Citi and McDonald's announced media networks on the same day. Only one can see the sale.

On 23 September 2026, two companies that are not media businesses announced that they now are. Citi launched Citi Commerce Media, built on 6.5 billion card transactions a year across more than 700 spending categories and more than 70 million US customers. The same day, at its investor day, McDonald's presented McDonald's Media Network and an aspiration to build it into "a billion-dollar business across the McDonald's system over time", with nearly 220 million loyalty members active in the last 90 days behind the pitch. Both will be filed under one heading: everyone with an audience now has a media network. The heading is accurate and tells an advertiser almost nothing, because the two crowds are different kinds of number.

On the architecture the two launches look alike. Each owner builds on its own customer data. McDonald's runs its own screens, starting with the drive-through boards, and sells the placements itself. Citi agreed in August to acquire Kard, a commerce media and rewards platform; once the deal closes, it would bring more of the targeting and measurement work inside Citi. Neither sells the underlying records: brands buy placements, the transaction file stays home. On configuration and on the boundary both are owner-enriched builds, and in both the rake is owned. If the five decisions stopped there, the two networks would look like the same business.

They separate at the first decision, the buyer's. The buyer of a media network is the brand that pays for the placement, and the decision that brand is paid to make is where next quarter's budget goes. That decision turns on one fact: whether the campaign sold anything. A network earns more than a screen price only when its data can answer that question for the brand paying.

McDonald's records what its own customers order at McDonald's. For a brand whose product is on the menu, a soft drink or a sauce, that record answers the question, and the answer is worth a premium. For a car insurer on the drive-through board, McDonald's sees a car in the lane and a burger on the receipt. It does not see a policy. To tell the insurer what happened next, McDonald's needs the insurer's own sales data and a way to match it to the people who saw the ad, borrowed campaign by campaign. For every advertiser outside the menu, the network sells attention at a high-frequency location, priced against billboards and other screens, with a data story on top that only a short list of suppliers can use without lending their own.

Citi sees the other half. A cardholder who buys a sofa, a flight or an insurance policy on a Citi card leaves the purchase in Citi's records, whoever the merchant was. The record is not the whole wallet, only what goes on the card, and it shows the merchant and the amount rather than the basket. It is still a record of the sale, for brands Citi has never met. Citi can tell a brand whether the people who saw its offer bought from it afterwards, which is what its launch calls closed-loop measurement. The record is wide where McDonald's is deep. What Citi lacks is a surface built for shopping: its own are Citi.com and the Citi Mobile app, alongside paid media elsewhere, and people open a banking app to check a balance. Kard is the likely answer to that weakness. An offer attached to the card itself needs less attention than a billboard does, because the card is present at the purchase it is trying to cause.

I ran a version of this at Reward with RBS in the UK, and the lesson that held was that the bank's offer earned its price at the till, not in the app. In February, Rezolve AI paid $230 million in cash for Reward.

Value goes where the work is, and in a media network the work is closing the loop between the ad and the sale. McDonald's has the better surface and a narrow record. Citi has the wider record and a thinner surface. Advertisers pay for the loop, and a thin surface can be worked around: an offer that sits in the card's own rewards needs little audience to work. The drive-through board has the opposite problem. The surface is real, and for most advertisers the sale it would need to see sits on someone else's books, available only on loan.

The billion-dollar aspiration says the same thing in money. The pilot started in August on 450 company-owned US restaurants, out of roughly 14,000 in the US, most of them franchised and not yet in the network. Reaching the number across the system means signing franchisees into a revenue split and selling screen time in volume, which is a media sales business. That is a legitimate business. It is a different business from the one the word "commerce" promises the brand, and unless the loop closes for advertisers off the menu, the price per impression will settle where screens settle.

The rush will produce two kinds of network under one label: those that can show a brand its own sale, and those that can show it a crowd. The first prices on outcomes and the second on reach. Borrowed data turns the second into the first only for as long as the advertiser agrees to lend it. When a company tells you it has a media network, ask whose sales it can see.

An audience is what you sell when you cannot see the sale.