Learn & Understand

Blended ROAS Hides Diminishing Returns - Incrementality Testing Reveals Them

A campaign can report a perfectly healthy ROAS while generating almost no genuinely new revenue at all - because ROAS, as commonly calculated, measures revenue that happened alongside ad exposure, not revenue that happened because of it.

The Gap Between Correlation and Causation in ROAS

Standard ROAS attributes revenue to ads based on some attribution model (as discussed in this category's CPA guide, the same attribution complexity applies directly to ROAS too), crediting a purchase to an ad if the customer was exposed to or interacted with it before converting. But some portion of that "attributed" revenue would very likely have happened anyway, even without the ad - a customer already planning to buy a well-known brand's product who happens to also see a retargeting ad gets fully counted as an ad-driven conversion, even though the ad may have had little or no actual causal effect on their decision.

Incrementality Testing: Measuring What Actually Changed

Incrementality testing addresses this directly by running a controlled experiment: a randomly selected holdout group of the audience is deliberately withheld from seeing the ads (or a specific campaign), while the rest of the audience sees them as normal, and the resulting difference in conversion rate or revenue between the exposed group and the holdout group isolates the ad's genuine causal lift - the revenue that specifically would not have happened without the advertising, rather than revenue merely correlated with having been shown an ad.

Why This Often Reveals a Much Smaller "Real" ROAS

It's common for incrementality testing to reveal that attributed ROAS substantially overstates true incremental ROAS, particularly for retargeting and brand-search campaigns that tend to reach audiences already inclined to convert - sometimes finding that a campaign reporting a strong 5x attributed ROAS is delivering meaningfully less actual incremental lift once a proper holdout comparison isolates the ad's real causal contribution from purchases that would have happened regardless.

Why Most Advertisers Still Use Blended ROAS Day to Day

Incrementality testing requires meaningful scale, careful experimental design, and a willingness to deliberately withhold ads from a portion of a valuable audience, which makes it impractical to run continuously for every campaign - most businesses use standard attributed ROAS for routine, day-to-day optimization and reserve incrementality testing for periodic, more rigorous checks on their biggest spending channels, treating the two as complementary tools rather than direct replacements for each other.

Applying This to a Healthy-Looking ROAS Figure

A strong attributed ROAS, particularly on retargeting or brand-focused campaigns, is worth treating as a starting signal rather than final proof of ad effectiveness - periodically validating the biggest-spending channels with even a simple holdout test reveals whether the reported number reflects genuine incremental business impact or largely revenue that attribution happened to credit to an ad exposure that wasn't actually decisive.