The playbook
Ad Campaign Experiments: Metrics and Measurement
Measure ad campaign experiments by defining evidence supporting the campaign decision precisely enough that two people can reproduce the result from the same records. Keep the reporting unit, date basis, and scope visible. Without those definitions, a change in the report may reflect a changed calculation rather than a changed customer outcome.
Put the task in context
Write the hypothesis, comparison, outcome, and decision rule before launch. Record concurrent changes and allow for delayed outcomes. If results are uncertain, say so and decide whether the next useful action is more observation, a revised test, or stopping.
Create a metric definition
| Definition field | What it must explain |
|---|---|
| Measure | evidence supporting the campaign decision |
| Unit | What one campaign or conversion record or completed outcome represents |
| Included records | The scope identified by hypothesis and comparison |
| Date basis | The event date or period used to include a record |
| Exclusions | Tests, duplicates, or other records outside the agreed definition |
| Source owner | budget owner |
Separate delivery checks from outcome measures
A delivery check asks whether the team completed a specific task, such as state the decision the experiment will inform. An outcome measure asks what happened for the user or business afterward. Keep both visible. Completing the task is useful evidence of execution, but it does not by itself prove that the broader outcome improved.
Worked arithmetic example
Suppose an illustrative review includes 80 eligible records, and 60 satisfy the explicitly defined completion condition. The completion share is 60 ÷ 80 × 100 = 75%. This is hypothetical arithmetic, not an industry benchmark or a predicted improvement. It describes only the population and condition you defined.
For ad campaign experiments, explain the condition before applying that calculation. If the question is whether the team can document duration uncertainty and rollout criteria, a record counts only when the evidence demonstrates that condition. Do not replace missing outcomes with successful ones just to complete a report.
Compare like with like
Keep the scope, date basis, and exclusions consistent between observations. If comparison changes meaning between periods, note the change and rebuild a comparable view where possible. A rate can also move because the composition of the underlying group changed, even when individual performance did not.
Investigate an unexpected result
- Check whether the underlying hypothesis records are complete.
- Look for this known failure: stopping as soon as one variant temporarily appears ahead.
- Verify that the process still includes: choose meaningful outcomes and comparable conditions.
- Reconcile a sample with the original record before changing the operating plan.
Use the report to choose an action
If the evidence supports a correction, use a preplanned review and consider variability before deciding. Record when the change occurred and what should be checked next. Keep uncertainty visible when the sample is small or the observation period is incomplete. The purpose of measurement is a better decision about evidence supporting the campaign decision, not a favorable-looking number.
Keep advertising spend, platform-reported conversions, and confirmed business outcomes distinct. Different platforms may claim credit for the same sale. A combined report needs reconciliation or a clear explanation that figures cannot be added.
Skip the manual steps. Let AgentixFLO run this for every client.
Start now