Conversion lift: measure the real impact of paid campaigns

A retailer spends $45,000 on advertising, and the platform tells them each purchase cost $7.50. A conversion lift test reveals the real cost is $50: almost seven times more. If you don't measure incrementality, you're likely paying for conversions that would have happened anyway.

Is it really smart to trust the numbers the ad platform shows you?

No, not always. Standard reports attribute all conversions that follow ads over time to them, even those that would have occurred without them. Conversion lift is the tool that separates conversions actually caused by advertising from those that would have arrived anyway.

Each advertising platform — Google, Meta, TikTok — shows you a report with conversions attributed to your campaigns. The problem is that these numbers count all conversions that occurred after a user saw or clicked an ad, regardless of whether the ad actually influenced the purchase decision.

The conversion lift (literally: "increase in conversions caused by advertising") solves this problem with a controlled experiment. The audience is divided into two random groups: one sees the ads, the other does not. The difference between the conversion rates of the two groups is the lift, i.e., the real effect of the campaign.

Without this test, you risk optimizing budgets on channels that look like they're performing well, but are actually just catching people who were already ready to buy on their own.

How does a conversion lift test actually work?

The audience is divided into two random groups: the treatment group sees the ads, the control group does not. The difference in conversions between the two groups, scaled to the same size, is the incremental lift.

A test produces three core numbers. The absolute lift (or incremental conversions) is the difference between the conversions of the group that saw the ads and those of the control group, scaled to the same size. The relative lift divide that difference by the control group's conversions. The incremental cost per conversion — called iCPA by Google and CPIC by TikTok — divides total spend solely by incremental conversions.

When conversions have an economic value, the same logic produces the iROAS (incremental return on ad spend): how much you make for every euro spent, counting only the sales truly caused by the ads.

The source example clarifies everything. A retailer spends $45,000 on one million people, keeping 10% as a control group. The group that saw the ads produces 9,900 purchases; the control group produces 1,000, which scaled becomes 9,000. The absolute lift is 900 purchases, the relative lift is 10%, and the cost per incremental conversion is $50. The platform's standard report attributed 6,000 purchases with a CPA of $7.50: almost seven times more optimistic than the real data. Source: PPC Land .

How far are platform numbers from reality?

Often a lot. Meta detected a median underestimation of 31% with traditional attribution across 307 studies. For retargeting, the actual iROAS can be 40% to 70% lower than what the platform reports.

The gap between platform numbers and test results isn't a fluke. Meta looked at 307 Conversion Lift studies across 54 advertisers between March 2022 and November 2024 and found a median undervaluation of 31% with rule-based attribution. Source: PPC Land / Meta "suite of truth".

Measurement platform Haus analyzed 640 controlled experiments on Meta and found that average incremental lift of 19% on the brand's primary KPI. For retargeting campaigns — those that show ads to people who have already visited the site — the Real iROAS typically turns out 40% to 70% lower to the numbers reported by the platform. Source: Haus, July 2025.

The reason is simple: those who have already visited your site are already interested and would probably have bought anyway. Showing them an ad doesn't change much, but the platform counts it as an attributed conversion.

An Amazon Ads study on nearly 2,000 Meta campaigns found that without controlled experiments, observational models mess up the estimates of ad impact between 488% and 948% . Source: Amazon Ads / arXiv, August 2025.

Which platforms offer conversion lift and what do they require?

Google, Meta, TikTok, Pinterest, and Snap all offer conversion lift tools, with different requirements for budget, conversion volume, and test duration.

The biggest turning point for Italian SMEs came in May 2025: Google lowered the minimum budget to kick off a Conversion Lift Study from about 100,000 dollars to 5,000 dollars , adopting Bayesian statistics — a method that allows for reliable results even with less data compared to the traditional approach. Source: PPC Land / Google Marketing Live, May 2025.

This is a game-changer: a tool that until yesterday was reserved for big advertisers is now within reach for those managing medium budgets.

Three concrete scenarios: what changes for an e-commerce, a manufacturing company, and a professional firm?

Conversion lift changes budget decisions differently depending on the industry and the type of conversion measured. Here are three illustrative scenarios to understand how to apply it.

Illustrative examples: numbers are meant to show the reasoning, they are not real customer data.

Which channels really have the best incremental iROAS?

Data from 225 geo-based tests show that branded search — ads for your brand name — often has an iROAS below the break-even point, while CTV and Google Performance Max lead the rankings.

The Stella platform conducted 225 geo-based incrementality tests between August 2024 and December 2025 on DTC (direct-to-consumer) brands. The results overturn some common beliefs. The branded search — the ads that appear when someone is already searching for your brand name — recorded a Median iROAS of 0.70x , below the break-even point of 1.0x. Source: Stella, December 2025.

In practice: spending on branded search often doesn't generate additional sales, because those who search for your name would have bought anyway. This doesn't mean it should always be cut — it defends your position from competitors — but it should be evaluated with different eyes.

The overall median iROAS across all tested channels was 2.31x. This data pertains to DTC brands primarily on Shopify and is not generalizable to all sectors, but it offers a useful benchmark for those looking to compare their performance. To learn more about measuring ROI across different channels, the article on IRL events and creator marketing offers another perspective on measuring real impact.

Why are people talking about this now: is the market really changing its approach to measurement?

Yep. Privacy rules, the death of third-party cookies, and standard tracking getting less reliable every day are pushing more and more businesses toward incremental measurement.

According to Nielsen Annual Marketing Report 2025 , 85% of global marketers say they are very confident in their ability to measure ROI holistically. But only 32% actually measure spending across digital and traditional channels in an integrated way — down from 38% in 2024. In Europe, the percentage drops to 23%. There's a huge gap between the perception and reality of measurement.

A TransUnion survey cited by eMarketer in 2026 shows that 52% of US brand and agency marketers already use incrementality testing, with an additional 36% planning to increase investment in this direction. In retail media, 71% of advertisers consider incrementality the most important KPI. Source: Digital Applied / eMarketer / TransUnion, May 2026.

On the regulatory front, IAB and IAB Europe published on November 3, 2025 the «Guidelines for Incremental Measurement in Commerce Media» , the first official framework that classifies incrementality data provided directly by platforms as the lowest level of reliability: «hybrid proxies». Independent controlled experiments are at the highest level. Source: IAB / IAB Europe, November 2025.

This is also relevant for those managing campaigns on Amazon: as explained in the article on Amazon Ads measurement , the choice of measurement tools directly influences budget decisions.

What are the risks and limitations to know before trusting a test?

Conversion lift tests have significant structural limitations: the seller of the advertising often also manages the test, and a single test on one platform does not measure the overall effect of your advertising presence.

The most critical limit is structural: the platform selling the advertising space usually designs the test, manages the control group, and calculates the result. IAB and IAB Europe classify these tests at the lowest reliability level. The right question to ask is: who controls the control group and can the test be verified by third parties?

A second limitation concerns the partial view. A user in Meta's control group continues to see your brand on YouTube, in organic search, and in other channels. The test measures Meta's incremental contribution. in addition to everything else, not in isolation. This is helpful, but it doesn't tell you anything about the overall effect of your channel mix.

How to get started concretely: the steps to launch the first test?

It starts with feasibility checks, then configuring the test on the platform, waiting for the minimum recommended time, and reading the results with the three key numbers: absolute lift, relative lift, and iCPA.

What are the most common mistakes beginners make with conversion lift?

The most common mistake is comparing the test's iCPA with the CPA reported by the platform as if they were the same thing. They aren't: they measure different things.

• Launching a test without enough budget or conversion volume. A test with insufficient statistical power doesn't produce reliable results. Google now shows a feasibility rating: if it's "Low", wait.
• Modify the campaign during the test. Any changes to budget, targeting, or creatives will invalidate the comparison between the groups.
• Trusting a single platform's test to make decisions for your whole mix. Each platform only measures its own contribution. For a complete view, you need an independent marketing mix model.
• Ignore conversion lag. If your customers take weeks to decide (typical in B2B or high-value products), a 7-day test won't capture the real effect. Use at least 14 days.
• Cutting branded search without a geo-test. A low iROAS on branded search doesn't always mean it should be turned off: it might be protecting your position from competitors. Test it first on limited geographic areas.
• Accepting vendor results without verification. Always ask who's running the control group and if an independent third party can verify the method. IAB frameworks rank native platform tests at the lowest level of reliability.

How to tell if the test is working and the results are reliable?

Look at the statistical power of the test and the certainty level of the lift. On Google, a result is reliable when the certainty exceeds 90%. Below 50%, the platform shows 'No lift,' but this doesn't prove the campaign isn't working.

So, is it worth it for your company?

It is worth it if you spend at least €5,000/month on a single paid channel, have more than 150 monthly conversions, and want to understand if the budget is really generating additional sales or just capturing those who would have bought anyway.

A middle ground You are below the minimum budget or conversion threshold but have doubts about a channel's effectiveness: you can start with a manual geo-test (turn the campaign on/off in different areas for alternating weeks) or wait until you meet the minimum requirements.

Not yet if you have a very limited budget (under €2,000/month per channel), little conversion data, or very long purchase cycles without the possibility of extending the test: the risk of obtaining inconclusive results is high, and the cost of the holdout (10-30% of the budget not used for ads) is not justified.

How we see it at SHM

For us, conversion lift is not a technical tool for specialists: it's the answer to the question every entrepreneur should ask before increasing their advertising budget. «Is this money really generating sales that I wouldn't have had anyway?»

We believe AI and advanced measurement tools should be used to provide clarity, not add complexity. The problem isn't a lack of data — platforms produce it in abundance — but a lack of the right data. A standard attribution report tells you how many conversions followed an ad. Lift tells you how many it caused. It's a difference worth tens of thousands of euros per year for those managing significant budgets.

In our work with companies managing paid campaigns across multiple channels, we often see the same pattern: increasing budgets, seemingly great CPAs, but sales that don't grow proportionally. The first question we always ask is: have you ever tested what would happen if you turned off that channel for two weeks in a specific geographic area?

If you want to understand how to integrate incremental measurement into your strategy, you can explore our approach to Strategic AI integration , at business intelligence and data processing and to the google ads campaigns . Find all cluster articles Marketing and ADV performance to dive deeper into every aspect of paid measurement.

Frequently Asked Questions

The problem isn't how many conversions the platform shows you: it's how many of those conversions would exist without you.