Metric governance
Give every critical metric a meaning, an owner and a change record.
Metric governance establishes which definition is valid for which decision, how it is produced and what must happen when the logic or source changes.
The actual problem
The dispute is often semantic before it is technical.
Two dashboards can query clean data and still disagree because they answer different questions. One may count accounts, another users; one may use event time, another processing time; one may restate history after late arrivals while another freezes the period.
Metric governance makes those choices explicit. It does not force one universal definition where the business legitimately needs several. It ensures that each measure is named precisely and used for the decisions it was designed to support.
- Same label, different formula“Active customer” or “revenue” changes meaning across teams and reports.
- Dashboard reconciliationLeadership time is spent explaining differences rather than interpreting performance.
- Silent definition driftA source or transformation changes and historical comparability breaks without notice.
- No decision ownerEngineering owns the pipeline, but nobody can approve the business meaning.
- Metric proliferationNew measures are added faster than obsolete or duplicate measures are retired.
- Threshold ambiguityA KPI is monitored, but no action or escalation follows when it moves.
Method
Govern a small set of decision-critical metrics deeply.
A catalogue containing hundreds of undocumented fields is not governance. The work begins with the measures used for material decisions and gives them enough semantic and operational detail to be reproduced and challenged.
- Usage mapping
- Identify where a metric appears, who consumes it and which decisions, commitments or external statements depend on it.
- Semantic contract
- Specify purpose, formula, grain, population, exclusions, time semantics, dimensions and expected interpretation.
- Technical trace
- Connect the approved meaning to source fields, transformation logic, refresh behaviour and quality controls.
- Ownership
- Separate the business owner of meaning from technical stewardship and from the forum that approves material changes.
- Change control
- Assess impact, version definitions, communicate effective dates and decide whether historical results must be restated.
- Use and retirement
- Review adoption, duplicate measures and whether the metric still supports a live decision or obligation.
Evidence required
What is needed to establish a metric contract.
The contract is built from actual use and actual code. A written definition that cannot be connected to production logic does not resolve trust.
| Input | Question answered | Expected evidence |
|---|---|---|
| Decision and reporting context | Why does this metric exist? | Meeting pack, target, operational rule or external commitment. |
| Current definitions | Where do meanings diverge? | Glossaries, dashboard labels, analyst notes and stakeholder explanations. |
| Query and transformation logic | How is the number produced? | SQL, semantic model, calculation field or reproducible transformation. |
| Source-system behaviour | What can alter the population? | Event generation, update rules, deletions, backfills and late arrivals. |
| Historical outputs | Has comparability changed? | Time series, release notes and known breaks in series. |
| Ownership and controls | Who can decide and who responds? | Named business owner, steward, quality responder and approval forum. |
Outputs
A governance system that can be used in normal work.
The outputs are deliberately compact. Their purpose is to remove ambiguity at the point where a metric is built, interpreted or changed.
- Metric contractPurpose, formula, grain, population, exclusions, time semantics, dimensions, source and caveats.
- Ownership recordBusiness owner, technical steward, quality responder, consumers and approval authority.
- Lineage referenceA trace from presented value to semantic model, transformation and source fields.
- Quality expectationsFreshness, completeness, uniqueness or reconciliation rules appropriate to the decision.
- Change protocolImpact assessment, versioning, communication, effective date and historical restatement decision.
- Metric registerA prioritised inventory showing approved, provisional, deprecated and retired measures.
Worked example
“Active customer” has no useful meaning without a decision.
A digital service has 50,000 customer accounts. Three teams report a different active-customer figure, and each number is reproducible.
| Proposed measure | Count | Definition | Suitable use |
|---|---|---|---|
| Logged-in accounts, 30 days | 42,800 | Account with at least one authenticated session. | Authentication infrastructure and broad reach. |
| Core-action accounts, 30 days | 18,600 | Account completing at least one defined value-producing action. | Product adoption and service utilisation. |
| Contracted core-action accounts | 11,200 | Account covered by an active customer contract and completing the core action. | Customer success capacity and commercial retention. |
Contracted share of core-action accounts = 11,200 / 18,600 = 60.2%
None of the three counts is inherently wrong. The error is calling all three “active customers” and allowing the label to travel between infrastructure, product and retention decisions.
The governance decision: approve three explicit names, assign a use case to each, document the core event and prevent dashboards from presenting the generic label without a qualifier.
Limits
Governance cannot manufacture agreement or clean data by itself.
The framework makes accountability and disagreement visible. Owners still need the authority and time to resolve them.
- Not a universal glossary projectLow-value fields do not need the same control as decision-critical metrics.
- Not automatic data qualityA contract defines expectations; engineering and operational controls must enforce them.
- Not one metric for every purposeDifferent decisions may require different populations, windows and levels of conservatism.
- Not governance by committeeConsultation can be broad, but final meaning and change authority must be explicit.