Campaign experiment decision records for careful learning
Document campaign experiments with a frozen hypothesis, exposure boundaries, evidence limits, accountable interpretation, and follow-up decisions.
Campaign experiment decision record
Write the decision record before launch so the team knows what question the campaign experiment can answer. Capture the hypothesis, variants, eligible audience, assignment method, primary measure, guardrails, owner, and decision date.
This keeps the team from redefining a result after inspection. The record should also state what the design cannot establish, especially when assignment, identity, or sample boundaries are imperfect.
Define exposure and exclusions
Describe who may enter, how a person or account receives a variant, when exposure begins, and what counts as participation. Exclude staff, tests, bots, duplicate identities, prior exposures, and prohibited audiences through approved rules.
Map consent, suppression, region, customer, partner, and frequency controls before activation. An experiment label does not relax privacy, messaging, or channel governance.
Lock measures and observation rules
Name the authoritative source, event definition, attribution window, denominator, missing data treatment, and observation cutoff for each measure. Separate delivery, engagement, conversion, pipeline, and revenue measures instead of blending unlike stages.
Record instrumentation and campaign configuration versions. If a tracking change occurs during the run, preserve the boundary and ask the accountable analyst how interpretation should change.
Monitor execution without rewriting it
Check assignment balance, delivery failures, broken destinations, duplicate exposure, consent exceptions, and material customer complaints. Use predefined safety rules to pause harmful behavior without selecting a preferred result early.
Log every operational change with time, reason, actor, and affected population. A correction may improve execution while also limiting comparability, so retain both facts.
Interpret the result cautiously
Report eligible, assigned, exposed, observed, excluded, and unresolved populations before discussing outcomes. Compare the predefined measure and guardrails, then label exploratory observations separately.
Avoid claims of causation when assignment or measurement does not support them. Do not turn a directional difference into a universal audience rule without review of uncertainty, context, and repeatability.
Record the decision and next step
Name the decision owner and record whether the team adopts, rejects, repeats, narrows, or stops the tested change. Link the decision to evidence, limitations, affected assets, and an effective date.
Preserve the original plan, configuration, result, and decision as one versioned package. Carry unresolved questions into a new test rather than quietly changing the meaning of the completed experiment.
Sources
- Test with confidence with the Experiments page, Google Ads Help. Provides official guidance on hypotheses, campaign experiment setup, and review.
Related GTM guides
Frequently asked questions
Can the primary measure change after results appear?
Keep the original measure intact. Treat any later measure as exploratory and document it separately.
Should a broken destination invalidate the experiment?
Record the affected population and timing, correct unsafe execution, and let the accountable analyst determine the interpretation limit.