A measurement plan states what question will be answered, how each metric is calculated and which conclusions the data cannot support. Write it before publishing. That reduces the temptation to select a flattering number after the event.
Start with a question, not a dashboard
“Did it work?” is too broad. Ask whether a defined audience completed a defined action more or less often during a stated observation window, or whether one message produced a different pattern from another under recorded conditions. Name who will use the answer and what decision follows.
Define every metric
| Field | Example structure |
|---|---|
| Metric | Plain-language name and formula |
| Event | The observable action counted once |
| Source | System, property and report |
| Scope | Channel, page, audience or campaign filter |
| Window | Start, end and timezone |
| Exclusions | Internal, test, bot or duplicate activity |
| Owner | Person responsible for extraction and notes |
Record the baseline using the same definition and filters intended for the later observation. If the tracking setup changes, annotate the date; do not present the resulting break as user behaviour.
Use campaign names consistently
Google’s official campaign URL guidance explains that UTM parameters can identify referring campaigns and recommends consistent values for source, medium and campaign. Create a short naming register before links are distributed. Treat case, spacing and punctuation as controlled choices because inconsistent values can fragment reporting.
Record the destination URL, tagged URL, source, medium, campaign, content variant, owner and creation date. Test the final redirect and landing page. Do not put personal or sensitive information into query parameters.
Interpret search metrics carefully
Google’s Search Console Performance documentation describes clicks, impressions, average click-through rate and average position for the selected period. Its metric definitions explain that position is averaged across impressions and that data is generally assigned to the canonical URL. Record the Search type, date range, page/query filters and comparison period.
Do not treat average position as a fixed rank seen by every person. Google recommends looking at trends in impressions and clicks as well as position. Search Console also omits some anonymised queries, so a query table need not equal the full property total.
Include guardrails and quality checks
- Verify the event fires once under the defined action.
- Test campaign links and naming values before release.
- Document consent and privacy controls with the responsible owner.
- Check whether seasonality or another simultaneous change affects comparison.
- Retain the raw export or query settings used for the summary.
Write the result in layers
First state the observation: metric, count, period and comparison. Then state the interpretation and its uncertainty. Finally record the decision. A change observed after a campaign is not automatically caused by the campaign. If the sample is too small or instrumentation failed, the correct result may be “insufficient evidence”.
Use the experiment log for controlled changes and the content checklist before publishing measurement claims.