The fastest-growing line item in the ecosystem is buying attention from retailers. It has to be re-bought every year — and the rent increase is sold as efficiency.
Twelve times. Across 152 companies and a decade of dated calls, executives said “retail media” 12 times. That’s the finding — it needs no comparison.
Within constant sources there’s no 2020–2025 surge: omni-talk ran ~26→30%, unofficial-shopify-podcast sat at 0% every year 2021–2025, taste-radio 0% in 2024–2025. The pooled 0.7%→19.5% rise is harvesting, disclosed as such.
One caveat, and I’ll own it: Amazon’s ad revenue covers every advertiser — sellers, vendors, publishers, authors, per the 10-K. This is the toll booth’s total take, not CPG-only spend. And it’s global: Amazon doesn’t disclose its US retail-media slice, so it can’t be placed inside the $62B below — the two charts share a scale, not a scope.
Amazon is the biggest booth on the road, not the whole road. The wider market matters because the rent isn’t one landlord’s pricing decision — it’s the structure of the shelf. One scope note: Amazon’s $68.6B is global — every advertiser, every ad product — while the $62B is US retail media only. Amazon’s US retail-media slice sits inside that $62B, but Amazon doesn’t disclose it.
Frames open wallets: “growth” gets funded, “toll” gets negotiated. Practitioners sell the channel in the language that buys budgets — a vocabulary you almost never hear on earnings calls.
“Investment” in plain English: buying the consumer’s attention, close to the purchase.
The balance he describes is between two pockets of the same customer. On a podcast it’s candor; on an earnings call it would be a margin question — which is why it stays off the calls.
The placement is rented; whether it creates equity depends on what remains in consumers’ minds after the placement ends. The rent is the part you have to re-buy every year.
If they stopped spending tomorrow, what would still be there in a year?
One call makes the loop explicit: an efficiency program and retail-media ROAS, back to back. The savings fund the spend; the spend justifies itself as efficiency. The rent increase never has to call itself a rent increase.
One honest note: execs barely say “retail media” on earnings calls — 12 of 2,592. The efficiency justification for the channel is documented in practitioner language (825 transcripts in a topic-selected corpus of commerce and ecommerce shows), not on calls. Don’t generalize this call into a call-language finding.
“This includes activation of our brands in retail media to convey our superiority and value messages close to point of consumer purchase decision.”
“…that balance of media spend between national and retail media, paying the piper of the customer…”
“Our garden e-commerce performance was very strong, growing sales high teens, supported by improvements in retail media, return on ad spend or ROAS, and an expanded assortment.”
I keep a corpus of 5,834 unique CPG transcripts — earnings calls, podcasts, interviews, SHA-256 deduped. I swept 2,592 dated earnings calls (2016–2026) for the exact phrase “retail media”: 12 calls, 0.5%. The practitioner side — 825 transcripts — comes from a topic-selected corpus of commerce and ecommerce shows, harvested precisely because retail media is their beat. From those 825, I human-labeled a random 100 passages (seed 20260922) as growth-framed (41), neutral/explainer (51), or toll-framed (8).
Amazon’s advertising-services revenue is official 10-K fact ($68.6B in 2025, 2.2× 2021); its line covers every advertiser — sellers, vendors, publishers, authors — not CPG-only. EMARKETER US retail-media figures (~$62B by 2025) are third-party forecasts of mixed vintages — context, not my data, and not spliced into a single series. The P&G, MGA/Brave Commerce, and Central Garden & Pet quotes are verbatim from the published call/episode transcripts.
Honest limits, and I’ll own all of them. One: Amazon’s ad revenue is the toll booth’s total take, not CPG-only spend. Two: the two corpora were assembled differently — the earnings-call corpus is company-selected and topic-blind, the practitioner corpus is topic-selected — so cross-corpus shares are not directly comparable; the earnings-call absence claim stands on its own. The efficiency justification for retail media is documented in practitioner language, not on calls — execs barely say the phrase there, so don’t read the Central Garden & Pet call as representative of call language. Three: the framing sample is 100 passages, human-labeled — directionally clear, not precise. Every quote on this page comes from the conversation; every number from the official filing — never the same source for both.