can be gross or net, booked or settled. A metric that is genuinely specific to one use can stay local.

Not this

  • Not a mandate for four physical layers or four tools.
  • Not a requirement to govern every number anyone calculates.
  • Not a choice of product. Any tool may evaluate or display these concerns; none of them owns the meaning.

Unresolved

  • How should measures, metrics, KPIs and presentation-specific calculations differ in lifecycle and versioning?
  • How should semi-additive and non-additive quantities be classified, and how should dependencies between metrics be resolved?

Join the argument

Agree, challenge or add evidence

Architecture improves when its assumptions are challenged. Say which kind of contribution you're making:

  • Challenge I disagree because…
  • Evidence We've seen this too…
  • Question How would this work when…?
  • Alternative Another way to approach this…
  • Extension This also implies…

Please keep clients, employers and colleagues unidentifiable. Strong contributions may be quoted and credited on this site. Read the contribution guidelines and the privacy notice.

Comments are provided by giscus and stored on GitHub. Posting requires a GitHub account.

Next in the argument Semantic authority is independent of consumption Whichever tool computes a number quietly becomes the owner of its definition. Data as a first-class citizen

Connected thinking

Principle · draft Semantic authority is independent of consumption Business concepts and metrics need governed identities that exist independently of every tool that consumes them. Direction · draft Put computation where it belongs Recurring analytical work belongs in the data product's serving model and the query layer, not in each consumption tool. Pattern · draft Process ledger fact Model a coherent business process as a ledger of atomic, additive events at an explicit grain, and compose metrics from it. Diary · published Where does a KPI live?