One Number, Three Narratives: A Custody Model for Contested Business Metrics
Photo: Rex Needle , CC BY-SA 2.0, via Wikimedia Commons
The board pack problem is usually a custody problem
A familiar scene in UK mid-market firms: the monthly board pack contains three revenue or churn figures, each produced by a different team. Finance works from the billing system, operations from the service desk, marketing from the CRM. The numbers disagree by a few percentage points and the meeting spends twenty minutes reconciling rather than deciding. No one is acting in bad faith. Each team has built a defensible definition for its own purpose. The failure is not in the pipeline; it is in the absence of custody.
Metric stewardship is the cheapest data quality intervention available to most organisations. It does not require a new platform, a master data programme or a large consultancy engagement. It requires named owners, published definitions, a short conflict path and the discipline to retire metrics that no longer inform a decision. In cash terms, the cost is a few hours a month from people already on the payroll. In decision terms, the return is a board that can argue about strategy rather than arithmetic.
Why another dashboard will not fix it
When three teams report different truths, the instinct is to buy a semantic layer or a BI governance tool. Those tools help, but they automate a decision that has not yet been made. A semantic layer cannot choose between two legitimate definitions of active customer. It can only publish the choice someone else has made. Without a custodian, the tool becomes another place where definitions compete.
The same applies to data quality dashboards. They will show that the finance churn number and the marketing churn number diverge, but they will not say which one the board should use. Stewardship supplies the missing editorial authority.
A lightweight RACI for metric custody
The model needs four roles, not a new department. The metric custodian is accountable for the definition and its certification. The business owner is accountable for how the metric is used and prioritised. The data steward is responsible for implementation, monitoring and evidence. The data governance lead is consulted on standards and informed of exceptions. For financial metrics, a finance business partner should be consulted before publication.
| Activity | Metric custodian | Business owner | Data steward | Governance lead |
|---|---|---|---|---|
| Approve definition | A | C | R | C |
| Certify publication | R | A | C | I |
| Triage variance | R | A | C | I |
| Approve definition change | A | C | R | C |
| Retire metric | C | A | I | R |
One person should not hold every A. The custodian owns the definition; the business owner owns the consequence. That separation prevents a common failure in which a data team publishes a definition that the business has never agreed to defend.
A conflict-escalation path that ends
Disputes should be resolved in days, not quarters. A five-step path is enough.
- Log the dispute in the metric register, with the competing definitions, the evidence behind each and the decision affected.
- The metric custodian convenes the relevant teams within two working days and proposes a single definition or a documented split.
- If the custodian cannot resolve it, the business owner and a finance representative decide within three working days.
- Publish the interim status as contested, with a named decision date. Do not allow parallel certified dashboards.
- Record the decision, update the definition version and notify downstream report owners.
The timebox matters more than the org chart. A contested metric is not a technical incident; it is a decision waiting to be made.
The certified definition template
A certified definition should fit on one page. Publish it where the dashboard links to it, not in a folder no one visits.
| Field | What to publish |
|---|---|
| Metric ID and name | A stable identifier and the label used in reports |
| Decision question | The business question the metric answers |
| Formula | The calculation in plain language and in SQL or equivalent |
| Grain | The unit of analysis, such as customer, order or month |
| Inclusions and exclusions | What is counted, what is excluded and why |
| Source systems | The systems of record and any transformations |
| Custodian and business owner | Named individuals, not team names |
| Refresh and latency | How often the metric updates and how late data can be |
| Quality threshold | The tolerance for completeness, timeliness and accuracy |
| Version and approval date | The current version and who approved it |
| Change log | A short history of material changes |
| Status | Certified, contested or deprecated |
The template is deliberately boring. Boring definitions are easier to trust, audit and retire.
Retire the zombie KPIs
Every metric register accumulates zombie KPIs: numbers that appear in reports but no longer inform a decision. They consume engineering time, confuse new joiners and dilute the credibility of the metrics that matter. A quarterly review should ask four questions. Has this metric changed a decision in the last two quarters? Does it have a named business owner? Is it duplicated elsewhere? Does its maintenance cost exceed its decision value? If the answer is no, yes, yes and yes, deprecate it and redirect the dashboard to a certified alternative.
Start with three
Choose the three metrics that cause the most debate in the board pack. Assign custodians, publish definitions and run the escalation path for one quarter. The aim is not a perfect metric catalogue. It is a visible reduction in the time spent reconciling numbers and a visible increase in the time spent using them. That is a data quality intervention most UK mid-market firms can afford this quarter.
