Document enough information for another reviewer to compute the same metric from the same inputs. Name the formula, unit, aggregation, population, reporting clock, attribution rule, source, exclusions, approver, and change date. Resolve disagreements before signing off a report.
A template you can actually fill in
| Field | What to record |
|---|---|
| Name and purpose | What is measured and which decision it supports |
| Unit and grain | USD, count, percent; one row per account/campaign/day, for example |
| Formula | Numerator, denominator, eligible population, and zero/missing rules |
| Aggregation | Sum, count, weighted ratio, or other explicitly defined operation |
| Attribution | Which system assigns credit; model, window, and overlap limits |
| Period and time zone | Inclusive start, exclusive end, named time zone, event/interaction clock |
| Exclusions | Statuses, channels, refunds, test traffic, and other agreed exclusions |
| Source and refresh | Export/field reference, extraction time, and late-arrival policy |
| Approval | Responsible owner, definitions approver, version, change date, and acceptance rule |
Two synthetic definitions
RP-MET-01 — Spend per platform-reported conversion. USD spend divided by that same platform’s reported conversions for the same account, window, attribution settings, and currency. With $1,000 spend and 20 reported conversions, the result is $50. Sum spend and conversions first, then divide. If conversions are zero or missing, show unavailable with the reason. Approver: fictional agency analytics lead; definition version 1, October 27, 2026.
RP-MET-02 — CRM-qualified lead rate. Count of unique eligible leads marked qualified by the agreed snapshot cutoff divided by unique eligible leads created in the period, multiplied by 100. The fictional eligible cohort has 100 lead IDs and 30 qualified IDs, so the rate is 30%. Exclude labeled test IDs; document the exact qualification status and deduplication key. Approver: fictional agency account lead; definition version 1, October 27, 2026.
These are different measures. Twenty platform conversions and thirty CRM-qualified leads do not establish a discrepancy without a common population and crediting rule. Platform-reported conversions may overlap across platforms and do not by themselves establish unique leads or incremental impact.
Check aggregation before copying a ratio
Synthetic campaigns spend $600 for 10 conversions and $400 for 10 conversions. The combined cost per reported conversion is $1,000 ÷ 20 = $50. Averaging campaign ratios is unsafe when their denominators differ. With $600/10 and $400/20, the correct combined figure is $1,000/30 ≈ $33.33; the unweighted mean of $60 and $20 is $40.
Keep ownership visible
When a definition changes, record why, the effective period, who approved it, and whether historical reports were restated. Recheck dashboards, spreadsheet formulas, slides, and footnotes against the same version. An unknown approver remains an unresolved sign-off item.
Sources and reproducibility
Technical references checked October 27, 2026. Worked examples are synthetic ReportPatch fixtures, checked by automated calculations.