Formula and method
- Scaling ratio = treatment participants ÷ control participants
- Incremental revenue = treatment revenue − (control revenue × scaling ratio)
- Incremental cost = treatment media spend − (control media spend × scaling ratio)
- iROAS = incremental revenue ÷ positive incremental cost
- Relative revenue lift (%) = incremental revenue ÷ scaled control revenue × 100
Worked example
Illustrative treatment: 10,000 participants, $30,000 revenue and $5,000 media spend. Control: 5,000 participants, $10,000 revenue and no spend.
The control scales by 2, giving a $20,000 baseline. Estimated incremental revenue is $10,000. Incremental media cost is $5,000, so iROAS is 2× and revenue lift is 50%. This arithmetic does not establish that the example is statistically reliable.
Use assignment groups, not buyers or ad viewers
Enter the number of eligible units assigned to each group. Restricting the denominator to people who purchased or saw an impression can introduce selection bias. Keep the observation period and revenue definition identical across the two groups.
Unequal random group allocation is compatible with this scaling arithmetic. Comparing a high-value customer audience with an unrelated prospect audience is not. Multiplying one group's totals does not make different populations comparable.
Revenue should follow the experiment's outcome definition, including activity that an ad platform did not attribute to itself. If you subtract two platform-attributed reports, the result can describe attribution differences rather than incremental business impact.
Distinguish revenue efficiency from profitability
An iROAS of 2× means two units of estimated incremental revenue per unit of incremental media cost in this scenario. It does not mean two units of profit. Product, service, fulfillment and other costs still matter.
Google's geography-based lift documentation defines iROAS using incremental conversion value divided by incremental cost. Its geo studies use methods designed for geographic differences. This simpler worksheet does not reproduce those models and should not be used to replace their adjusted results.
A negative result is allowed. A zero or negative incremental media cost makes this positive-spend ratio unsuitable, so the calculator requests a different scenario instead of presenting an infinite or misleading efficiency value.
Keep uncertainty with the decision
Aggregate revenue and participant counts do not contain the individual outcome variation needed for a reliable revenue confidence interval. Keep the experiment platform's uncertainty estimates, allocation checks and data-quality notes beside this worksheet.
Before changing a large budget, ask whether the result is precise enough for that decision, whether the study represents the next spending level, and whether demand or operating capacity has changed. A point estimate from one spend level is not a forecast of marginal returns at every budget.
Download the inputs and output, then add the study identifier, observation dates and exclusions in your own analysis file. Keep personal records in your controlled systems; the public worksheet needs aggregate totals only.