The next challenge for AI marketing is collaboration without lock-in

A recent piece in this publication makes a compelling case for data collaboration as the foundational layer of AI-powered marketing. The argument is well made: as agentic AI moves from concept to deployment, connected and permissioned data becomes the difference between an AI that operates on a complete picture and one that optimizes against a fragment of it. 

That framing is right. But there is a question it doesn’t fully answer, and it is the one that will matter most as these environments become genuinely strategic: when the collaboration ends, who keeps the intelligence it produced? 

Most enterprises are not yet asking this. But they should be. 

AI readiness is a data collaboration issue 

In Cisco’s 2025 study of 8,000 leaders, just 13% of organizations were found to be fully ready to capture AI’s value, with data foundations identified as the defining gap between leaders and the rest. 64% said they struggle to centralize their data, the step that has to come before any of it can be joined securely with a partner’s. 

The most valuable signal is precisely the kind no single party holds. A retailer and a brand, a bank and a payments partner, a hospital and a research group: each sees only a fragment of the picture.Joined securely, those fragments train better models and sharper measurement than any one dataset can. That is why the readiness gap is, at its core, a collaboration gap. 

So far, so agreed. The question is what happens next. 

Collaboration is not the same as custody 

Here is a distinction that too few enterprises draw when evaluating collaboration infrastructure. Running a joint computation is not the same as retaining permanent custody of the intelligence it creates. 

A collaborative environment lets two parties answer a question together: which audiences overlap, which treatment lifted outcomes, which cohort responded. But the models, the matched audiences, and the learned patterns that fall out of that work are durable assets. They outlive the project. Whoever holds them holds an advantage. 

This is what intelligence custody means: not who runs the computation, but who ends up holding the durable result. An enterprise can run a perfectly privacy-preserving collaboration and still cede that long-term intelligence to the platform sitting in the middle. 

Two legal teams will spend weeks on who may touch the raw data, and almost no time on who keeps the model it trains. As agentic AI raises the value of those trained models dramatically (agents operating on unified, high-quality data will outperform agents operating on fragments by an increasing margin), that asymmetry becomes a strategic liability. 

The agentic context raises the stakes 

The author of the article I referenced earlier is right that the agentic era creates a critical dependency on connected data. An agent is only as effective as the data it can see. The IAB Tech Lab’s data clean room guidance points in the right direction: as these environments mature, common interoperability standards become essential so that data and intelligence can move between them. 

But there is a specific risk in the agentic context that collaboration infrastructure with a stake in the data doesn’t address. As agents automate media buying, audience activation, and measurement at scale, the outputs they generate (refined audience segments, campaign-level attribution models, cross-partner behavioral patterns) accumulate into genuinely valuable proprietary intelligence. If the platform running that infrastructure has a commercial interest in the data flowing through it, those assets don’t stay with the enterprise that created them. 

The cookie deprecation cycle offers a structural lesson here. For half a decade, the advertising and measurement industry rebuilt itself around the coming end of the third-party cookie. Then, in April 2025, Google decided to keep third-party cookies in Chrome after all. An entire market had reorganized its data strategy around a single company’s roadmap, and that roadmap changed. The dependency was the risk. Apply that to AI infrastructure and the exposure is an order of magnitude larger. 

Neutrality has to be structural 

This is why neutrality is moving from a talking point to a procurement criterion. When a collaboration layer becomes strategic, the right question is not whether the operator seems trustworthy. It is whether the operator has a commercial interest in the data flowing through it at all. 

Europe has already answered part of this in law. The EU Data Governance Act, in force since 2023, requires data-intermediation providers to remain neutral toward the data they handle, bars them from using that data for their own purposes, and obliges them to run the service through a separate legal entity. Neutrality, in other words, can be structural rather than a matter of trust. 

A neutral and scalable environment is one where the operator cannot quietly become a competitor, cannot repurpose data to train its own products, and has no claim on the intelligence generated. The credible version of that is enforced by design and governance, not promised in a deck. 

Interoperability is the test of whether you’ve built or rented 

Neutrality answers who owns the room. Interoperability answers whether you can leave it. The EU Data Act, which has applied since September 2025, requires data-processing services to make switching free, fast, and fluid: data exported in machine-readable formats, open interfaces between providers, and switching charges removed entirely from January 2027. Lock-in is being treated as a defect to engineer out. 

An enterprise that cannot move its audiences, models, and measurement between environments has not built an AI capability. It has rented one instead. 

Other industries show what neutral infrastructure looks like in practice 

Facing exactly this problem across thousands of suppliers, the automotive sector built Catena-X, a shared data ecosystem governed by an industry association rather than by any one manufacturer or cloud provider. Its design values are explicit: trust, self-sovereignty, interoperability, and industry governance. Carmakers and suppliers who compete in the showroom still share supply-chain, emissions, and quality data through it, because none of them owns the road they all drive on. 

That model works precisely because neutrality is what makes participation safe. When the infrastructure belongs to the industry rather than to one member of it, collaboration stops meaning you hand a rival or a platform the keys. 

Marketing and media haven’t had an industry consortium to build that for them. What they do have is a growing set of vendor-neutral collaboration platforms whose commercial model depends on not having a stake in the data, and whose architecture enforces that at the technical level rather than relying on contractual promises. 

Build for collaboration you don’t have to own 

Data collaboration as foundational AI infrastructure is the right frame. But before committing strategic AI plans to any shared environment, three questions are worth pressing. 

Does the operator have a commercial interest in the data flowing through it, or is it genuinely neutral? Can data, models, and insights be extracted in standard formats and moved elsewhere? And hardest of all: when the collaboration ends, who keeps the intelligence it produced? 

The enterprises that insist on clear answers to those questions now will be the ones still in control of their own intelligence a decade from now.