Lift Studies for E-Commerce: How to Know If Your Campaigns Are Actually Working

Here's a question most brands can't honestly answer: if you turned off your best-performing channel tomorrow, would revenue actually drop?

Most performance marketers assume yes. Their attribution dashboards show healthy ROAS, the channel logs attributed conversions, and it looks indispensable. But attribution models — whether last-click, linear, or data-driven — measure credit allocation, not causality. They tell you who got credit for a sale, not whether the campaign caused it. Lift studies do the latter. For brands spending serious money on commerce media, affiliate, or paid social, they're the most underused measurement tool in the stack.

What a Lift Study Actually Tests

A lift study (also called a holdout test or geo experiment) measures the incremental impact of a campaign by comparing two groups: people exposed to your advertising and people who weren't. The difference in purchase behavior between those groups is your lift — the revenue you can causally attribute to the campaign, not just correlate with it.

The mechanics vary by approach:

• Conversion lift tests (available natively on Meta, Google, and TikTok) randomly split eligible audiences into exposed and holdout groups at the platform level.
• Geo holdout tests identify matched geographic markets and run campaigns in some regions but not others, then compare purchase outcomes over a defined window.
• User-level holdouts suppress ads for a randomly selected percentage of your audience and measure their behavior against the exposed group.

The goal in every case is the same: create a clean counterfactual. What would have happened without the campaign? That's the number your reported ROAS can never give you.

If your reported ROAS is 4x but your measured lift is 1.2x, you're not running a profitable channel — you're paying for customers who were already going to buy.

Why Attribution Models Alone Fall Short

Attribution models are built to assign credit. They're useful for understanding the customer journey and making relative budget decisions within a channel. But they can't tell you whether your spending caused a purchase or simply touched it on the way to an inevitable conversion.

The problem is structural. Every attribution model — even the most sophisticated algorithmic one — can only work with the touchpoints it observes. It cannot account for organic demand, brand awareness building, or the simple reality that a segment of your customer base would have converted anyway. High-intent buyers already deep in the funnel will often click a retargeting ad seconds before purchasing — a purchase they'd already decided to make. Your model records a conversion. Your lift study records no incremental impact.

This dynamic is especially pronounced in channels with inherently late-funnel audiences: affiliate cashback and coupon sites, branded search, and retargeting. Reported ROAS in these channels tends to look strong precisely because the audience is already close to buying. Actual incrementality is often substantially lower.

Running a Lift Study Without an Expensive MMP

You don't need a sophisticated measurement platform to run a meaningful lift test. A few practical approaches most brands can execute today:

• Start with geo holdouts. Identify 10–15 pairs of matched DMA markets with similar demographics, historical conversion rates, and seasonal patterns. Run campaigns in half, pause in the other half for four to six weeks, and compare conversion rates. Simple, interpretable, no third-party tool required.
• Use platform-native lift tools. Meta Conversion Lift, Google Conversion Lift Studies, and TikTok Conversion Lift are accessible to most advertisers and don't require MMP integrations. They're imperfect, but they're directionally reliable and far better than reported ROAS alone.
• Test one channel at a time. Running lift tests across multiple channels simultaneously creates interference and muddies results. Clean, isolated tests are more actionable than comprehensive but noisy ones.
• Size your holdout appropriately. A 10% holdout on a campaign spending $50k per month typically provides enough statistical power for a reliable result. Too small and results are noisy; too large and you're burning budget unnecessarily.

Interpreting Results and Acting on Them

A lift study produces one of two valuable outcomes: it validates your spending or exposes a problem. Both are worth knowing.

Strong lift — exposed users converting at meaningfully higher rates than holdout users — gives you causal evidence to invest more in the channel. Weak or negligible lift means you have a budget reallocation problem that's better identified now than at end-of-quarter when targets are missed.

The right response to a low-lift result is rarely to kill the channel outright. It's usually to reshape the strategy within it: shift budget from retargeting to prospecting, move spend from coupon publishers to editorial content partners, or tighten audience targeting to reach genuinely unconverted users rather than those already in-market.

At Quantum Digital, lift studies inform every significant budget recommendation we make. Reported metrics tell you what happened. Lift tells you what you actually caused. Brands that optimize on the former while ignoring the latter end up spending more each quarter and growing less — and the attribution dashboard makes it look like everything's working fine right up until it isn't.