An Integrated Operations Center can display many indicators while still lacking a shared management model. Two areas may use the same metric name with different time windows or denominators. Neither dashboard necessarily contains a calculation error; the disagreement may begin in the definition.
This is a proposed design process grounded in the public scope of industrial data and KPI governance work. The examples are illustrative. It does not reproduce an employer's catalog, approval structure or operating targets.
Start from a decision, not an available tag
For each proposed KPI, identify the decision it informs, who can act and how quickly the information must arrive. An hourly view may support shift intervention, while a monthly view supports a different management conversation. A high refresh rate does not automatically make a metric actionable.
Separate an outcome indicator from a diagnostic variable. A process measurement can explain an outcome without becoming a management KPI itself. Keeping that distinction reduces the number of indicators that need formal targets and escalation rules.
Write a definition that survives implementation
A catalog entry should be specific enough that two implementers can calculate the same result from the same approved inputs.
| Field | Question it must answer |
|---|---|
| Identifier and version | Which definition does this result use? |
| Operational purpose | Which decision or intervention does it support? |
| Scope and unit | Which equipment, process and measurement unit apply? |
| Numerator and denominator | What is counted, and against which exposure? |
| Window and aggregation | Shift, day or month; ratio of totals or mean of ratios? |
| Exclusions and quality | Which records are excluded and when is the result withheld? |
| Ownership | Who approves meaning, maintains inputs and operates the calculation? |
| Freshness and revision | When is a result provisional, final or restated? |
For a synthetic example, define a delay share as approved delay minutes divided by observed eligible equipment minutes. That is an illustrative metric, not a universal availability definition. Explicitly define eligibility, overlapping states and missing coverage before using it for comparison.
Assign responsibility to the actual workflow
Use a small RACI matrix per activity. A proposed arrangement is: the process owner is accountable for the definition; an operational specialist prepares it; a data steward maintains mappings and quality rules; an engineer implements the calculation. Consult adjacent process owners where boundaries cross, and inform report consumers when meaning changes.
Validate that arrangement with the real organization. “The IOC owns everything” is not a substitute for naming an accountable role. Definition approval, source correction and dashboard maintenance are separate activities and may have different responsible teams.
Reconcile inside and across processes
Within one process, reconcile shift and daily results. If valid observations have different denominators, a simple average of percentages can be misleading. Consider two synthetic groups: one has 10 delay minutes in 100 observed minutes, the other 40 in 200. Their combined share is 50/300, approximately 16.7%; the unweighted average of 10% and 20% is 15%.
Across mine and plant, check whether the same material, equipment boundary and accounting period are represented. Matching labels do not justify adding quantities or treating simultaneous observations as a single material flow. Record a reconciliation rule, or explain why direct reconciliation is unavailable. The Mine-to-Mill architecture note develops that limitation.
Treat a definition change as a release
Changing an exclusion rule can move a KPI without any operational change. Version the definition, record its effective date and approve whether historical periods should be recalculated. Preserve the prior result and explanation when a published value is restated.
Before release, compare the implementation against a small approved example set: normal intervals, zero denominator, missing inputs, boundary events and late corrections. Show insufficient data explicitly instead of converting it to a favorable zero. Publish valid coverage alongside the metric.
Deliver a catalog people can maintain
Begin with a bounded set of decisions and complete their definitions before expanding. A shared spreadsheet may be sufficient for initial review, provided it has an owner and revision control. A database or catalog tool becomes useful when relationships, access and change volume justify it; software cannot resolve disagreement about meaning.
The deliverable is an agreed metric contract, a responsibility map and evidence that the calculation follows the contract. Connect it to historian data quality rules so consumers can distinguish a process issue from an input-quality issue.