Same campaign, 54 different iROAS numbers. Here's why.
Albertsons Media Collective, Ovative, and Northwestern's Kellogg School just put incremental ROAS through the wringer. Without an industry standard, the results swing wildly. Quoting it without the fine print? Let's fix that.
Plain ROAS takes credit for every sale near an ad, earned or not. Incremental ROAS was meant to count only the sales the ad actually drove.
How iROAS actually works
To know what the ad caused, you need to know what would've happened without it. Nobody can rerun reality, so analysts fake it: they match shoppers who saw the ad against similar shoppers who didn't, then call the gap between the two groups "incremental."
That matching step is a judgement call, not a fixed formula. There are dozens of legitimate ways to build the comparison group and calculate the gap.
Same campaign, different verdict
|
6.5x average swing in iROAS output for the exact same campaign, just from changing methodology |
83% of campaigns flipped result entirely, winner under one method, loser under another |
54 different iROAS values produced from a single campaign, depending purely on measurement choices |
Not a different campaign. Not different data. Just a different way of doing the maths, and a completely different verdict.
The label is doing too much work
“An iROAS number means very little without the method behind it and a benchmark to compare it to. What makes it useful is consistency. We measure every campaign the same way and build our benchmarks on that same method, so when a campaign lands above or below benchmark, that's a real signal. It starts a conversation about why. It doesn't end one. Every brand is really asking one question: did it sell? You don't answer that by picking the most flattering of 54 numbers. You answer it by measuring every campaign the same way, every time, and comparing like with like. We'd rather give a brand an honest number and dig into why it landed there than go hunting for the method that makes it look good”.
Noah Antoun Data and AI Architect, QSIC
What does this look like in the real world?
We measure every campaign two ways, and they're built to answer different questions.
|
Trial and control Trial and control compares stores playing the ad against matched stores that aren't, across the whole flight. It's a wide-angle lens. |
Moment lift Moment lift compares shoppers who checked out in the minutes after an ad played with shoppers in the same store who didn't hear it. That's a microscope. |
Take an energy drink campaign we ran in convenience stores earlier this year. Shoppers who heard the ad bought the product about 10% more than comparable shoppers who didn't. Across the whole flight, stores playing the ad grew that product about 4% more than matched stores that weren't. Same campaign. The wide-angle number counts the shopper who heard it Tuesday and grabbed a can on the spot. It also counts them coming back to buy again, and it counts every shopper who walked in when no ad was playing. That's why the percentage is smaller. Put 10% next to 4% without that context and it looks like one of them must be wrong. Neither is. They're answering different questions, and if you swap between them depending on which looks better, the number stops meaning anything.
QSIC is doing things differently
It's why QSIC leads with methodology, not a headline stat. Our measurement links audio to real, closed-loop in-store transactions, and we show the mechanism behind every result rather than handing over a number and asking you to trust it. When one campaign can produce 54 defensible answers, showing your working is the actual credibility play.
Want to see how QSIC Intelligence measures incrementality, methodology and all?
Talk to our teamSource: "Retail Media iROAS Demystified," Albertsons Media Collective, Ovative Group and Northwestern Kellogg School of Management, March 2026. Full paper.
