Incremental Lift: The Metric That Separates Real Growth from Coincidence

Incremental lift measures the additional outcome your marketing activity actually caused, stripped of everything that would have happened anyway. It answers a deceptively simple question: if you had not run this campaign, what would have changed? The gap between what happened with your marketing and what would have happened without it is your incremental lift, and it is the closest thing to a true measure of marketing effectiveness that exists.

Most marketing measurement does not answer that question. It measures what happened, not what you caused. Those are very different things, and the gap between them is where marketing budgets quietly disappear.

Why Standard Reporting Hides the Real Question

Early in my career at lastminute.com, I launched a paid search campaign for a music festival. Six figures of revenue came in within roughly a day. The numbers looked extraordinary. The temptation was to declare it a triumph and move on. But even then, I had a nagging question: how many of those people were going to buy the tickets anyway? They had already searched for the event by name. They were in market. The campaign may have captured their purchase rather than created it.

That distinction, between capturing demand and generating it, sits at the heart of incremental lift. Standard reporting tools are built to show you what happened, not to isolate your causal contribution. A conversion dashboard tells you that 4,000 people converted after seeing your ad. It does not tell you how many of those 4,000 would have converted regardless.

This is not a minor technical quibble. It determines whether your marketing is actually growing the business or simply taking credit for growth that was already happening. When I was running agencies and reviewing client performance reports, I saw this problem constantly. Campaigns reporting strong ROAS while the underlying business was flat. Channels claiming credit for the same conversion simultaneously. Boards being shown dashboards that looked like proof of performance but were, in practice, proof of activity.

The wider landscape of attribution sits beneath this problem. If you want to understand why so much marketing measurement produces misleading signals, the piece on attribution theory in marketing covers the conceptual foundations in detail. Incremental lift is, in many ways, the practical answer to the attribution problem: instead of arguing about how to divide credit among channels, you test whether the channel contributed anything at all.

The broader question of what analytics can and cannot tell you is something I write about throughout the Marketing Analytics hub, which covers measurement frameworks, GA4, and the organisational habits that determine whether data actually drives decisions.

What a Proper Incrementality Test Actually Looks Like

The cleanest way to measure incremental lift is a holdout test. You split your audience into two groups: one that receives your marketing activity and one that does not. Everything else is held as equal as possible. After the test period, you compare outcomes between the two groups. The difference is your incremental lift.

In practice, there are a few ways to structure this. User-level holdouts work well in digital channels where you can suppress ads to a defined audience segment. Geo-based experiments work better for channels where user-level control is difficult, such as television, out-of-home, or broad display campaigns. You select matched geographic markets, run your activity in some and not others, and compare results. The matched-market approach requires careful selection of control regions, but it is more strong than relying on platform-reported attribution data.

A/B testing methodology is closely related here. If you are running experiments in GA4, the Semrush guide to A/B testing in GA4 covers the mechanics of how to set up controlled comparisons within the platform. The principles translate directly to holdout test design: clean split, sufficient sample size, defined success metrics before you start, and a test period long enough to capture meaningful signal.

The most common mistake I see in incrementality testing is running the test for too short a period. Two weeks of data in a channel with a long consideration cycle tells you almost nothing. If your customers typically take six weeks from first exposure to purchase, a two-week holdout will understate the true lift, sometimes dramatically. Define your test duration based on the actual purchase cycle of your product, not on the impatience of your reporting calendar.

The second most common mistake is not protecting the test from contamination. If your holdout group is exposed to your activity through other channels, or if there is a significant external event during the test period, the results become unreliable. This requires coordination across your media plan, which is harder than it sounds in organisations where channels are managed in silos.

The Channels That Fail Incrementality Tests Most Often

Not all channels are equally likely to produce genuine incremental lift. Some channels are structurally better at capturing existing demand than creating new demand. Understanding which is which is commercially important.

Branded search is the most obvious example. When someone types your brand name into Google, they already know you exist. They are likely already in the purchase funnel. Bidding on your own brand terms may protect you from competitor conquesting, and that is a legitimate reason to do it, but the incremental contribution to revenue is often very low. You are paying to intercept a customer who was already coming to you. I have seen clients running substantial branded search budgets that, when holdout tested, showed incremental lift close to zero. The revenue was real. The attribution was accurate. The incrementality was not there.

Retargeting has a similar problem. By definition, you are targeting people who have already visited your site or engaged with your brand. Some of those people will convert because of the retargeting. But many of them were already going to convert. The overlap between “people who were going to buy anyway” and “people in a retargeting pool” is significant, and platform attribution does not separate them.

This does not mean branded search and retargeting have no value. It means their value needs to be tested rather than assumed. The answer might be to reduce spend, restructure bidding, or accept a lower ROAS target on the understanding that the true incremental contribution is smaller than the platform reports. What it should not do is continue unchallenged because the dashboard looks good.

Upper-funnel channels often tell the opposite story. Display, video, and paid social campaigns targeting new audiences frequently show lower attributed ROAS but stronger incremental lift when tested properly. They are reaching people who would not have found you otherwise. The attribution model undervalues them because the conversion happens later, through a different touchpoint, and that touchpoint takes the credit.

This is one reason why email marketing, when done well, tends to show strong incrementality. You are communicating with people who have opted in to hear from you, and the timing of your message can genuinely influence when and whether they purchase. The HubSpot guide to email marketing reporting covers the metrics worth tracking, and incrementality sits above all of them as the measure that contextualises the rest.

Incrementality and the Affiliate Channel Problem

Affiliate marketing deserves specific attention here because it has a structural incrementality problem that is widely understood in the industry but rarely acted on. Many affiliate programmes are dominated by cashback, voucher code, and comparison sites that sit at the end of the purchase funnel. A customer has already decided to buy. They search for a discount code, find it on a cashback site, and that site claims the last-click conversion. The affiliate gets paid. The brand’s reported ROAS looks fine. The incremental contribution of the affiliate to that sale is often close to zero.

This is not a hypothetical. I have seen affiliate programmes that appeared profitable on a cost-per-acquisition basis but, when incrementality was tested, were largely paying commission on sales that would have happened anyway. The practical question of how to measure this properly is covered in detail in the article on how to measure affiliate marketing incrementality. The methodology matters, and the results often require a difficult conversation with affiliate partners about commission structure.

What GA4 Can and Cannot Do for Incrementality

GA4 is a useful analytics platform with genuine improvements over Universal Analytics, particularly in its event-based data model and cross-device reporting. But it is not an incrementality measurement tool. It measures what happened on your site and in your apps. It cannot tell you what would have happened in the absence of your marketing.

That is not a criticism of GA4 specifically. No analytics platform can answer the counterfactual question through passive measurement alone. The counterfactual requires active experimentation: a control group, a test group, and a clean comparison. GA4 can support that experimentation by providing the conversion data you need to compare outcomes, but the experimental design has to come from you.

It is worth understanding what GA4 does well and where its limits sit. The Moz overview of GA4 features worth using is a practical reference for getting more out of the platform within its actual capabilities. And it is also worth being clear about what GA4 cannot track at all. There are specific data types and user behaviours that fall outside its measurement scope, which the article on what data Google Analytics goals are unable to track covers directly. Knowing those gaps matters when you are designing a measurement approach that relies on GA4 as an input.

For incrementality specifically, the most useful GA4 features are audience segmentation and conversion event tracking, which allow you to define your holdout group and measure outcomes consistently across test and control. The Moz breakdown of GA4 audiences is a good starting point for understanding how to build the audience segments you need for a structured holdout test.

Incrementality in Emerging Channels

The incrementality question does not go away for newer or less conventional channels. If anything, it becomes more pressing, because the measurement infrastructure for newer channels is less mature and the temptation to rely on proxy metrics is stronger.

AI-generated content and AI avatars in marketing are a current example. Brands are investing in these formats without always having a clear framework for measuring whether they are producing outcomes beyond engagement metrics. The article on how to measure the effectiveness of AI avatars in marketing addresses this directly. The same incrementality logic applies: engagement is not lift, and lift requires a counterfactual.

Generative engine optimisation presents a similar challenge. As more search behaviour shifts toward AI-generated answers, the question of whether your GEO activity is actually driving incremental traffic and conversions becomes harder to answer with standard analytics. The piece on how to measure the success of generative engine optimisation campaigns covers the measurement approaches being developed for this channel. The incrementality question is live and largely unsolved, which is an honest place to start.

Making Incrementality Actionable Without a Data Science Team

I am aware that “run a geo-based holdout test with matched market selection” sounds like something that requires a team of econometricians and six months of planning. For large advertisers with complex media mixes, that level of rigour is appropriate. But incrementality thinking does not require that level of infrastructure to be useful.

When I was building out the performance marketing function at iProspect, growing the team from around 20 people to over 100, one of the disciplines I tried to embed was the habit of asking “what would have happened anyway?” before drawing conclusions from campaign data. It is a mindset before it is a methodology. It does not require a PhD. It requires the intellectual honesty to challenge a good-looking number rather than simply report it.

For smaller teams, practical incrementality testing can start simply. Pause a channel for a defined period and observe what happens to conversions. Run your campaign in some geographic markets and not others. Suppress ads to a randomly selected 10% of your retargeting audience for a month and compare conversion rates. None of these are perfect experiments. They all have confounds. But they are substantially better than assuming that everything your attribution model reports is genuinely incremental.

The goal is honest approximation, not false precision. Marketing does not need perfect measurement. It needs measurement that is directionally correct and honest about its limitations. Incremental lift, even when measured imperfectly, gets you closer to understanding what your marketing is actually doing than any attribution model running on the same data it is trying to evaluate.

Inbound marketing is a channel where this thinking is particularly valuable. Inbound activity is often diffuse and hard to attribute, but the question of what revenue it generates incrementally is answerable with the right experimental design. The article on inbound marketing ROI covers how to think about returns from inbound investment, and incrementality sits at the foundation of that calculation.

The Organisational Resistance to Incrementality

There is one more thing worth saying about incremental lift, and it is not technical. It is political.

Incrementality testing is genuinely threatening to people whose performance is measured by the metrics that incrementality testing often deflates. If you run a retargeting programme and your ROAS looks strong, you have a strong incentive not to run a holdout test that might reveal the ROAS is largely capturing demand rather than creating it. If you manage an affiliate programme and your CPA looks efficient, there is a real risk that an incrementality test reveals you are paying commission on sales that were already coming.

I have been in rooms where incrementality tests were proposed and quietly shelved because nobody wanted to know the answer. The data would have been useful. The political consequences of the data were unwelcome. This is a real dynamic, and pretending it does not exist does not help anyone.

The organisations that get the most value from incrementality measurement are the ones where the culture rewards finding the truth over protecting the numbers. That is an organisational question as much as a technical one. You can have the best experimental design in the industry, but if the results get ignored or buried because they are inconvenient, the measurement exercise has no value.

The first time I asked for a budget to build something and was told no, I built it myself. The lesson was not about coding. It was about not letting the absence of easy resources be a reason to avoid doing the right thing. Incrementality testing is harder than reading a platform dashboard. It produces results that are sometimes uncomfortable. It requires defending a methodology to stakeholders who would prefer a simpler story. But it is the right thing to do if you want to know whether your marketing is working.

If you want to go deeper on the measurement frameworks and tools that sit around incremental lift, the full range of topics is covered across the Marketing Analytics and GA4 hub, from attribution and channel measurement to the practical limits of the tools most teams rely on.

About the Author

Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.

Frequently Asked Questions