# Performance review and the next test

## Use this workflow

Use after a campaign has run. Read campaign-brief.md for the objective. Ask for actual performance exports and the relevant asset versions. Possible inputs include ads, email, landing-page conversions, sales, enquiries and customer replies.

Confirm the date range, metric definitions, attribution window and whether the audiences and spend are comparable. Ask about ambiguous columns. Do not invent missing performance data or use the example below as a real result.

## 1. Check the data

Flag duplicates, missing values, inconsistent definitions and zero denominators. Missing is not zero. Do not merge different attribution windows or sum incompatible metrics. Distinguish all enquiries from qualified enquiries if the objective depends on qualification.

## 2. Explain what happened

Measure against the stated objective. Show counts and rates with the numerator and denominator. Calculate cost per outcome only when the relevant spend is available and comparable.

Identify stronger observed results where the comparison supports them. Note sample-size limits. Do not force a winner from weak evidence or treat a correlation as causation.

## 3. Explain what might have caused it

For each explanation, label it as an observation, supported interpretation or hypothesis. Link it to the data or customer responses. Name plausible alternative explanations and the information needed to distinguish them.

Look for messages that get attention without the intended conversion, unclear next steps, questions the landing page does not answer, and meaningful differences between audience segments. Treat email open rates cautiously; use more direct outcome measures when available.

## 4. Propose the next test

Return performance-review.md with findings, calculations, limitations and up to three ranked tests. For each test, state:
- Finding or hypothesis being tested.
- One deliberate change.
- Assets and audience affected.
- Expected signal and success metric.
- Comparison period and information needed to judge the result.
- The result that would weaken the hypothesis.

Stop and wait for the user to choose a test.

## 5. Apply the approved learning

Read the existing assets before revising them. Preserve the core campaign idea and unrelated elements. Update only the assets affected by the chosen test. For a CTA-clarity test, the change may travel through content, social creative and email.

Show previous and revised versions side by side. Save iteration-notes.md with the change, reason, affected asset IDs and evaluation plan. Ask for review; do not publish the changes.

Propose lasting context updates only when supported and approved. Save customer evidence in audience context, brand preferences in brand instructions and one-off campaign findings with the campaign. Record source and review date.

## Worked example from the video

Illustrative data, not an actual campaign result:

| Creative | Clicks | Enquiries | Enquiries / clicks |
| --- | ---: | ---: | ---: |
| A | 100 | 2 | 2% |
| B | 60 | 6 | 10% |

Calculation: 2 / 100 = 2%; 6 / 60 = 10%.

B has fewer clicks but more enquiries and a higher observed enquiry rate. That makes it worth investigating when enquiries are the objective. We cannot claim a reliable business winner without comparable spend, lead quality, attribution and enough data. The table does not explain why the difference occurred.