The fastest-growing line in advertising is not a new audience or a new format. It is a toll. Retail media, now politely rebranded commerce media as banks and airlines join in, is the fastest-growing digital channel again in 2026, with Dentsu forecasting 14 percent growth in a trillion-dollar global ad year and US spend heading toward roughly $69 billion by eMarketer's count, with Amazon absorbing the overwhelming majority of it. Every retailer with checkout data has noticed that selling ads against that data carries software margins, and their CFOs have noticed harder.
It helps to see the channel from the retailer's side for one paragraph, because the incentive structure explains everything a brand experiences inside it. Retail runs on low-single-digit margins; ad networks run on margins that would make a software company blush. A retailer that converts even two percent of its revenue into ad sales can add more profit than a good year of merchandising, which is why every grocer, pharmacy, airline, and bank built a media network in the span of four years, and why the pressure on brands to fund those networks arrives through the merchant relationship, not the marketing one. The toll is not a media product that happens to be sold by your retailer. It is a margin product that happens to be denominated in media, and the seller's negotiating leverage is your shelf space.
None of this means brands should not pay the toll. Often you must: for endemic CPG and increasingly for anyone whose customer shops inside a marketplace, commerce media is shelf placement, and refusing to buy shelf has never been a strategy. What it means is that you should buy it the way you would negotiate with any counterparty whose incentive is to expand the toll, and most teams do not, because commerce media budgets frequently come out of trade or shopper funds where the media discipline never arrives. The channel grew up in the seam between two departments, and it is priced accordingly.
How to pay the toll without getting robbed
Start with incrementality, because closed-loop measurement is the channel's best sales asset and its biggest trap. The retailer grades its own homework: it serves the ad, hosts the purchase, and writes the attribution report. A sponsored placement on a search for your own brand name will post a spectacular ROAS made almost entirely of sales you already owned, and the network will renew that placement forever because it is the easiest number it will ever print. The question is never what the dashboard attributes, it is what disappears when you turn the spend off, and you find that with holdouts and geo splits, the same discipline I apply to any attribution system that grades itself. The practical split that keeps the buy honest: separate branded from non-branded placements in every report the network gives you, because blended, the branded ROAS launders the non-branded performance, and the blend is usually the default view for a reason.
Then read the supply chain, because history is rhyming. On-site inventory is finite, so the growth is moving off-site: retailer data buying audiences across the open web, through the same intermediary stack that already eats a third of programmatic budgets. An off-site commerce media buy is a programmatic buy wearing a retailer's badge, and it deserves the same log-level scrutiny and the same questions about fees. Ask what percentage of your dollar reaches an impression. The silence is informative.
Finally, put commerce media inside the margin conversation instead of beside it. Retailer ad fees, trade funds, and price concessions are one negotiation from the retailer's side of the table, and they should be one analysis on yours. A brand that funds a retail media network while losing the same points on trade terms has not bought media, it has restructured its margin. I do not present media plans before I know the margin, and commerce media is the channel where that rule earns its keep, because the seller of the media also sets your cost of goods sold. The organizational fix is unglamorous and decisive: one person or one meeting where the trade terms and the media terms are visible together, because the retailer's team already operates that way, and a negotiation where one side sees the whole board and the other sees half of it has a predictable winner.
Here is the medium-term thing worth watching, because it decides how much leverage you will have in three years: the networks are consolidating measurement standards slowly and reluctantly, and the brands funding them are the only constituency that can force the pace. Every incrementality requirement you write into a joint business plan, every refusal to accept self-graded ROAS as renewal evidence, moves the channel toward being priced like media instead of priced like leverage. The toll booths are real, the roads behind them are real, and some of them are worth every cent. Just remember who built the booth, and why.