A dashboard cannot resolve a disagreement about what a metric means. ‘Revenue,’ ‘conversion’ or ‘active customer’ can refer to different definitions across teams. AI can organize the discussion and expose conflicts, but the business owner must decide what a metric represents.
Gather existing reports, source-field descriptions, calculation logic and examples. Start with a few metrics used in an actual decision. The result should be a small, versioned dictionary that a marketer, analyst and manager can read the same way.
What you will leave with
A metric dictionary with formulas, grain, exclusions, sources, ownership and worked checks.
Step 1
Name the decision the metric supports
Ask who uses the metric and what they do when it changes. Remove measures with no current decision or monitoring purpose from the first pass. Use plain definitions before implementation detail: for example, what qualifies an action as a conversion and when it is counted.
Each metric supports a named decision or monitoring need.
Step 2
Expose competing definitions
Have AI extract definitions from existing reports without choosing a winner. Compare formulas, date basis, filters, attribution, currency and exclusions. Keep different business concepts separate even when they share a familiar label. Document unresolved differences for the accountable owner.
Conflicting definitions are visible instead of silently combined.
Step 3
Write the metric contract
For each accepted metric, record its name, business definition, formula, unit, grain, source fields, inclusion rules, exclusions, zero or missing-data behavior, refresh expectation and owner. For ratios, define numerator and denominator separately. State which dimensions can be used to slice the metric without changing its meaning.
Try this instruction
Draft a KPI dictionary from these approved definitions. Include business meaning, formula, unit, grain, numerator, denominator, source fields, filters, exclusions, missing-data behavior, freshness, owner and a worked example. Flag unresolved conflicts; do not choose policy for us.
The definition is precise enough for two analysts to implement consistently.
Step 4
Test with a small worked example
Create clearly labeled sample rows and calculate the expected result by hand or with an independent formula. Include a zero denominator, a missing value and a boundary date where relevant. Compare the implementation to those expectations. A familiar metric name is not a substitute for a shared calculation.
The metric passes representative and edge-case examples.
Step 5
Publish ownership and change rules
Put the dictionary where reports and playbooks can link to it. Record who can approve a change and how affected dashboards will be updated. When a definition changes, preserve the previous version and explain whether historical comparisons remain valid. The dictionary is part of the operating workflow, not an attachment everyone forgets.
The metric has an owner, a stable reference and a change process.
When you get stuck
Two dashboards use the same name for different numbers.
Compare the contracts. Rename distinct concepts or reconcile the implementation to the accepted definition.
The total of a rate is calculated by averaging row rates.
Define the aggregation rule explicitly. For many business ratios, calculate from total numerator and denominator.
Terms explained
- Metric contract
- An agreed definition and calculation rule for a measure.
- Dimension
- A field used to group or filter observations, such as date, channel or product.
Sources and further reading
The workflow and templates are teaching guidance. The sources below support the underlying tools and concepts; product setup and availability can change.
Take it into your next task
Your completion checklist
0 / 5 checked
Checks are saved on this browser. Mark them only after you inspect the evidence. These are your checks, not an automated certification.