The short answer
In most cases, trust the experiment's incremental ROAS of 4 when judging how much additional revenue your advertising generated. Attribution's ROAS of 8 can include purchases that would have happened anyway. First check that the experiment is credible and that both numbers cover comparable campaigns, sales, and dates.
Here's a summary video from Sellforte CEO, Juha Nuutinen:
This article is part of Asked by Marketers, a series answering real questions from marketing leaders.
Sellforte's team holds more than 1,450 meetings each year with marketing leaders in Ecommerce and Retail about Marketing Mix Modeling and incrementality testing. Each week, we anonymize at least one question from those conversations and answer it in depth, based on what marketers are actually struggling with, not what keyword tools suggest. About the series
Why marketers ask this
Your attribution dashboard says every dollar of advertising returned eight dollars in sales. Then an incrementality test reports four. The budget meeting is approaching, and the team needs to explain the difference. If campaign targets have been built around attributed ROAS, accepting the experiment's result also means reconsidering what counts as good performance.
Why should I usually trust the experiment?
A well-designed experiment estimates how much additional business the advertising caused. That is the evidence you need when assessing what the business received for its marketing spend.
Attribution assigns credit to ads based on tracked interactions and the rules of the attribution model. It can credit a purchase to a campaign even when the customer would have purchased without that campaign. For example, someone who already intends to buy may click a retargeting ad on the way back to your store. The purchase appears in the campaign report, but the click alone cannot tell you whether the ad changed the outcome.
Incrementality testing addresses that missing comparison. In an audience holdout test, one group is eligible for the advertising and another is held out. With a sound assignment and analysis, the difference in sales estimates the advertising's additional contribution. Geo tests use a comparable principle across regions.
So, when a credible experiment reports an incremental return on ad spend, or iROAS, of 4, use 4 as your estimate of incremental revenue per dollar for the activity and period tested. The attributed ROAS of 8 remains a record of credited sales. It does not establish that all eight dollars were caused by the advertising.
When should I question the experiment's result?
Question it when the design, data, or scope does not support the decision you want to make. The word “experiment” does not make a result reliable on its own.
- Check the comparison. Did the test and control groups provide a fair basis for estimating what would have happened without the advertising? Did the planned spending change happen, and did a promotion or tracking problem affect one group differently?
- Read the uncertainty interval. A headline iROAS of 4 may be precise enough to guide a decision, or it may sit inside a range so wide that the decision remains unresolved. Keep an inconclusive result in the evidence record and investigate what a better test would require.
- Match the scope. Compare the same campaigns and dates, using the same sales definition and relevant spend. A web-only study cannot establish the full effect on web and app sales. Revenue before returns also differs from net revenue.
- Check the observation window. Purchases can arrive after advertising stops. Confirm that the agreed measurement period captures the effects relevant to your decision, and distinguish that measured effect from any longer-term value the test did not observe.
If one of these checks fails, investigate it before using 4 as a firm budget assumption. Keep 8 labeled as attributed ROAS while that work is underway. Our guide to judging whether an incrementality test is credible explains the checks in more detail.
How this looks in practice
Consider a hypothetical campaign with $10,000 in ad spend. Assume a credible holdout test, comparable groups after the required adjustments, no spend on the tested ads in the holdout, and matching campaign coverage, revenue definitions, and measurement dates. All figures below are illustrative.
| Measure | Amount |
|---|---|
| Spend on the tested advertising | $10,000 |
| Revenue credited by attribution | $80,000 |
| Attributed ROAS | $80,000 ÷ $10,000 = 8 |
| Total revenue in the treatment group | $240,000 |
| Estimated treatment-group revenue without the tested ads | $200,000 |
| Estimated incremental revenue | $240,000 − $200,000 = $40,000 |
| Incremental ROAS | $40,000 ÷ $10,000 = 4 |
The experiment estimates that the campaign added $40,000 of revenue. Attribution credited it with $80,000. The sales themselves have not disappeared; the estimate of how much revenue the advertising caused is lower than the amount credited to it.
This also clarifies what “attributed sales” means. The $80,000 is the revenue assigned to this campaign under the attribution model. It is not the business's total revenue, and it is not automatically incremental revenue.
How should I use 4 in the budget meeting?
Present 4 as the estimated incremental return for the tested activity, together with its uncertainty interval and scope. Keep the attributed 8 visible so everyone can understand why the performance report and the experiment differ.
A useful explanation for the illustrative example is: “The campaign was credited with $80,000 in revenue. The experiment estimates that it generated $40,000 of additional revenue on $10,000 of spend. We are using an incremental ROAS of 4 to assess its contribution under the tested conditions.”
Compare that return with the business's requirements, including margins and relevant costs. An iROAS of 4 describes revenue per advertising dollar; it does not mean four dollars of profit. It also does not, by itself, tell you how much to increase or cut the budget.
For a spending change, assess the expected return on the dollars you plan to add or remove. An experiment comparing the existing campaign with no advertising measures a different decision from a test of a budget increase. Advertising response curves can help estimate how the return changes as spending changes.
What should I do with attribution after the test?
Use the experiment to calibrate the attribution numbers your team works with. For the matched example, the calibration factor is 4 ÷ 8 = 0.5. Multiplying the $80,000 of attributed revenue by 0.5 gives the $40,000 incremental estimate.
That factor belongs to the attribution source and activity you compared. A factor calculated against an ad platform's report cannot automatically be applied to a different attribution model. Nor should one result become a permanent 50% discount across every campaign. Traffic, awareness, and sales campaigns can need different corrections.
For ongoing planning, experiment-calibrated Marketing Mix Modeling can connect test evidence to a broader view of channel returns and spending decisions. Attribution then provides the campaign detail needed for execution. The resulting campaign estimates still depend on the model and calibration assumptions; a channel-level experiment does not independently test every ad.
Related questions
Can the experiment's ROAS ever be higher than attribution's?
Yes. Attribution may miss sales that advertising causes, including purchases without a trackable click or outside the sales channels it observes. An experiment can capture more of that effect when its outcome data and measurement window cover it. Our article on why platform ROAS and incremental ROAS differ explains how the gap can run in either direction.
How Sellforte helps
Sellforte brings experiment results and attributed performance into the same measurement workflow, with experiment-calibrated MMM supporting ongoing budget decisions. Teams can use the evidence to adjust campaign estimates and review how those estimates inform bids and budgets. Book a demo.
Authors
Lauri Potka is the Chief Operating Officer at Sellforte and has over 15 years of experience in Marketing Mix Modeling, marketing measurement, and media spend optimization. Before joining Sellforte, he worked as a management consultant at the Boston Consulting Group, advising some of the world's largest advertisers on data-driven marketing optimization. Follow Lauri on LinkedIn, where he is one of the leading voices in MMM and marketing measurement.
How should we weight multiple incrementality tests when calibrating an MMM?
When should an omnichannel retailer use MMM, MTA, or incrementality testing?
