Lift vs. incrementality: what the difference actually means for your budget

Step on a bathroom scale and you get a single, clean number: down three pounds since last month. That number doesn't explain itself. It could be water weight, muscle, or an actual change in body fat, and you'd need something more than the scale to know which one it was. Marketers hit the same wall with lift. A lift number, say a treatment group converting a point higher than a control group, tells you something moved, but not whether that gap would've shown up even without the campaign. That's the real incrementality question, and there's more than one way to go looking for that answer.

Looking for that answer in the right place matters a lot for brands. Budget decisions built on a lift number that's treated as if it settles the incrementality question tend to send spend toward channels that look good on paper but aren't fully earning it. That's especially true once a brand is running a test across several media channels and campaigns at once, where one test's results can end up steering budget for channels the test never actually touched.

Key takeaways

• Lift is the measurable number; incrementality is the underlying question of whether an outcome would've happened anyway.
• A holdout test is one common way to explore that question: it compares a treatment group that sees ads against a control group that doesn't, then reads the gap between them as lift.
• Holdout tests can be locally accurate but globally inaccurate: a result that holds for one campaign, region, or test window doesn't automatically scale to a full marketing budget.
• Test and control comparisons run into the retargeting credit problem, since they can't cleanly separate people who saw an ad from people who would've converted anyway.
• Halo effects, revenue a campaign drives indirectly through another channel, can slip past a narrow test, and that's true across the funnel, not just for top-of-funnel campaigns.
• A marketing mix model can answer the incrementality question directly, without a holdout test, by separating revenue that would've happened anyway from revenue marketing actually drove.
• Running a clean test across every media channel and brand campaign gets harder at scale, which is part of why brands often use a model alongside tests rather than relying on either one alone.

What is incrementality?

Incrementality is the concept sitting behind almost every effectiveness question marketers ask: did this actually work, or would it have happened anyway? It's less a metric you calculate and more a standard a campaign has to meet. When someone asks whether a campaign is truly incremental, they're asking whether a given sale, sign-up, or conversion happened because of that specific marketing effort, not because the customer was already headed toward a purchase.

That question matters for budget allocation because dollars spent on activity that would've converted regardless of any ad exposure aren't building new demand. You don't want to waste money on marketing efforts if the people you're reaching would have purchased even if those campaigns didn't exist. Marketers who can tell the difference between incremental revenue and revenue that reflects organic activity are in a much better position to decide where the next round of marketing dollars actually belongs, whether that's a paid media channel, a retail campaign, or a mix of both across several brands.

What is lift?

Lift is what you get once you've run that comparison: an actual number. It's the measurable gap in performance, conversions, sales, sign-ups, whatever outcome you're tracking, between an audience exposed to a campaign and a comparable audience that wasn't. If a treatment group converts at 4% and a control group converts at 3%, that one-point gap is your lift, and it's the same basic math whether you're measuring conversions on a single ad or across an entire media plan.

Lift is useful because it's concrete. A marketer can point to a percentage difference and say that's what moved. But lift on its own doesn't tell you whether that difference reflects genuine incrementality or just noise, seasonality, or an audience split that wasn't as clean as it looked. That's where understanding lift as a metric, and incrementality as the concept it's supposed to measure, starts to matter for how much weight you put on a single test's results.

How lift and incrementality work together

One common way marketers try to answer the incrementality question is by running a holdout test. The structure is pretty consistent, whether it's a small regional test or a full incrementality testing program run across several markets and media channels at once.

1. Split the audience. Marketers divide a customer base, region, or market into two groups: a treatment group that sees the campaign, and a control, or holdout, group that sees no ads, or a placebo version of the ad.
2. Track the outcomes. Both groups get measured against the same business outcome, whether that's conversions, sales, sign-ups, or another metric, over the same window of time. This is the step where a test actually measures something rather than just assuming it.
3. Calculate the gap. The difference in performance between the treatment group and the control group is the lift. If that gap holds up and isn't explained by something else going on in the market, it gets read as evidence of incrementality.

That structure is straightforward, which is part of why lift and incrementality get treated as interchangeable in practice. But the mechanics of a single test don't answer the bigger question of how much you can trust that result once you try to apply it to your entire marketing budget across every channel and brand you run.

Here's how the two terms break down side by side, since marketers and content writers alike tend to reach for them as synonyms even though they're doing different jobs.

What incrementality tests can't tell you

A well-run test and control group setup can tell you a lot, but treating it as the final word on whether a campaign or a channel actually works is where marketers tend to run into trouble.

A test result is locally accurate, but it isn't automatically globally accurate

A holdout test can show real lift for one campaign—in one region, over one test window, on one media channel—and still fail to represent how that channel performs once you scale the budget behind it or run the same tactic across an entire omnichannel footprint. What holds up in a two-week regional test doesn't necessarily hold once real budget decisions ride on it across every brand and channel in the mix.

Test and control groups struggle with the retargeting credit problem

Someone shown a retargeting ad who then converts might've been headed toward that purchase anyway. Because a test group and a control group can't cleanly separate people who saw the ad from people who would've bought regardless, incrementality testing isolates a lift number without fully resolving whether that lift reflects new demand or purchase intent that was already there.

Halo effects can slip past a narrow test entirely

A halo effect is impact (measured through revenue) that a campaign drives indirectly through another channel, like someone who sees a retargeting ad, doesn't click, and later buys through a search ad or in a retail store. That kind of activity doesn't show up cleanly in a single test's conversion count, and it's true across the funnel, not just for upper-funnel awareness campaigns. It's also a good reminder that a test measures one slice of media, not the full picture of how brands actually earn conversions.

Where MMM fits in

A holdout test isn't the only way to go after the incrementality question. A marketing mix model can answer it directly, without splitting anyone into a treatment group or a control group at all.

Instead of testing one campaign against a holdout audience, an advanced marketing mix model looks at a brand's ongoing revenue and spend across media, brands, and channels, and works out how much of that revenue would've shown up anyway, through seasonality, trend, and underlying demand, versus how much marketing actually drove. That's the same core question a holdout test is asking, would this have happened without the campaign, just answered from the whole business at once instead of from one isolated slice of it. Because the model runs continuously across every channel a brand uses, it can speak to incrementality at the scale an actual budget decision requires, not just for the one campaign, region, or test window a holdout test happened to cover.

NOTE: Not all marketing mix models are capable of this. Traditional MMMs struggle to separate marketing effects from baseline demand. You'll need an advanced MMM like Prescient's that doesn't fall prey to baseline leakage.

That doesn't make holdout tests worthless. Prescient's marketing mix model can fold a test's result in and show accuracy with and without it, so a brand can see whether that particular test lines up with what the rest of its data shows. If including it improves accuracy, that's a good sign the test caught something real. If it doesn't, a brand can still choose to trust the test and layer it in as a prior anyway. Either way, the model turns that call into something a marketer can actually see, instead of a guess.

That same model can also help answer the question a lift number raises next: what happens if you scale a channel's spend, or shift budget into it from other channels, while holding the total budget steady? That's a different question than whether a single test showed lift, and it's the one that actually determines where marketing dollars should go from here.

Where Prescient comes in

Prescient AI's marketing mix model answers the incrementality question directly by separating out what a brand's revenue would've been anyway, through seasonality, trend, and underlying demand, from what marketing actually drove. Brands don't have to run a holdout test to find out whether a channel is genuinely incremental. If they've already run one, Prescient can fold that result in and show whether it holds up against the rest of the data, and it tracks performance across tactics, including retargeting campaigns, so credit for a sale isn't decided by a two-group test alone.

That matters most for omnichannel brands trying to allocate budgets across dozens of channels at once, where a handful of isolated tests can't capture the full picture the way a model built on the whole marketing mix can. Book a demo to see these reports and the platform in action on a real brand's anonymized data.

FAQs

Is lift the same thing as incrementality?

Lift and incrementality are related, but they answer different questions. Lift is the number you get once you compare a treatment group against a control group. Incrementality is the broader question that number is supposed to answer: would this outcome have happened without the campaign? A campaign can show real lift and still leave the incrementality question only partly settled, especially if the test doesn't account for things like retargeting or halo effects.

How do you calculate incrementality lift?

Incrementality lift is calculated by comparing the outcome for a treatment group, the audience exposed to a campaign, against a control group that wasn't. Marketers typically track each group's conversion rate or performance over the same window of time, then express the difference as a percentage. A treatment group converting at 4% against a control group's 3% conversion rate, for example, works out to roughly a 33% relative lift, and the same math applies whether you're measuring a single test group or comparing conversion lift across several campaigns at once.

What counts as a good incremental lift percentage?

There's no universal benchmark, since a good incremental lift percentage depends heavily on channel, campaign type, audience size, and how the test itself was structured. Brands with tighter margins or smaller test groups sometimes see wide swings on a single test that look dramatic but aren't statistically reliable. A more useful gut check is whether a lift result holds up once it's checked against a marketing mix model, since a result that improves overall model accuracy is more trustworthy than a big percentage on its own.

Can a campaign be incremental without showing measurable lift?

Yes, and it's one of the trickier parts of incrementality measurement. A test with too small a sample, too short a test window, or a control group that wasn't cleanly isolated can fail to detect real incrementality even when a campaign is driving genuine new customers across several media channels. That's part of why a single test and control group study shouldn't be the only input into a bigger budget decision.

Do you need a control group to measure incrementality?

A control or holdout group is one way to measure incrementality, and it gives you a baseline for what would've happened without a specific campaign. It's not the only way, though. A marketing mix model like Prescient's can answer the same question by separating revenue driven by seasonality, trend, and underlying demand from revenue marketing actually drove, across a brand's entire business rather than one isolated test. Many brands use both, leaning on a marketing mix model as the ongoing read on incrementality and folding in test results as an extra check when they have them.

How is incrementality different from attribution?

Attribution and incrementality are trying to answer different questions. Attribution assigns credit for a conversion across the touchpoints a customer encountered, working backward from an observable outcome like a sale. Incrementality asks a more fundamental question: would that sale have happened at all without any marketing involved? A channel can look strong in an attribution model and still turn out to add little in the way of new, incremental revenue once it's actually tested.