Closing the Loop Outside the Walled Gardens: Who Is Actually Doing It
A survey of every serious attempt to rebuild the closed feedback loop on the open web, and an honest read on which ones are real.
Nate Woodman ยท July 24, 2026
A walled garden owns six links in a chain: identity resolution, decisioning, delivery, engagement measurement, conversion attribution, and the feedback of all of it into the next model version. That closed loop, not any single algorithm, is why the gardens outperform the open web. So the right question about the open web's future is concrete: who is actually rebuilding the loop out here, link by link, and which of those efforts are real versus marketing.
I have been tracking this across categories for the past year. Here is the honest map.
The identity spine has a neutrality problem
Identity resolution is link one, and without it nothing downstream closes. LiveRamp built the open web's dominant deterministic identity layer and then, in May 2026, Publicis bought it for roughly $2.2 billion. Whatever you think of the deal, the neutrality posture changed materially: the most successful single-vendor identity spine on the open web now belongs to one holding company. That leaves The Trade Desk's UID2, with its multi-operator governance and publisher adoption, as the most credible neutral identity option standing, and it makes first-party publisher and retailer graphs more strategically important, not less.
Clean rooms close links, one bilateral deal at a time
Clean rooms genuinely close the measurement and attribution links. Snowflake's broad deployment makes it the default neutral substrate, AWS runs a structurally similar offering, and independents like InfoSum serve buyers who want distance from cloud-platform politics. The honest read: a clean room closes the loop one bilateral relationship at a time. A brand running across six SSPs needs six matching arrangements or a hub. And almost none of the output feeds machine training. Clean-room results are read by humans doing attribution and MMM, which means the loop closes into a slide deck, not into the next bidder model. Real progress, wrong altitude.
Payment networks are the underweighted story
The most consequential new category of 2024 through 2026 is the one industry discussion weights least: payment networks monetizing verified purchase data. This is structurally different from everything else on the map, because payment networks see conversions across merchants, which is exactly the view retail media cannot provide and identity vendors do not own.
PayPal stood up an ads business and launched its Transaction Graph measurement program in January 2026: cross-merchant purchase truth across hundreds of millions of users, with deterministic login identity as the resolution key. Chase launched the first major bank-led media network in 2024, a genuinely closed loop of identity, spending data, activation, and measurement inside its customer base of roughly eighty million. Mastercard assembled the most mature integrated play, combining its transaction insights with personalization and loyalty assets and, most tellingly, partnering its purchase signal into the existing programmatic stack rather than competing with it.
Verified purchases are the cleanest optimization target in advertising. They are not proxies, they are the outcome itself. Any genuinely competitive open-web loop probably integrates payment-network signal rather than competing with it, and the operators moving that signal toward bidders, rather than toward dashboards, are the ones to watch.
Retail media: real loops, isolated from each other
Retail media is the largest real closed-loop story outside the gardens. Amazon is effectively a garden by another name. Walmart's acquisition of Vizio extended its loop into CTV measurement, one of the most structurally consequential moves of the period. Kroger closes a complete shopper-graph loop through its data science arm. The catch is fragmentation: a brand running across four retailers has four separate closed loops, each optimizing inside its own walls. Real closure, bounded scope, and no cross-retailer learning anywhere.
The holdcos are building walled gardens and nobody says it out loud
The quietest closed-loop development is happening at the agency holding companies. Publicis now owns first-party data, activation, retail media infrastructure, and, with LiveRamp, identity and clean rooms. The Omnicom and IPG merger consolidates data platforms, commerce media, and Acxiom's identity assets into the largest holdco in the market. These are vertical integrations that close the loop at the holdco level, for the holdco's book of business.
Call it what it is: holdco-as-walled-garden. For brands inside those supply chains the loop does close, with the caveat that it optimizes to the holdco's economics. For everyone else, DTC brands, in-house teams, independents, it does not help at all, and it removes formerly neutral infrastructure from the neutral column.
The substrate path is the only one that scales
The remaining path closes the loop across operators instead of inside one: containerized bidders running inside exchange infrastructure, standards for exchanging audience and context signal as embeddings, agent frameworks with registries for trust. The bet is that no single vendor has to own the loop if the standards layer lets data flow back into training across vendors. Composable bidders already let a brand run its own decisioning with its own measurement feedback without building a DSP from scratch, and the early co-location deployments are real.
This is the only architecture on the map that could compete with the gardens at open-web scale, and it is also the least mature. It depends on multi-vendor adoption, which takes years. Today it is more potential than reality. By 2027 or 2028 it could be the dominant pattern, if adoption compounds the way standards adoption sometimes suddenly does.
The honest bottom line
Real loop closure exists today in three places: vertical integration (retailers, CTV platforms, and now holdcos), specific bilateral partnerships, and payment networks extending purchase truth into measurement. The open-web-wide closed loop, the thing that would actually compete with walled gardens at scale, does not yet exist. Most of what markets itself as closed-loop is measurement without a path back into bidding, which is half a loop, and the half that was never the hard part.
If you are placing bets, mine are these. Neutrality is becoming scarce and therefore valuable: every acquisition that pulls shared infrastructure into one owner's loop raises the worth of whatever stays credibly neutral. Purchase-level signal will keep pulling toward the bidder, and whoever industrializes that path first sets the terms. And the substrate-and- standards work, unglamorous as it is, is where the compounding happens, because it is the only version of the loop that closes for everyone at once.