Your campaign came in at 130% iROAS, clean, holdout-validated. The competitor's came in at 1,300% on four times your budget. Both are incremental. Only one is building next quarter's organic position. Incrementality is the truth about your dollar — but the truth about your market lives in another layer.
The argument, in a hypothetical with numbers
Two brands, same category, same retailer, same month:
- Brand A (you): USD 200K in retail media. Clean geo holdout, iROAS 130%. Real incremental sales: USD 260K. The regression closes with documented bias, aligned with IAB/MRC, goes up to the CFO with confidence.
- Brand B (direct competitor): USD 800K in retail media. iROAS 1,300%. Real incremental sales: USD 10.4M. Their MMM also closes clean.
Both campaigns pass their own incrementality test. Both get signed off in finance as "spend that pays back." But the retailer's ranker — Amazon's A9, Mercado Libre's, the VTEX engines behind Éxito and Jumbo — doesn't read your incrementality coefficient. It reads relative signals: sales velocity relative to the category, conversion relative to neighboring listings, review accumulation relative to the cohort, content score relative to retailer benchmarks. And every one of those signals just moved much more in Brand B's favor than yours.
Ninety days later, your weighted organic position on the top 30 keywords dropped four spots. Your campaign was 100% incremental. Your category drifted away from you. Both statements are true at the same time, and the first does not protect against the second.
This article is the autopsy of why.
What incrementality does answer — and why you must keep measuring it
Before attacking the limit, acknowledge the strength. Incrementality answers a question that no ROAS, no retailer dashboard and no badly-specified MMM can answer: would these sales have happened anyway?
- It separates "sales you'd have gotten anyway" from "sales that only happen because of paid placement." The canonical Blake, Nosko & Tadelis (2015) holdout at eBay is the founding case: nearly all brand-keyword spend by an established brand is non-incremental. Without measuring incrementality you never find out.
- It calibrates ROAS bias. Skai (2026) documented true iROAS varying between 253% and 1,609% on the same reported spend. The error can be an order of magnitude.
- It gives finance a defensible answer about the dollar spent. A coefficient with confidence interval and documented bias is the only thing that survives a serious audit.
- It resolves brand-keyword cannibalization, defensive spend, stockouts and the content ceiling — the four confounders a flat ROAS will not catch.
Anyone telling you incrementality is academic theater is selling you something else — usually a ROAS with extra steps. Incrementality is necessary. The problem is that it isn't sufficient.
Why incrementality alone isn't enough: the structure of the gap
Three reasons, in order of severity.
Reason 1 — The question incrementality answers is univariate
Your attribution model answers exactly this: "if we had not spent, would these sales have happened?". It's a counterfactual about your dollar.
It does not answer: "if the competitor had not spent, would these sales have happened?". And, more importantly, it does not answer: "are we gaining or losing relative ground in the category?". Those are multi-actor questions. Incrementality, by design, doesn't cover them.
The practical result: two brands can simultaneously sign off "incremental spend" and still be moving in opposite directions in the ranker. Your coefficient describes your slope; it doesn't describe the slope of the floor beneath your feet.
Reason 2 — The ranker rewards relative signals, not absolute ones
Modern retail rankers — Amazon's A9, Mercado Libre's engine, the VTEX rankers behind Éxito, Jumbo and Olímpica — take as input signals that are intrinsically relative: CTR vs. the search page median, conversion rate vs. neighboring listings, category-normalized sales velocity, content completeness against retailer benchmarks.
A campaign with 10× more incrementality is, by construction, producing 10× more of the signals the ranker rewards: 10× more real clicks, 10× more conversions, 10× more accumulated reviews in the period. The competitor isn't just selling more units this week — they're buying next quarter's ranker memory.
This isn't theory. The Dash et al. (2024) paper on Amazon shows that sponsored placements feed back into organic rank via clicks: paid investment today buys organic position tomorrow. The brand with the more incremental campaign isn't playing for this month's sales — they're building the next six months of position. And because the ranker has memory, that premium persists.
Your incrementality coefficient doesn't incorporate any of this. It measures the causal effect on your sales; it doesn't measure the effect on the signal ecosystem that decides your next quarter.
Reason 3 — The MMM doesn't observe competitor incrementality
Most observational attribution models (in retail media and beyond) include "competitor actions" as a variable — but operationalize it with a weak proxy: visible launches, generic promos, share-of-voice in communications.
Almost none include the competitor's actual incrementality, for the obvious reason that no one publishes their iROAS. The result is that your 87% incremental coefficient is measured against a baseline that already incorporates the competitor's moves, but doesn't tell you whether the competitor moved more or less than you in relative terms.
The iROAS confidence interval captures the precision of your causal estimate. It doesn't capture the trajectory of the category. Two different variances — and only one shows up in the report.
Where the argument has limits — steelmanning the other side
To be honest: the counter-counter-argument has valid cases.
"You don't have the competitor's actual incrementality — you can't measure what you don't observe." True in absolute terms. But you can measure their paid SOV per keyword, their relative sales velocity, their organic position delta, their launches, their content score, their review density. It's not the competitor's incrementality — it's something different: the signals the ranker read from the competitor during the same period. And those are observable, with disciplined operational scraping.
"If all actors are rational and all measure incrementality, the market sorts itself out." True in aggregate over the long run; false at the campaign-design level, where the premium goes to who acts first or spends more efficiently. Aggregate market efficiency does nothing for an individual brand that lost three rank positions this quarter.
"In highly fragmented categories, the 10× competitor effect dilutes." True. In long-tail categories with 40+ competitive brands, no single competitor has 10×. But in LATAM CPG — where the universe of large competitors on any category keyword is countable on one hand — the 10× scenario isn't theory, it's Thursday.
Where incrementality alone is fine: brands with already-consolidated dominance and no structural competitor; niche categories without a competitive auction; brand-discovery moments where the ceiling is awareness rather than competition. If you're in one of those cases, this article doesn't apply. If not, the combinatorial of Reason 1 + Reason 2 + Reason 3 is exactly your problem.
What gets reported to finance: two numbers, one conversation
The operational consequence is direct.
1. Report two metrics in every retail media review, never one. Incrementality coefficient (Layer 1) + weighted position delta vs. the top 2 competitors (Layer 2). If you must give a single number for the deck, give the ratio: "incremental lift per dollar relative to the category leader moved from X to Y."
2. Diagnosis when the two numbers conflict:
| Reading | Interpretation |
|---|---|
| iROAS high + rank dropped | Competitor won the incrementality auction. Investigate competitor SOV, relative content score, launches in the window. |
| iROAS borderline + rank rose | Defensive incrementality. Protect the spend — you're not winning dollars but you're holding position. |
| iROAS high + rank rose | The winning combination. Document and replicate. |
| iROAS low + rank dropped | Misdirected spend (saturated keywords, non-converting PDP, uncompetitive offer). Change the mix before changing the bid. |
3. Calendar the joint business plan against competitor patterns. The competitor's campaign cadence is observable. Using it to plan your windows (anticipating their escalation, defending through their peaks) is the only way to stop always reacting.
4. Draw the line between "spend ROI" and "competitive ROI." The first is a function of incrementality. The second is a function of incrementality weighted by relative competitor movement. Two distinct KPIs for two distinct roles: the first to CMO/CFO, the second to the commercial lead and the retailer's KAM. Collapsing them into one — typically dressed up as ROAS — produces the contradictory decisions you see every Q4.
Close
Incrementality is the truth about the dollar. Position is the truth about the category. Both are necessary; neither is sufficient.
"Is it incremental?" is answered by an MMM, a geo holdout, or a well-specified observational regression. "Are we winning?" is answered by a continuous competitive-position thermometer — weighted rank by keyword × retailer, competitor paid SOV, SKU-level availability, relative content score. The CFO's question — "is this investment making us win in the market?" — needs both answers, read together, not fused into a single number that loses the information that matters.
The brand that survives the 2026 cycle is the one that shows up to every conversation with two figures: "our spend is 87% incremental (Layer 1) and our weighted position vs. the top 2 competitors moved +0.8 points (Layer 2). We're better off." Or its opposite, if the opposite is true. Either honest reading is actionable. iROAS alone is not.
ePerfectStore.com lives in Layer 2 — the relative-position thermometer no retailer hands you and no MMM models. Your attribution model lives elsewhere. Living separately is fine. What can't happen is that only one of them lives.
Sources
- Blake, Nosko & Tadelis (2015) — Consumer Heterogeneity and Paid Search Effectiveness. Canonical paid-search holdout at eBay. The empirical foundation for why incrementality is indispensable. Econometrica.
- Dash, Ghosh, Mukherjee, Chakraborty & Gummadi (2024) — Sponsored is the New Organic. Algorithmic audit showing how sponsored placements feed back into organic rank via clicks. The central evidence for why relative incrementality matters more than absolute. arXiv 2407.19099.
- Sorokina & Cantu-Paz (2016) — Amazon Search: The Joy of Ranking Products. The official A9 paper on how the ranker incorporates relative engagement signals. Amazon Science.
- Skai (2026) — The 2026 State of Retail Media Measurement and Incrementality. 75% of brands list incrementality as priority #1; only 21% rate themselves "good" at measuring it; iROAS ranges 253%-1,609%. Skai.
- Bain & Company (2024) — No More Easy Money on the Side. The performance-era thesis: incrementality as new baseline, with investment consolidation into fewer networks. Bain Brief.
- IAB/MRC — Retail Media Measurement Guidelines (January 2024). Standard that accepts observational attribution provided bias is documented. Official IAB PDF.
Your MMM says your retail media is incremental. But do you know if the competitor's is more so? ePerfectStore.com is Layer 2 of the stack: weighted rank by keyword, competitor SOV, availability and relative content score — the signals the ranker actually reads.