The AI Agency Race Is a Data Race

Publicis Groupe says it wants LiveRamp because the next growth market is “agentic transformation.” That sounds like the kind of phrase built for an earnings call, polished just enough to survive a keynote and vague enough to mean almost anything.

But this time the phrase points to something real. Publicis is not simply buying an AI story. It is buying the layer beneath the AI story. LiveRamp is not famous because it builds autonomous marketing agents that wake up in the morning, optimize a media plan, negotiate with publishers, and brief a creative team before lunch. LiveRamp is known for data collaboration, identity, clean-room infrastructure, partner connectivity, activation, and measurement. Those are less glamorous than agents. They are also the reason agents can become useful.

That is the part worth watching. The agency world is full of AI theater right now. Every holding company has a platform. Every platform has agents. Every agent promises to remove friction, accelerate decisions, and turn the marketing organization into something more predictive, automated, and intelligent. The language is often swollen, but the strategic fight underneath it is becoming clearer. The future of agency competition will not be won by the group with the flashiest AI interface. It will be won by the group that controls the data environment in which AI can act.

The AI agency race is becoming a data infrastructure race.

Agents without governed data are only demos

An AI agent in marketing is only impressive if it can do something with reality.

It can generate campaign ideas from a generic model. It can summarize a brief. It can draft copy, suggest audiences, produce reporting language, or answer a question about performance. That may be useful, but it does not change the agency business by itself. It is productivity software wearing a more ambitious costume.

The larger commercial opportunity begins when agents can work inside the actual marketing system. That means access to customer records, consent status, identity graphs, retail media signals, publisher data, CRM segments, media performance, transaction data, loyalty behavior, creative history, attribution models, suppression rules, market constraints, and channel-specific activation pathways.

That environment is not simple. It is fragmented by design and by accident. Client data sits in different systems. Retailers hold transaction data. Publishers hold audience data. Platforms hold performance signals. Agencies hold planning and media intelligence. Legal teams hold the rules. Privacy teams hold the red lines. Finance teams want proof. Marketing teams want speed. No serious enterprise client wants all of that dumped into an ungoverned AI system because someone in a workshop said agents were the future.

This is why LiveRamp matters to Publicis. The deal is not just about making AI smarter in the abstract. It is about giving Publicis more control over how client data, partner data, identity, permissions, and activation can be connected. Agents need that connective tissue. Without it, they produce plausible suggestions. With it, they can begin to support real workflows.

The word “agentic” makes the story sound like it belongs at the interface layer. In practice, the decisive layer sits below the interface. It is the data layer, the identity layer, the permissioning layer, the clean-room layer, and the activation layer.

Why LiveRamp fits Publicis

Publicis already made one of the most important data bets in the agency world when it bought Epsilon in 2019. That acquisition gave Publicis a major first-party data, identity, CRM, and personalization asset at a time when the advertising industry was moving away from easy cookie-based targeting and toward owned customer intelligence.

LiveRamp extends that logic. Its value is not that it gives Publicis one more database. Its value is that it helps companies connect data across organizations without turning collaboration into uncontrolled exposure. That distinction matters. Modern marketing increasingly depends on relationships between brands, retailers, publishers, platforms, media networks, data providers, and commerce partners. The value is not only inside one company’s data. The value often appears when one company’s data can be matched, enriched, analyzed, activated, or measured against another company’s data under strict controls.

That is the data collaboration market. It is also where the agency holding companies now see the next battle.

For Publicis, LiveRamp strengthens the machinery around identity resolution, privacy-preserving collaboration, audience creation, activation, and measurement. Those capabilities sit directly underneath the kind of agentic workflows Publicis wants to sell. If a client wants an AI system that can recommend which audiences to prioritize, where to shift spend, which customers to suppress, which retail media partners matter, or which creative variants deserve more investment, that system needs trusted inputs. It also needs permission to use them.

A generic AI assistant can talk about marketing. A connected AI workflow can operate inside marketing. Publicis is betting that clients will pay more for the second version.

The future agency wants to become an operating layer

The agency holding company model has always sat between clients and markets. Agencies translated brand ambition into campaigns, media plans, creative work, production systems, measurement frameworks, and platform execution. The old model was labor-heavy, relationship-heavy, and service-heavy. AI puts pressure on all three.

If AI can reduce the cost of producing content, reporting, analysis, planning, and optimization, then the traditional billable-service model becomes harder to defend. Agencies can still sell judgment, strategy, creativity, relationships, and execution, but the economic center shifts. Owning the workflow becomes more valuable than merely staffing it.

That is why the language around agency AI platforms has changed so quickly. Publicis talks about smarter agents and data co-creation. WPP acquired InfoSum and framed the deal around AI-driven data, privacy-enhancing connections, first-party data, media partners, audience targeting, marketing intelligence, and AI model training. Omnicom presents Omni as an AI-driven marketing intelligence platform built around identity, data infrastructure, media and commerce buying power, content production, and autonomous agent systems. dentsu describes dentsu.Connect 4.0 as an agentic AI-powered operating system that unifies the marketing lifecycle, synthesizes data in real time, automates routine activity, and keeps human governance in the loop.

The wording differs. The direction does not. Each group is trying to become the environment through which marketing work is planned, connected, automated, governed, measured, and improved. The agency is no longer presented only as a collection of specialist teams. It is becoming a software-like control layer across the client’s marketing operation.

That is a major change in power. A service provider can be replaced. A deeply embedded operating layer is harder to remove. A campaign can be moved from one agency to another. A data and workflow environment that connects customer intelligence, partner ecosystems, activation channels, creative production, measurement, and AI decision support becomes much stickier.

This is why the Publicis-LiveRamp deal deserves more attention than the usual acquisition cycle. It shows where the new lock-in may come from.

Clients get speed, but the real prize is control

The client benefit is easy to understand at the surface level. Publicis can tell clients that LiveRamp will help them use their data more effectively, collaborate more safely with partners, build better audiences, improve measurement, and make AI systems more useful.

That is probably true.

For a bank, the appeal could be more relevant customer engagement without throwing sensitive financial data into unsafe systems. For a retailer, it could mean better use of commerce signals across brand partnerships and media networks. For a healthcare or pharmaceutical company, it could mean more careful data collaboration in a highly regulated environment. For a consumer brand, it could mean faster segmentation, cleaner activation, and better measurement across a fragmented media market.

Marketing teams are drowning in systems that do not naturally cooperate. They are expected to personalize more, waste less, prove more, move faster, and comply with stricter rules. An agent that can help coordinate that work across governed data sources is not a toy. It is an operational asset. But the deeper client benefit is control.

The strongest version of this model gives clients a way to make AI useful without losing control over data rights, access, consent, and activation. It creates a path between two bad extremes. One extreme is the old manual marketing machine, where every insight requires a meeting, every data connection requires a project, and every partner collaboration moves slowly. The other extreme is reckless automation, where AI systems are granted broad access to sensitive information and everyone hopes the guardrails work.

The better path is governed acceleration. That means faster workflows inside defined boundaries. The agent can help, but the environment decides what the agent is allowed to know and do. That is the commercial promise behind the infrastructure race.

The risk is dependency

There is another side to this shift. If agency groups become the operating layer for AI-enabled marketing, clients need to ask harder questions about dependency. Who controls the identity spine? Who defines the permissions? Who owns the workflow logic? Who decides which data is usable? Who governs the agent’s access? Who validates the measurement layer? Who can audit the system when performance claims become disputed? Who can move the client out if the relationship changes?

This is not a reason to reject the model. It is a reason to treat it as infrastructure.

The more agencies become embedded in data collaboration, AI workflows, and activation systems, the more procurement and governance teams need to evaluate them like critical enterprise partners, not only as marketing vendors. That changes diligence. A pitch deck about agentic transformation is not enough. Clients need to understand architecture, interoperability, data rights, model governance, clean-room controls, consent enforcement, auditability, switching costs, and failure modes.

If an AI agent recommends a media shift, the client should know which data shaped the recommendation. If the agent creates an audience, the client should know which permissions allowed it. If the agent suppresses customers from a campaign, the client should know whether the rule came from strategy, privacy, brand safety, regulatory concern, or model inference. If the system claims better performance, the client should know whether the measurement layer is independent enough to trust.

Marketing has spent years learning that attribution can become a political instrument. AI will make that problem sharper if the same environment that recommends action also measures success without adequate transparency. This is where the next governance fight will sit.

Competitors are moving in the same direction

Publicis is not alone.

WPP’s InfoSum acquisition is the most direct comparison because it also centers on privacy-safe data collaboration. WPP did not buy a conventional creative agency or media shop. It bought infrastructure that helps connect data without exposing it, and it tied that acquisition directly to AI-driven marketing intelligence, audience targeting, and model training. That is the same strategic grammar as the Publicis-LiveRamp deal.

Omnicom is coming from a different angle, especially after its combination with IPG. Its Omni platform is framed around marketing intelligence, identity infrastructure, media and commerce buying power, content production, and autonomous agent systems across creativity, media, commerce, and measurement. The logic is scale plus data plus workflow. If Omnicom can integrate Acxiom and Omni into a broader operating system for clients, it has a powerful answer to Publicis’ Epsilon and LiveRamp combination.

dentsu is positioning dentsu.Connect 4.0 as an agentic AI-powered operating system for the marketing lifecycle. Its public language emphasizes interoperability, real-time data synthesis, automation, decision augmentation, and human governance. That matters because many clients do not want a closed system that forces them to rebuild their technology stack around one agency. They want something that connects to the systems they already have. dentsu’s answer is to make interoperability part of the pitch.

Different groups are using different words, assets, and acquisition paths. But they are converging around the same thesis. AI in marketing will not be won by loose tools sitting at the edge of the organization. It will be won by the systems that connect data, decisions, workflows, execution, and proof. That is why the agency race has moved below the surface.

The interface is not the moat

Most people will see the agents first.

They will see a dashboard, a chat interface, a planning assistant, a creative workflow, a reporting tool, or a system that claims to coordinate campaign activity across teams. That is the visible product. It is also the easiest part to copy.

The harder part is everything underneath. Which data is connected? Which partners are reachable? Which identity signals are reliable? Which permissions are enforceable? Which workflows are integrated? Which media channels can be activated? Which measurement systems can prove outcomes? Which governance controls prevent an agent from turning speed into liability? That is where the moat forms.

Publicis’ LiveRamp deal is important because it recognizes that AI agents do not become powerful because they sound intelligent. They become powerful when they are connected to the right data under the right controls and can turn that intelligence into action.

The agency of the future may still need brilliant strategists, planners, creatives, analysts, and operators. It will still need human judgment because brands do not run on automation alone. But the holding company that wins the next phase will not be the one that merely adds AI to existing services. It will be the one that turns data access, partner collaboration, agentic workflow, and measurement into a governed operating environment.

That is the real Publicis signal. Everyone is selling agents. Publicis is buying the layer that tells agents what they are allowed to know, where they are allowed to act, and how their actions become measurable business outcomes.

That is where the agency race is going.