How to Measure Campaign Performance: Shared KPIs, Deduplication, and GA4 Attribution
Shusaku Yosa

Measuring a campaign means agreeing on outcomes, costs, timing, and evidence. It does not mean adding every dashboard total together. Keep the channel reports, but establish a shared basis for decisions and identify what cannot be reconciled or observed.
Define the Campaign Objective and Counted Outcome
Start with the decision the measurement must support. For lead generation, distinguish submissions, qualified enquiries, opportunities, and wins. Decide whether two enquiries from the same company concern one buying need or separate opportunities. Counting enquiry IDs does not automatically count people.
Agree on the observation window before launch. A hypothetical plan might accept enquiries from October 1–31 and finalise qualification on November 15. The intake period and finalisation date are different fields. Check the actual buying cycle and reporting delay rather than assuming ads produce immediate outcomes while every webinar requires months. Preserve provisional and final snapshots.
For awareness, select an appropriate measure such as a consistently designed recall survey or qualified audience reach. An acquisition count alone cannot answer every awareness question.
Build a Measurement Dictionary for Each Channel
Reports can disagree because they use different outcome definitions, click-through or view-through rules, attribution windows, time zones, or cancellation updates. Document these differences before deciding which number belongs in the shared report. This is a hypothetical measurement plan.
Objective and metric | Counting basis | Period and finalisation | Missing data and owner |
|---|---|---|---|
Acquisition: qualified enquiries | Enquiry register; agreed company-and-purpose deduplication | October 1–31 intake; November 15 finalisation; JST | Unreviewed enquiries remain pending; sales and marketing |
Awareness: brand recall | Respondent percentage with consistent questions and audience criteria | Separate before-and-after surveys | Record sample or method changes; brand owner |
Traffic: landing sessions | GA4 session acquisition dimensions | Consistent dates and time zone | Consent, tags, and device limitations; analyst |
Maintain an immutable initiative ID separately from its display name and map it to platform IDs. Agree on UTM naming and check how manual tagging interacts with auto-tagging. Consistent identifiers support reconciliation; they do not create automatic joins between costs and outcomes.
Separate Duplicate Outcomes from Unobserved Touchpoints
More than one channel may claim credit for the same outcome. Deduplicate using the agreed outcome ID. In this hypothetical example, both channels use identical definitions and cutoffs, and every outcome ID in scope can be reconciled. Costs include media spend only, in JPY excluding tax.
Report | Media cost JPY | Reported qualified outcomes |
|---|---|---|
Channel A | 600,000 | 70 |
Channel B | 400,000 | 50 |
Outcome IDs appearing in both | Not an additional cost | 30 |
Deduplicated outcomes = 70 + 50 − 30 = 90. Combined CPA = (600,000 + 400,000) / 90 = JPY 11,111.11 per outcome, or approximately JPY 11,111 when rounded to whole yen. Adding or simply averaging channel CPAs does not produce this result.
Reconciled outcome IDs do not mean every advertising touchpoint is observed. Keep unreconciled outcomes separate instead of labelling the raw total of 120 as people. Self-reported event or referral sources can help, but recall errors, non-response, and multiple touchpoints remain limitations.
Distinguish First-User, Session, and Event Attribution in GA4
GA4 traffic-source dimensions answer different questions at different scopes. Choose the dimension and metric together.
Question | Example GA4 dimensions | Interpretation |
|---|---|---|
First observed user acquisition | First user source / medium | Original acquisition information; not a first-click attribution model |
Acquisition for each session | Session source / medium | A session acquisition series, distinct from event attribution settings |
Credit for key events | Source / medium with Key events | Credit under the reporting attribution model; fractional credit is not people |
Google states that first-click and several other legacy models have been unavailable since November 2023. First-user dimensions still existing does not restore the first-click model. Reporting model changes affect event-scoped attribution differently from user- and session-scoped dimensions.
If a separate register assigns reference credit to first or last touch, label it as your own analysis rule. Ranking highly under both rules does not establish the incremental effect of running the campaign. Also distinguish the attribution lookback window from the dates selected in the report.
Evaluate CPA and Awareness Signals on Consistent Terms
Compare CPA only after aligning outcome definitions, cost coverage, and dates. Media-only CPA differs from a measure that includes production. With zero outcomes, division is undefined; with pending outcomes, CPA remains provisional. A low acquisition cost does not establish sales quality or the capacity to handle demand.
Awareness signals also need a denominator. New-user share or direct traffic alone is not proof of stronger awareness: identification gaps, missing source information, and seasonality may affect them. For list acquisition, review duplicates, contactability, audience fit, and necessary consent alongside volume.
Use attribution to describe credited touchpoints. Treat incrementality as a separate question requiring an appropriate experiment or comparison. Where that is not feasible, report the observed change and the untested explanation separately.
Move from Measurement Checks to Review Decisions
- Agree on success, exclusion, cancellation, and deduplication rules and name their owner.
- Use an authorised test to check consent, events, destinations, UTMs, and the outcome register.
- Record cost coverage, as-of date, time zone, and finalisation date.
- Log missing data and setting changes without confusing a measurement issue with campaign performance.
- Review final counts, quality, and limitations; assign the next owner, observation period, and continuation conditions.
When totals disagree, trace one outcome ID through the reports and inspect its inclusion rules. Choosing whichever model makes the figures look consistent undermines future comparisons. Retain the previous conditions and change date whenever a definition or setting changes.
References (checked October 10, 2026)
- Google Analytics: Select attribution settings
- Google Analytics: Scopes of traffic-source dimensions
- Google Analytics: Traffic-source dimensions, manual tagging, and auto-tagging
Related reading