Here's a number that should make every CPG marketing leader uncomfortable: 20-40% of retail media spend is non-incremental. That means one out of every three to five ad dollars you're spending on Amazon, Walmart Connect, or any retail media network is paying for a sale that would have happened without the ad. You're not buying growth. You're buying attribution reports that tell you the growth was yours.
This isn't a new finding. Third-party incrementality studies have been landing this number consistently for three years. What's new in 2026 is that the gap between what attribution reports say and what incrementality tests reveal has gotten wide enough that the brands running tests are making fundamentally different spending decisions than the brands relying on platform-reported ROAS. One group is optimizing against real signal. The other is optimizing against noise dressed up as data.
Retail media incrementality measures whether an ad-driven sale would have occurred without the ad. An incremental sale is one that the ad directly caused. A non-incremental sale is one the customer would have completed anyway — the ad claimed credit but didn't create demand. The distinction between the two is the difference between a marketing investment and a marketing tax.
This piece lays out the testing framework we use at Neato across our brand portfolio — the Incrementality Testing Cadence — and shows which ad types are genuinely incremental, which are mostly attribution capture, and how to build a quarterly test rotation that replaces reported ROAS with real ROAS.
What is retail media incrementality — and why is 2026 the reckoning year?
Incrementality has always mattered. What changed is the scale.
Retail media networks — Amazon Advertising, Walmart Connect, Instacart Ads, Kroger Precision Marketing — collectively passed $60 billion in US ad spend in 2025 and are projected to hit $75 billion in 2026. Every one of these networks uses attribution models that favor the platform. Customer searches "dog food," sees a Sponsored Product ad, clicks, buys — the platform counts an attributed sale. But the customer was going to buy dog food. The question is whether the ad changed which dog food they bought.
The 2026 reckoning: brands that test are discovering their actual ROAS is 30-60% lower than reported ROAS on certain ad types. Not because the platform is lying. Because attribution and incrementality measure different things.
Attribution answers: did the customer see or click the ad before purchasing?
Incrementality answers: would the customer have purchased without the ad?
A brand optimizing on attribution alone over-invests in ad types that capture existing demand and under-invests in ad types that create new demand. That's the structural mistake — and it compounds every quarter you don't test.
Which ad types are actually incremental?
Not all retail media spend is created equal. The incrementality gap varies dramatically by ad format, keyword type, and audience targeting. Here's what the data shows:
Ad Type | Typical Incrementality Range | What It Means |
|---|---|---|
Sponsored Products — branded keywords | 5-20% | Mostly capturing sales that would have happened anyway |
Sponsored Products — category keywords | 40-60% | Moderate incrementality; depends on competitive density |
Sponsored Products — competitor keywords | 55-70% | Higher incrementality; stealing share from competitors |
Sponsored Brands — branded | 15-30% | Low incrementality; reinforces existing awareness |
Sponsored Brands — category/non-branded | 60-75% | High incrementality; drives discovery and consideration |
Sponsored Brand Video | 50-65% | Strong creative impact drives genuine demand shift |
Amazon DSP — retargeting | 10-25% | Mostly capturing demand that already exists |
Amazon DSP — prospecting | 45-60% | Higher incrementality; new audience exposure |
The pattern is consistent: branded keyword ads have low incrementality because the customer was already searching for your brand. Category and competitor keyword ads have higher incrementality because the ad is influencing a decision the customer hasn't made yet. Prospecting campaigns outperform retargeting on incrementality because the audience hasn't already decided to buy.
The implication for budget allocation is significant. A brand spending 60% of its Amazon ad budget on branded Sponsored Products — a common pattern — is likely spending 60% of its budget on the least incremental ad type available. The reported ROAS looks great. The real ROAS is a fraction of it.
What is the Incrementality Testing Cadence?
The Incrementality Testing Cadence is a quarterly rotation of three test types that progressively reveal your true ROAS across ad formats and channels. Most brands run none of these. The ones who run all three make structurally different budget decisions.
Quarter 1: Geo holdout test. Split your market into matched geographic regions. Run ads in test regions, suppress in holdout regions. Compare sales velocity over 4-6 weeks. The difference is your incremental lift. Requires 500+ orders per region for statistical significance.
Quarter 2: Temporal holdout test. Two weeks ads-on, two weeks ads-off, two weeks ads-on. Compare sales during the off window to on windows, adjusting for organic demand trends. Simpler to execute than geo holdout. Limitation: can't isolate individual ad format effects.
Quarter 3: Audience holdout test (DSP-specific). Create a randomly selected control group within DSP campaigns that sees no ads. Compare conversion rates between exposed and control groups. The cleanest incrementality test because randomization controls for selection bias. Amazon offers this as a built-in feature for DSP advertisers above spend thresholds.
Quarter 4: Synthesize and reallocate. Compile results from the three tests. Reallocate away from low-incrementality formats (branded Sponsored Products, DSP retargeting) toward high-incrementality formats (category Sponsored Brands, competitor keywords, DSP prospecting). Re-test annually.
The cadence produces a rolling picture of true ROAS that improves each quarter. After two full annual cycles, a brand has enough data to make confident reallocation decisions rather than guessing.
What happens when brands actually test — and what goes wrong when they don't?
Brands that run the cadence and reallocate typically see one of two outcomes: same revenue with 20-30% less spend (cutting non-incremental branded keyword ads), or same spend with 15-25% more incremental revenue (reallocating to category Sponsored Brands and DSP prospecting). Either one is a material P&L improvement. Both are invisible to brands that never test.
The most common mistake is trusting platform-reported ROAS as ground truth. Amazon's attributed ROAS for branded Sponsored Products might show 8x-12x. The incremental ROAS is often 1x-3x. The gap is where the waste hides. The second mistake is testing once and assuming the results hold — incrementality shifts with competitive intensity and seasonality. A Sponsored Brand campaign showing 65% incrementality in Q1 might show 40% in Q4 when the category is saturated with holiday advertisers. The cadence is quarterly for a reason.
The Neato point of view
Neato runs the Incrementality Testing Cadence across every brand in our portfolio. It's not optional. It's how we make advertising allocation decisions.
The reason this matters in a 2P model specifically: Neato buys inventory and sells at margin. Wasted ad spend doesn't just reduce the brand's marketing efficiency — it reduces our margin. We're structurally incentivized to find and eliminate non-incremental spend because every dollar wasted on an ad that didn't create a sale is a dollar off our P&L. That's the alignment difference between a 2P partner and an agency. Agencies get paid whether the ad spend is incremental or not. We don't.
Our Impact analytics layer runs geo holdout tests on a rolling basis across the brand portfolio. When a test reveals that a specific ad format or keyword strategy is less than 30% incremental, we reallocate. Not in the next quarterly review. Within the current sprint. That's what having skin in the game produces — speed against waste.
