A practical guide to choosing an incrementality question, treatment, control, outcome and decision rule before trusting attributed conversions.
Quick answer
Attribution assigns credit within a measurement model; incrementality asks what additional outcome happened because of the activity. Start with one causal question, define an eligible population, create a credible treatment and control, measure the same downstream outcome in both groups and decide in advance how the result will change budget or execution.
Attributed conversions are useful for reporting journeys, but they do not automatically show what would have happened without the campaign. That distinction matters when teams use platform totals to justify marginal spend.
Incrementality is not a replacement for everyday reporting. It is a focused way to answer expensive questions about causal impact.
State one decision-sized question
Frame the study around a choice the business can actually make.
Examples include whether a channel creates additional qualified leads, whether a new campaign expands demand or captures existing demand, and whether a website treatment improves completion among eligible visitors.
Name the population, intervention, outcome and time window. Avoid questions so broad that different mechanisms and audiences become inseparable.
- Who is eligible?
- What changes for treatment?
- What is withheld or unchanged for control?
- Which outcome would alter the decision?
Separate attribution from causal lift
Do not compare numbers that answer different questions as if they were interchangeable.
Google describes Conversion Lift as the difference in conversions between groups exposed and not exposed to ads. Standard attributed conversions instead follow the conversion action's attribution settings and windows.
The two views can coexist. Attribution helps operate campaigns; lift evidence helps estimate whether activity produced additional outcomes. Differences are a reason to investigate, not to select whichever number is larger.
Protect the comparison
A useful control should differ from treatment mainly in the intervention being tested.
Watch for overlapping campaigns, audience contamination, uneven budgets, seasonality and operational changes during the test. Record them rather than explaining them away afterwards.
Google Ads experiments split traffic or budget between an original and trial campaign, but auction dynamics can still produce unequal exposure. Review actual delivery and eligibility before interpreting the result.
Make uncertainty visible in the decision
A test result is a range of plausible effects under assumptions, not a permanent truth.
Predefine the primary metric, guardrails, minimum useful effect and stopping approach. Avoid repeatedly checking and ending only when the preferred answer appears.
Document what the study cannot tell you, then apply the result at the same level as the test. A result for one audience or period does not automatically justify an account-wide rule.
Further reading
Sources
- About Conversion Lift — Google Ads Help
- Test with confidence with the Experiments page — Google Ads Help
- Monitor your experiments — Google Ads Help
