Optimization

Incremental ROAS: are your ads really driving sales?

The high ROAS in your dashboard often bills you for sales that were coming anyway. Incremental ROAS measures the revenue your ads actually create, here's how.

What platform ROAS doesn't count

The ROAS in your ad dashboard is really an attribution number: it counts conversions that happened after someone saw or clicked an ad. But some of those people would have bought even if they'd never seen the ad. When the platform credits itself with those sales too, ROAS looks higher than it is. Industry research shows that up to ~60% of conversions credited to paid ads can come from users who were already intending to buy.

~60%
Of conversions credited to ads may come from users who would buy anyway
30–70%
Typical gap between platform ROAS and incremental ROAS
2.3×
Median incremental ROAS reported across geo tests (industry)

The gap isn't small. Independent geo tests routinely find a 30–70% difference between reported ROAS and true incremental ROAS. A channel showing in the dashboard might actually be contributing 2×, the rest is revenue that was already in your pocket.

Platform ROAS vs incremental ROAS (by channel)Platform ROASIncremental ROAS (iROAS)8,42,1Retargeting111,8Brand search2,62,2Prospecting5,22,7PMax42,9ShoppingIllustrative industry values
The highest-attributed channels fall the furthest: retargeting and brand search shine on the platform and collapse on incremental value. Prospecting barely moves.
Attribution tells you what happened after someone saw an ad. Incrementality tells you what wouldn't have happened without it.The essence of measurement

What exactly is incremental ROAS (iROAS)?

Incremental ROAS (iROAS) divides only the extra revenue your ads generated by your spend: incremental revenue ÷ spend. It rests on one question: 'What would have happened if I'd never run this ad?' A practical estimate is: iROAS = platform ROAS × incrementality factor, where the factor (0–1) is how much of attributed revenue is genuinely ad-driven.

Which channels inflate the most?

Not all channels inflate equally. As a rule, the closer to purchase a user is captured, the more misleading the attributed ROAS:

  • Retargeting. Shown to people who already have items in the cart and already know you. Impressive ROAS, low incrementality: most were coming back anyway.
  • Brand search. Someone typing your name is already looking for you. Highest-attributed, often lowest-incremental channel (you're 'buying back' the organic result).
  • PMax / Shopping. Blends brand and generic traffic; the overlap inflates attribution. Without a breakdown the true contribution stays hidden.
  • Prospecting. New audiences who don't know you. Usually the most incremental channel; even if platform ROAS looks low, this is where real growth comes from.

See your channels' real contribution

Unify Meta, Google and GA4 in one panel; watch the total-revenue signal and before/after results.

Join the Beta →

How do you measure incrementality? 4 methods

They all share one idea: create a control group that doesn't see the ad, then compare the total outcome of the two groups. The difference is pure incremental lift.

  • Geo holdout. Match regions; cut/pause ads in one group and keep the other normal. Channel-agnostic and post-cookie friendly; it compares total sales.
  • In-platform conversion lift. Meta, Google and TikTok randomly split users into test and holdout (ghost ad / ghost bid). Easy to set up, but the platform still owns the measurement.
  • PSA / ghost holdout. The control group sees a neutral public-service ad instead of yours; mostly for upper-funnel video and brand campaigns.
  • Synthetic control. When there's no clean matched region (GeoLift, CausalImpact), it builds a weighted 'what-if' scenario from several markets and compares it to reality.

How to design a clean test

A clean incrementality test: 5 steps1Pick thechannelStart with thehighest-attributed:retargeting or2Build amatchedSimilargeo/audience;past sales trends3Change onevariableCut/pause thatchannel; holdeverything else,4Compare totalsalesRead theaccount's totalrevenue, not the
Whatever the method, the discipline is the same: one variable, a matched control, enough time, and the right metric.
  1. One variable: change only the channel under test; keep creative, bids, price and promotions constant.
  2. Enough power: aim for the conversion volume needed to detect a 5–10% minimum detectable effect at 80% statistical power.
  3. Right duration: at least 2–4 weeks; 4–8 weeks for products with a long consideration window.
  4. Right metric: read the account's total revenue, not the platform's attribution report.

How to read the result

When the test ends, produce a single number: the incrementality factor = test iROAS ÷ platform ROAS. This ratio tells you how much of the revenue that channel attributes is real.

Share of attributed revenue that is truly incremental42%Incrementality factorIllustrative, varies 20–90% by channel
In this example only 42% of attributed revenue is truly incremental. The other 58% was coming anyway, you're 'buying it back'.

The action is then clear: if the factor is low (e.g. heavy retargeting), cut the budget and shift it to more incremental channels, usually prospecting; if it's higher than expected, scale with confidence. You decide with evidence, not with the feel of attribution.

Where Ads Sensor fits in this process

Let's be honest: Ads Sensor doesn't run the randomized geo experiment for you, that field work stays with you. But it gives you the foundation that makes incrementality thinking practical. It unifies Meta, Google Ads, GA4 (plus TikTok and Criteo) in one panel and shows the account's total revenue and blended ROAS curve, the very signal a geo test relies on, not a single platform's report. On every recommendation you apply, it runs automatic before/after tracking, a lightweight, always-on pre/post measurement for each change.

The AI flags channels whose platform ROAS looks inflated (heavy retargeting, brand-search cannibalization) and places MER / blended ROAS next to in-platform ROAS, so you know which test to run first. Join the beta and start reading your channels' real contribution today.

Frequently asked questions

What's the difference between incremental ROAS (iROAS) and platform ROAS?
Platform ROAS counts every conversion that happens after an ad, that's correlation. Incremental ROAS counts only revenue that wouldn't have happened without the ad, that's causation. iROAS is almost always lower because it excludes customers who would have bought anyway.
Which channel inflates platform ROAS the most?
Usually retargeting and brand search. Both capture people who are already close to buying, so they show high attributed ROAS while their incremental contribution can be small. Prospecting campaigns are typically the most incremental.
Can you run an incrementality test on a small budget?
Yes. Geo holdout tests are now feasible for smaller accounts, platforms have cut minimum test budgets sharply. What matters isn't budget but enough conversion volume and a clean control group, usually over a 2–4 week window.
How long should a test run?
At least 2–4 weeks for most tests; 4–8 weeks for products with a long consideration window. You need enough data to detect a 5–10% minimum detectable effect at 80% statistical power.
Does Ads Sensor run the geo test for me?
No, you run the randomized field experiment. Ads Sensor provides the foundation it relies on: a cross-platform total revenue/ROAS view, automatic before/after tracking on applied recommendations, and flagging of channels whose platform ROAS looks inflated.

Make your next budget call on evidence, not attribution

Ads Sensor turns incrementality thinking into a habit with a cross-platform total view and before/after tracking on applied recommendations.

Join the Beta →