Understand what attributed ROAS cannot prove about incremental sales

Credit for a sale is not proof of causing it

A credible control approximates what would happen without the treatment. Depending on the design, units can be people, locations or another suitable population, with assignment and outcome measurement planned in advance. Google describes Conversion Lift as a controlled comparison for incremental outcomes and notes that availability is not universal. This does not mean that every account has access or that a self-selected untreated region provides a valid substitute.

Educational example, not a measured lift study: advertising costs 1,000 units and receives 5,000 units of attributed value, giving attributed ROAS 5. In a hypothetical valid, normalized comparison, the treated population produces 8,000 in business revenue and its counterfactual estimate is 6,000. The incremental revenue estimate is 2,000, giving incremental revenue/spend 2. If incremental contribution before advertising is 40% of that revenue, it is 800, below the 1,000 advertising cost. The lesson is that attributed return, incremental revenue and incremental contribution answer different questions.

Raw treated and control totals are not comparable when population sizes or baseline demand differ. Explain normalization and the assumptions behind it. Watch for contamination: customers cross regions, shared advertising reaches control units, or a broad promotion changes both populations. A contemporaneous control helps address shared conditions but cannot automatically correct unequal treatments, inventory differences or outcome measurement failures.

Ask the counterfactual question

• Keep attributed ROAS as attributed evidence rather than discarding it. It can support operational diagnosis while a separate design investigates causal impact.
• Select the business outcome independently of platform credit where the design requires it. Otherwise the measurement can merely repeat the attribution assumption it is supposed to examine.
• Use contribution economics for the business decision. Incremental revenue can still be financially insufficient after variable fulfilment and acquisition costs.

Plan a defensible incrementality assessment

1. Define the treatment, population and causal question. Specify whether the decision concerns a campaign, channel or budget increment, since those are different counterfactuals.
2. Choose a credible control design and document assignment, baseline comparability and possible spillover. Check actual platform eligibility before proposing a provider-managed study.
3. Set the outcome definition, observation window, conversion maturity and analysis approach before examining results. Verify that both groups have consistent collection and business operations.
4. Inspect contamination, promotions, stock availability and sales capacity during the assessment. Maintain an incident log so that an interruption can qualify or invalidate the interpretation.
5. Report the incremental estimate with uncertainty and the exact population and period tested. AdAce Ads records can support read-only preparation, but a stored ROAS report alone must not be described as an incrementality experiment.

A lift estimate has boundaries

Sources and further reading

• Google Ads: Conversion Lift and account availability