Use attribution to follow conversion journeys and experiments to estimate additional impact. Recovering more conversions through better tracking does not, by itself, mean advertising caused more sales.
Attribution and incrementality answer different questions
Attribution assigns conversion credit to touchpoints using rules or a model. Incrementality asks how the observed outcome differs from what would have happened without the advertising intervention. Estimating that alternative requires a suitable research design.
A customer already familiar with a brand might buy after clicking its branded search ad. Crediting the campaign helps describe the journey; it does not prove the customer would otherwise have abandoned the purchase. That is a hypothesis to test, not a conclusion that branded search has no value.
Where the Google update fits
On 10 September 2026, Google announced global general availability of Meridian GeoX following its beta. This open-source library supports geographic experiments and can operate independently or help calibrate marketing mix models. Google announcement.
This guide draws on an earlier update; it is not presented as news announced today. Google describes a publisher-agnostic solution that requires experiment design and analysis. Global availability does not establish that every business has enough suitable data. GeoX documentation.
A worked example
Consider a hypothetical experiment with equally sized, comparable groups. The treatment group records 120 purchases, while an appropriate design estimates it would have recorded 100 without the advertising. The estimated incremental impact is 20 purchases, a 20% lift over the counterfactual.
If those additional sales generate 2,000 monetary units and incremental advertising costs 1,000, incremental return on ad spend, or iROAS, is 2. That is a revenue return, not net profit. Product margins and costs still matter.
These figures are illustrative. Simply subtracting sales in two different regions is insufficient: baseline differences and uncertainty must be addressed. A small dataset may yield an inconclusive result.
Choosing a measurement approach
| Question | Approach to evaluate |
|---|---|
| Does a specific campaign add conversions? | A treatment-and-control experiment where available and appropriate |
| What happens when spending changes across regions? | A geo experiment that checks comparability and exposure spillover |
| How do channels contribute over a longer period? | Marketing mix modeling (MMM), informed by historical data and calibrated with experiments |
This is a CROMBA planning framework, not a universal tool specification. There is no single budget or duration suitable for everyone. Detectable effect size, sales volatility and conversion lag influence the design.
Before increasing the budget
- Define one decision the test could change, such as increasing a particular channel budget.
- Agree on the outcome: paid sales after cancellations, or leads qualified against a clear standard.
- Predefine groups, duration and success criteria; inspect historical data quality.
- Record promotions, price changes, stock issues and concurrent campaigns.
- Report the impact estimate, uncertainty and cost, then relate them to profitability.
If data is unstable, repair measurement first. Our guide to measuring lead generation in GA4 provides a useful starting point. An inconclusive result means the test did not settle the question; it does not establish zero advertising impact.
Frequently asked questions
Should platform ROAS be abandoned?
No. It remains useful for operational monitoring, while experiments add a causal perspective.
Is MMM a guaranteed substitute for experiments?
No. It depends on data and assumptions; experimental results can help calibrate it.
Sources and editorial review
Original sources used in this article. Last editorial review: 8 October 2026.
- Google — Data and measurement updates10 September 2026
- Google for Developers — Meridian GeoXReviewed 8 October 2026
