Publicis Is Making Itself Harder to Leave
The Scorecard Already Looks Different
AI was supposed to make the advertising holding company easier to dismantle. Generative tools can create drafts, variations, scripts, and visual assets quickly. Clients could reasonably ask why they should keep paying large networks for work that software appears capable of accelerating.
Publicis has supplied a more complicated answer. The group reported 5.6% organic growth for 2025, an 18.2% operating margin, and €2.03 billion in free cash flow before changes in working capital. Its investor presentation also reported a 98% retention rate among its top 100 clients. Those figures do not establish a clean line of causation from AI to performance. Client relationships are long-lived, and financial results reflect far more than software deployment. Still, the pattern is hard to ignore.
The company is turning AI into an argument for keeping more of the marketing operation connected. That is a more durable proposition than a demonstration of faster copy generation. And it raises the question this article is really about: when an agency becomes harder to leave, is the client buying more value, or simply more dependency? The two are not the same, and Publicis's own numbers do not tell you which one you are looking at.
Data Did the Difficult Work First
Generic AI tools are widely available. A holding company can license the same foundational models as its competitors, and a brand can do the same. The strategic question is whether the model can work inside the client's actual marketing environment.
That environment is full of constraints. Customer data may be usable for one purpose and prohibited for another. Identity needs to be resolved across channels without treating consent as a technical afterthought. Media decisions require measurement that can be defended when the budget moves. A useful AI system has to function inside those conditions rather than merely produce convincing output.
Publicis's Epsilon business gave it a substantial identity and data foundation before generative AI became the industry's preferred vocabulary. Its Connected Media operation, powered by Epsilon, grew at a high-single-digit rate, according to the company, and now represents 60% of net revenue. The group's AI production platform also grew at a double-digit rate.
The proposed LiveRamp acquisition makes the strategy clearer. Publicis agreed on May 17 to acquire the data-collaboration platform in an all-cash transaction valued at about $2.17 billion on an enterprise-value basis. The deal is not a formality: it remains subject to LiveRamp shareholder approval, antitrust clearance in multiple jurisdictions, and CFIUS review, with an outside date in 2027. But its logic is already visible. Publicis is pursuing a larger role in the infrastructure that allows companies to connect, activate, and measure data across a fragmented marketing ecosystem.
That is where the agency contest has moved. An AI model can be bought. Permissioned data relationships, usable identity, and a reliable connection to media and commerce operations take longer to assemble. This is also the first turn of the screw on switching costs: the more of a client's data plumbing runs through the agency, the more leaving means rebuilding.
Can a Workflow Keep an Account?
Retention is usually treated as a quiet operational metric. It deserves more attention in the AI race, because it is where the value-versus-dependency question stops being abstract.
An agency relationship becomes harder to replace when its systems touch the work that determines how a brand sees audiences, activates campaigns, measures performance, and adapts creative. Replacing the agency then involves more than choosing a new creative partner. It means rebuilding processes and deciding what can travel safely from one operating environment to another.
That can create genuine value for a client. A connected workflow can reduce handoffs, make decisions faster, and provide a clearer view of what is working. It can also create dependency. The two outcomes can coexist inside the same 98% retention number — which is exactly why that number should be read carefully. It is not proof that AI is responsible for client loyalty. It does, however, support the idea that the company's services are becoming more deeply integrated into client operations. Investors see a business with recurring relationships and wider scopes of work. Clients should see a relationship that requires clear contractual boundaries before the integration becomes irreversible.
The Agency Model Has Started to Move
For decades, the holding-company model depended heavily on labor. Agencies sold time, specialized judgment, media scale, and creative reputation. AI changes the cost and speed of some of that work, which places pressure on the old logic of the billable team.
Publicis is presenting a different model: an agency as a marketing operating system that coordinates decisions across data, media, content, commerce, and customer engagement. Human judgment remains central, especially where brand risk, creative quality, and accountability are involved. Yet more of the underlying work can be routed through a common technical layer.
The financial appeal is straightforward, and the mechanics are worth stating precisely. Publicis's operating margin rose to 18.2% in 2025, up 20 basis points from 18.0%. According to the company, AI and its platform organization unlocked close to 50 basis points of operating leverage during the year; roughly 30 of those points were reinvested in new business, AI, and talent, and about 20 dropped through to margin. AI, in this version of the story, is expected to expand capacity while creating room for further investment — not merely to cut costs.
That is a better outcome for the company than simply cutting labor and hoping the savings hold. A holding company that uses AI only to make agency work cheaper may find itself competing on price against a growing number of smaller, specialized firms. A company that uses AI to become more useful in the client's operating model has a chance to protect both margin and relevance. Whether that usefulness is worth the lock-in it produces is a judgment the client has to make, not the agency.
What Rivals Still Need to Show
Every large agency group now has an AI platform story. WPP describes WPP Open as an agentic marketing platform, and other rivals have made similar commitments around data, workflow, and automation. The language is no longer scarce.
The operational evidence is harder to come by across the board — and that caution should cut both ways. Publicis's 2025 was strong, but a single strong year does not prove that AI, rather than a well-timed accumulation of assets and new-business wins, is what produced the retention and the margin. By the same logic, one weak year does not condemn a competitor. WPP's 2025 revenue less pass-through costs fell 5.4% on a like-for-like basis, and its headline operating margin declined to 13.0%. That is a difficult result, but it is shaped by account losses, restructuring, and uneven client spending as much as by any verdict on its platform. What it does establish is that broad claims about AI are now easy to make and hard to substantiate, at any group.
The bar, then, is the same for everyone, Publicis included. Rivals need to show that their systems are being used inside major accounts, with approved data and clear rights. They need to demonstrate that AI improves the work clients actually buy, rather than producing a polished layer of presentation around existing services. They also need to show whether AI-led work expands client scope, improves renewal behavior, or creates a measurable economic advantage. The distinction between a platform and a presentation becomes visible when a client has to decide where the next year's budget will go.
What CMOs Should Question Now
CMOs should welcome better workflow integration, but they should resist treating convenience as a substitute for control. The first question is who owns the data connections, audience definitions, agent configurations, and performance history created inside the relationship. If the answer is vague, the apparent efficiency may already be turning into an expensive dependency.
The next question concerns portability. A client does not need every operational detail to be portable on day one. It does need a practical transition path if an agency changes, a technology vendor fails, or a commercial relationship deteriorates. That path should be defined before the system becomes indispensable, not after.
Finally, CMOs should examine how AI-driven savings are shared. If automation reduces production time or improves media decisions, the commercial model should reflect the changed economics. An agency that becomes more valuable through better infrastructure may deserve a broader role. It should still be able to explain where the value is created and how the client can verify it — the test that separates a partner from a landlord.
Publicis may be building a lasting lead, or it may be benefiting from a well-timed combination of assets and execution. The next procurement cycle will reveal more than the next AI demonstration. The decisive documents may be the data-rights provisions, workflow ownership terms, and exit clauses attached to the client relationship — because that is where "harder to leave" is finally priced, as either value or dependency.