How to Prove Incremental ROAS: A Working Methodology

Ask a room of retail media buyers whether their campaigns drive incremental sales and most will say yes. Ask them to prove it and the room gets quiet.

That silence is expensive. Buyers and finance teams now rank incrementality as the most important retail media KPI, per the ANA, yet in Skai and the Path to Purchase Institute's State of Retail Media research, 44% of CPG marketers name accuracy concerns as the top barrier to actually measuring it. Everyone agrees on the destination. Almost nobody trusts the map.

So here is a map. This is the working methodology for proving incremental ROAS, the same logic our platform runs daily for 7 of the 10 largest CPG brands in the world.

Why you can't get there from ROAS

The most common mistake is starting with attributed sales and trying to discount your way to incrementality. Take ROAS, apply a haircut, call it iROAS. It doesn't work, and the reason is structural.

ROAS starts from ad exposure and counts everything a click or impression touched. A shopper who was going to buy your product anyway, who has bought it every month for two years, still gets counted the moment she scrolls past your sponsored placement on the way to her usual purchase. The attribution is technically correct and causally empty. No adjustment factor can fix it, because the missing information (what she would have done without the ad) was never collected.

Proving incrementality means flipping the starting point. You begin with what was sold, all of it, and work backward to what caused it.

The methodology, step by step

Step 1: Assemble total sales, not attributed sales. Pull retailer sales data across every channel you can get: online, in-store where available, every retailer you sell through. This is the foundation, and it is where most measurement approaches quietly fail. Bid platforms optimizing off retailer-reported attribution never see total sales, so they are doing arithmetic on a partial ledger.

Step 2: Establish the organic baseline. Model what sales would have been with zero advertising. Seasonality, category trends, distribution changes, historical velocity. This baseline is the counterfactual everything else gets measured against, and it has to be built per product, because a hero SKU and a long-tail SKU have completely different organic gravity.

Step 3: Strip out merchandising effects. Price changes, promotions, coupons, ratings and reviews, organic search rank, share of shelf. The digital shelf moves sales constantly, and every one of those movements will masquerade as ad performance if you let it. A 20% discount running alongside a sponsored products campaign will make that campaign look brilliant. The discount did the work.

Step 4: What remains is incremental. After baseline and merchandising are accounted for, the residual lift is what advertising caused. Divide it by spend and you have incremental ROAS. Divide incremental profit by spend and you have iROI. These are the numbers a CFO can take to a budget meeting without an asterisk.

Step 5: Build response curves and retrain continuously. A single incrementality number is a snapshot. The useful version is a curve per campaign showing how incremental sales respond as spend rises, because that is what tells you where the next dollar should go. Campaign A may be saturated at current spend while Campaign B has headroom. Retrain daily and the curves stay honest as competitors, pricing, and seasonality move. One women's intimates brand ran more than 1,500 daily budget optimizations this way across thousands of SKUs and came away with a 63% iROI improvement worth $3.3M in annualized incremental sales.

Where do holdout tests fit?

Geo-holdouts and matched-market experiments are legitimate, and if you have never measured incrementality, a well-run holdout is a fine way to get your first defensible number. Their limits show up at scale. Tests take weeks, cover one campaign or channel at a time, require withholding ads from real customers, and expire the moment conditions change. Use experiments to validate the model, and use the model to run the business. The two together are stronger than either alone, a point eMarketer's own guidance on incrementality testing lands on as well.

What proof looks like in practice

When this methodology runs against real portfolios, the findings are rarely subtle. Church & Dwight, working with Wavemaker and Skai, fed daily causal signals directly into automated bidding and saw up to 309% incremental sales improvement and up to 122% iROI gains across TheraBreath, Batiste, and Arm & Hammer Laundry, all on a fixed budget. The work won The Drum's Grand Prix. Bayer set out to improve Amazon media efficiency by 10% while holding spend flat and delivered 32%. Neither result came from spending more. They came from finally seeing which spend was doing anything.

And the pattern the numbers keep confirming: swapping ROAS for incrementality as the optimization target typically produces 15-20% more sales from the same budget. The money was always there. It was just being graded on the wrong test.

The one-sentence version

You cannot prove incremental ROAS by adjusting attributed sales; you prove it by decomposing total sales into baseline, merchandising, and advertising, and letting the advertising claim only what is left.

If you want to see what that decomposition says about your own media mix, try our iROI calculator or get in touch and we'll run the assessment.