The cart was never just a UI element. It was the last first-party event before revenue — the place where intent became measurable on your infrastructure, in your units. Google just asked to host it.

At Google I/O and Google Marketing Live 2026, Google announced Universal Cart — one persistent cart that follows the shopper across Search, YouTube, Gmail, and Gemini — built on the Universal Commerce Protocol, an open standard that lets AI agents check availability, manage carts, and complete checkout against any participating merchant. Nike is in. TechCrunch read it as a privacy story: a cart that follows your entire shopping journey. For retail marketers, it's a different story. When I wrote the Retail Decision Stack, every layer of it stood on an assumption so obvious nobody would have thought to state it: the cart is your event, on your site, in your data. That assumption just acquired an expiration date.

What shipped — read as one decision, not three

Three announcements are circulating as three product notes. They're one architectural move.

The first is Universal Cart itself: a shopper adds your product from a YouTube video on Tuesday, sees it again in Gmail on Thursday, and checks out from Gemini on Saturday — without once loading your website. The cart persists across surfaces because it doesn't live on any of them. It lives in Google's commerce layer.

The second is UCP, the protocol underneath: a shared commerce language through which an AI agent can transact with any merchant that speaks it — availability, cart, checkout, order management. Open standard, genuinely; that part of Google's framing is fair. But protocols are like auctions: open rules, and still somebody hosts the floor. The agent completing the transaction is, overwhelmingly, going to be Google's, operating inside Google's surfaces, logging events into Google's reporting.

The third is Direct Offers inside AI Mode — ads that present an exclusive offer to a shopper who is ready to buy, with the purchase completing right there. This is the part that should reorganize how you read the other two: the ad is no longer the start of a journey to your store. The ad is the store.

Put the three together and the structural read is plain: the purchase funnel is being relocated into the platform's interior. This is the commerce twin of what I described in The Sealed Auction — there, the advertiser's inputs moved inside the prediction layer; here, the shopper's transaction does. Same building, new wing.

The Decision Stack assumed you owned the cart

The Retail Decision Stack replaced Cost Per Click with a hierarchy that runs in the units retail actually earns in — Cost Per Cart, retained margin, customer lifetime margin. Its working layer, the one practitioners actually manage to, is Cost Per Cart: media cost divided by margin-qualified add-to-carts. The metric works because the add-to-cart event has three properties no platform metric has. It's observed, not modeled — a discrete first-party event on your infrastructure. It's margin-legible — you know the SKUs in the cart and what they earn. And it's yours — no vendor sits between you and the event.

Universal Cart doesn't degrade those properties. It relocates them. The add-to-cart still happens — inside YouTube, logged by Google, reported to you through Merchant Center's UCP reporting at whatever granularity and latency the platform chooses. Your GA4 funnel — already quietly lying to you in its default configuration — doesn't go slightly darker. It goes dark at precisely the layer the Decision Stack measures. A purchase that begins in a YouTube ad and completes in Gemini is, from your website's point of view, a purchase that never happened to a visitor who never came.

The cart was the last first-party event before revenue. Universal Cart moves it inside the platform — and takes the Decision Stack's ground truth with it.

The steelman: why retailers will say yes, and won't be wrong

Be precise about the upside, because it's real and the retailers signing on aren't naive. Checkout friction is the largest single leak in retail conversion; a persistent cross-surface cart attacks it directly. Signals captured inside Google's environment are cleaner than signals leaked across a consent-walled, ad-blocked, cross-device web — the same logic that makes PMax's signal loop compound. UCP is an open standard, not a proprietary lock-in, and UCP-specific reporting in Merchant Center is a genuine attempt at transparency. And the demand is going where the agents are: if shoppers delegate buying to Gemini, being transactable by Gemini isn't optional. A Google commerce PM would say Universal Cart trades a measurement inconvenience for a conversion windfall — and for many retailers, this quarter, that trade will price out positive.

The problem isn't the trade. It's that most retailers will make it without noticing what was traded. The conversion lift arrives in the dashboard; the measurement loss arrives eighteen months later, in a budget meeting, when someone asks which media actually drives margin and the honest answer has become: the platform's reporting says the platform's media does. That's the correlated-evidence problem wearing a shopping cart — the system that sells you the media now also hosts the event that proves the media worked.

Move the ground truth to the order

The Decision Stack survives this. But its anchor has to move one layer down, from the event Google now hosts to the one thing Google cannot intermediate: the fulfilled order. Your order management system knows what was actually bought, at what margin, returned or kept, repeated or abandoned. No protocol relocates that. The retailers who keep their measurement through the agentic-commerce shift will be the ones who re-anchor on it deliberately, now, in four moves.

Anchor on order truth. Re-base Cost Per Cart's downstream layers — retained margin, lifetime margin — on server-side order data from your OMS and CRM, not on web analytics. The cart event becomes an upstream signal; the order becomes the metric. This is the retail version of the independent outcome definition every Cost Per Decision implementation starts with.

Reconcile UCP weekly. Run UCP order flows into the same evidence layer as site orders, and compare platform-reported cart and checkout metrics against fulfilled-order truth on a weekly cadence. The gap between what Merchant Center reports and what your OMS confirms is a number with a name: it's the price of the convenience, and you should know it before your QBR does.

Keep margin legibility. Universal Cart knows the SKU; it doesn't know your margin structure, return rates by category, or repeat-purchase curves. Feed margin-weighted values into the platform where it helps bidding — that's signal architecture — but keep the margin model itself on your side of the protocol. What you send becomes the platform's optimization target; what you keep remains your negotiating position.

Price the dependency before it prices you. Direct Offers and agent-completed checkout will grow as a share of revenue. Decide now what share of your transactions you're willing to have complete inside surfaces you can't independently measure — and what data symmetry (order-level detail, event latency, abandonment visibility) you require as that share grows. Retailers negotiated these exact terms with marketplaces for a decade. The agentic version of that negotiation starts this year, and the retailers who arrive with their own numbers will get terms; the ones who arrive with Merchant Center screenshots will get rates.

The org question underneath

Notice that none of those four moves is a media optimization. They're data architecture, finance reconciliation, and vendor negotiation — work that, in most retail orgs, no single team owns. The marketing team sees the conversion lift. The analytics team sees the funnel gap. Finance sees neither until the annual media review. Agentic commerce doesn't just relocate the cart; it relocates the marketing organization's accountability boundary, and orgs that route decisions the old way will discover the gap at the speed of their slowest committee. The retailers that handle this well will treat the cart's departure the way they'd treat any loss of a strategic asset: a named owner, a measured exposure, and terms.

Here's the version that survives the board meeting. For twenty years, the cart was free ground truth — so universally yours that no one thought of it as an asset. It is now becoming rented shelf space inside someone else's store, and the rent isn't paid in fees. It's paid in evidence. Take the conversion lift; it's real. But move your ground truth to the order before the cart finishes moving to Google — because the retailer that owns its outcomes can rent anything, and the retailer that rents its evidence ends up owned by it.

Share LinkedIn Post