Skip to main content

Impersonally

🏆 Winners of the People’s Choice Award at TravelTech Show 2026              Read our latest research →

Incrementality

A causal measurement that isolates the conversions a campaign actually caused, rather than those it would have received anyway, revealing how much attributed revenue is truly incremental and how much spend is wasted on demand that already existed.

What is Incrementality?

Incrementality measures the causal lift a marketing channel produces over what would have happened without it. Google Ads Help describes it as “knowing exactly what happens because of your marketing, and what would not have happened otherwise.” When a customer converts after seeing a brand search ad, an automated campaign, or an affiliate link, the platform reports that conversion to the channel as if the channel caused it. Incrementality asks the harder question: would that customer have converted anyway? It separates the demand the marketing actually created from the demand it merely intercepted on the way to a purchase that was already likely.

Incrementality sits next to three concepts that answer different questions. Attribution assigns credit for a conversion that happened, regardless of cause. ROAS divides revenue by ad spend, but that revenue may be incremental or not, so a high ROAS can be entirely non-causal. Impression share measures whether your ad was eligible to show, not whether it changed behavior. The boundary that matters: attribution tells you who gets credit, incrementality tells you what actually happened because of the marketing.

You are encountering this term now because measuring it became affordable. Google says it lowered the minimum cost of an incrementality experiment from a previous high of $100,000 to a minimum of $5,000, according to Think with Google in February 2026. At the same time, Performance Max and other automated campaigns can serve on your own brand searches, and affiliates can be credited with brand traffic, so the gap between credited and caused revenue is easy to miss. Performance marketers can now run the experiments that were previously out of reach.

How Incrementality works

Isolate the causal effect

Google says its incrementality experiments provide “the precise, causal data for specific campaigns and channels.” The test compares a treated group that saw the ad against a control group that did not, holding everything else constant. The difference between the two groups is the lift the campaign actually produced, stripped of everything else that happened in the period.

Separate credit from causation

Attribution assigns credit for a conversion that happened; incrementality asks whether it would have happened without the ad. A brand search ad can carry high attributed ROAS and low incrementality at the same time, because people running a branded search were already looking for it. The click and the purchase intent are correlated, so last-click or data-driven models hand the sale to the ad even when the customer would have clicked the organic result and bought anyway.

Factor in competitor presence

The incrementality of a brand ad depends on who else is on the results page. With no rival bidding on the brand term, the organic link captures most clicks anyway, so the paid ad adds little. With rivals brand bidding on your name, the brand ad keeps clicks that would otherwise go to competitors. The two large field studies reconcile on this point: the answer changes with the competitive situation, so a single number does not describe brand search incrementality universally.

How to measure Incrementality

Run a true holdout

Every valid test needs a holdout: a group that could have seen the ad but did not. Without it, any comparison of periods mixes the ad’s effect with seasonality and demand changes. Google’s examples of what such tests separate include shoppers who saw a YouTube ad versus eligible shoppers who did not, search-driven responses versus organic brand-seeking, and new customers versus existing ones. Compare conversion rates or revenue between the treated and holdout groups over the same window.

Test brand search against the competitive field

Do not assume brand search is non-incremental without checking who else bids on your terms. Simonov, Nosko, and Rao, in Marketing Science (2018), ran large-scale Bing experiments covering thousands of brands and found that with no competitors bidding, a brand’s own ad gave a small 1-4% lift in clicks, but when the brand did not advertise on its own name, competitors took 18-42% of clicks. Run your test while competitors are active, or the result will not reflect your real defensive value.

Audit affiliates on brand traffic they claim

Check whether affiliates are earning commissions on conversions from customers already searching your brand. A coupon site or an affiliate bidding on your brand terms can be credited with a sale the customer was already making, so its commissions look earned while adding little. Test it with a holdout: pause a partner’s brand-term traffic for a test group and see how many of those sales still arrive through other routes.

ImpersonAlly shows which conversions affiliates actually drove and which they intercepted, so commissions are paid on earned sales rather than on demand you created.

How to read Incrementality

Read the eBay field experiment correctly

Blake, Nosko, and Tadelis, in Econometrica (2015), ran large-scale field experiments at eBay and found that brand keyword ads had no measurable short-term benefits. Because search clicks and purchase intent are correlated, experimental returns from paid search were a fraction of non-experimental estimates. On non-brand keywords, new and infrequent users were positively influenced by ads, but frequent users whose purchases were not influenced accounted for most of the spend, making average returns negative. The non-brand result is the more general lesson: an average return can hide who was actually influenced.

Let competitor presence change your interpretation

Simonov, Nosko, and Rao concluded that defensive brand advertising has a strongly positive return when competitors bid on the brand term, and argued that eBay, a strong brand facing no competitors, was not typical. Read the two studies together: with no rival, your organic link catches most clicks anyway; with rivals, your ad keeps clicks that would otherwise go to them. A low incrementality reading in a no-competitor test does not predict what happens when competitors enter.

ImpersonAlly monitors your brand’s ad presence continuously, showing who is bidding on your name, what they are running, and when it started.

Treat results as local to channel, period, and market

A result from one test window does not transfer automatically to another season, market, or competitive situation, especially on brand terms where competitor presence changes the answer. Do not apply a Q4 brand search incrementality number to Q1, or a US result to a new geography, without running a fresh holdout in that context.

Common mistakes with Incrementality

Reading attributed ROAS as incremental

Attributed ROAS divides revenue by spend, but that revenue includes sales that would have happened anyway. Treat it as a credit ledger, not a causal measure, and validate with a holdout before defending budget with it.

Cutting brand search ads on one no-competitor study alone

A single field experiment found no short-term lift for brand ads in a market with no rival bidding, but that context is not typical. Check competitor presence first, or you will cut defensive spend that protects clicks from rivals.

Comparing before and after without a holdout

Time-based comparisons mix the ad’s effect with seasonality and demand shifts. Always run a concurrent holdout group, or the result measures the calendar, not the campaign.

Applying one test result to every market and season

Incrementality varies by channel, period, and competitive situation. Run a fresh test per market and season instead of reusing a single number across all of them.

Paying affiliate commissions on demand you already had

Commissions on brand-term or coupon traffic often pay for sales the customer was making anyway. Audit affiliate-attributed conversions against a holdout, and use commission reversal where your program terms allow it for sales that were never earned.

Terminating all affiliates when only brand bidders are the problem

Blanket termination cuts legitimate incremental partners along with the ones doing affiliate hijacking on your brand terms. Prune the specific affiliates bidding on your trademark and coupon terms instead of ending the whole program.

Where Incrementality is heading

The cost of causal testing is falling, which moves incrementality from an annual audit to a standard part of campaign management. As experiments become cheaper and more automated, more teams will hold out control groups routinely rather than relying on attribution models. The constraint will shift from affording the test to affording the dark periods: every holdout means deliberately not showing ads to some customers, and that trade-off will need justification in budget reviews.

Automated campaigns are the next test case. When these systems serve on own-brand searches, they report the conversions of people already looking for the brand, so their reported results overstate what they add. The gap between credited and incremental revenue will widen unless teams explicitly hold out brand terms from automated campaigns. Any system that optimizes on attributed conversions has a reason to find that existing demand, which is why brand traffic is the first place to run a holdout on an automated campaign.

What will stay hard is the counterfactual itself. No model can observe the same customer both with and without the ad, so every incrementality estimate rests on a holdout assumption that the control group behaves like the treated group would have. Seasonal shocks, competitor moves, and price changes can break that assumption. Technology can make the test cheaper and faster, but it cannot remove the need for a control group or the uncertainty around what would have happened otherwise.

Further reading

Frequently Asked Questions

What is the difference between incrementality and attribution?

Attribution assigns credit for a conversion that happened; incrementality asks whether the conversion would have happened without the ad. A brand search ad can show high attributed ROAS while having low incrementality, because the customer was already searching for the brand. Attribution tells you who gets credit in the ledger; incrementality tells you what the marketing actually caused.

Are brand search ads incremental?

It depends on competitor presence. Blake, Nosko, and Tadelis found no measurable short-term benefit for eBay's brand keyword ads. Simonov, Nosko, and Rao argued eBay, a strong brand with no competitors bidding, was not typical, and found that when a brand did not advertise on its own name, competitors took 18-42% of clicks, which makes the brand ad defensively incremental. Test in your own competitive context rather than assuming a universal answer.

How much does an incrementality test cost on Google Ads?

Google says it lowered the minimum cost of an incrementality experiment from a previous high of $100,000 to a minimum of $5,000, according to Think with Google in February 2026. That is Google's stated minimum, not a quote for your account: what a test costs in practice depends on the campaign and how much traffic the holdout needs to give a usable answer. On brand search, the hidden cost is the holdout itself, the customers who do not see your ad while the test runs.

Why does my brand campaign show high ROAS if it is not incremental?

Because ROAS divides attributed revenue by spend, and attribution credits the brand campaign with customers who were already searching for you and would have converted through organic search anyway. The high ROAS reflects correlation between brand search intent and purchase, not causation. Run a holdout to see how much of that revenue disappears when the ads are off.

Metrics & Impact

Your Brand Protection Ally

Are your Ads being Impersonated by others?

Use the Fraud Tool on our homepage to check your domain — free, in real time.

Check Here →