Governed Metric Set

Attempting to govern every number produces a bottleneck: every new question requires a definition, a review and an implementation, so analysts route around the platform. Governance that is routed around provides less assurance than none, because it exists on paper and not in the system.

The workable boundary is a small governed set — the numbers in board reporting, regulatory filings, team objectives and customer commitments, typically around twenty — with everything else self-service and explicitly labelled as ungoverned.

Why it matters

It is the difference between a governance programme that operates and one that is nominally in place. It also directs the effort at the numbers where a disagreement is expensive, rather than spreading it evenly across analysis where a disagreement costs nothing.

Implementation patterns

• Choose the set by which numbers cause the argument — the same metric reported differently by three teams, each confident, unresolvable by inspecting the numbers — rather than by which are most queried.
• Define each once in a semantic layer, version-controlled and reviewed like code, with dimensions defined alongside since most disagreements are about the slice rather than the measure.
• Consumed rather than copied. A layer producing documentation for humans to reimplement solves nothing.
• A named owner per metric, because a definition with no owner cannot be changed or defended.
• A promotion path from ungoverned to governed, since without one, important metrics stay ungoverned because governing them is a project.
• Label the ungoverned tier clearly, give it the same underlying data so it is not working from a worse copy, and apply cost quotas since analytical workloads have no natural ceiling.

Industry example

Marketplaces such as Meesho and analytics platforms such as Amplitude both encounter the three-teams-three- numbers problem at scale. The resolution that works is not more governance but less, applied precisely — and the boundary itself is what has to be communicated, since an unstated boundary means every analyst assumes a different one.

Failure scenarios

• Governing everything, producing a bottleneck and abandonment.
• Governing nothing, producing irreconcilable numbers in board reporting.
• No promotion path, so the governed set ossifies while the important metrics move.
• The ungoverned tier unlabelled, so its numbers are quoted as authoritative.
• One definition imposed where two are genuinely needed — if finance and product mean different things by "active customer", the answer is two metrics with distinct names, since an imposed single definition produces a number neither trusts and reappears as a shadow metric within a quarter.

Trade-offs

A small governed set means most analysis is ungoverned, and an ungoverned number will occasionally be quoted in a decision it should not have been. That is a real cost.

The alternative is a governed set nobody uses, which is worse — and the mitigation is the labelling and the promotion path rather than expanding the set until it collapses under its own review burden.

Interview question

"Your organisation has four hundred dashboards. How many metrics would you govern, how would you choose them, and what would you tell everyone about the rest?"