What the agentic AI era means for ad agencies, with Omnicom’s Jonathan Nelson

A major talking point among the brand and agency executives in attendance was the onset of the so-called agentic era of artificial intelligence, in which AI tools handle multi-step tasks for people like booking a full travel itinerary — or firing off a client brief. In this era, data will be at even more of a premium than it is today

“If you think about the IPG acquisition, we will have a broader platform to to do things. We will have the broadest dataset on the buy side anywhere in the world, and more expertise, more clients,” Jonathan Nelson, CEO of the agency holding company’s digital arm Omnicom Digital, said on the latest Digiday Podcast, which was recorded in person at CES.

The combined company will also have Omni AI, a product that Omnicom is developing to combine various foundational large language models. “We’re putting that on every employee’s desktop in Omnicom right now,” Nelson said.

Which gets at another aspect of how AI will affect agencies’ business. As agencies effectively outsource tasks to AI tools, the traditional agency compensation model — in which agencies are paid in accordance with the time it takes to complete client projects — will be under pressure. This is again where Omnicom is counting on the combination with IPG and the corresponding dataset — as well as its previous acquisition of commerce platform Flywheel — to be able to adopt a model in which its client fees are contingent on the results of its work rather than the time it takes to complete that work.

“Here we are sitting on this massive dataset. It’s coming together across audience, activation, outcomes. It has that purpose, which is driving towards outcomes remuneration,” said Nelson.

Here are a few highlights from the conversation, which have been edited for length and clarity.

The post-IPG data reserve

We are now sitting on what is the largest and arguably best, the gold standard of audience data in Acxiom. We have Omni, which is our media orchestration platform. Sitting inside of Omni, sitting on tens of petabytes of behavioral inventory pricing data. And then connecting it to outcomes, connecting it to commerce, connecting it to the Flywheel Commerce Cloud, which has tens of petabytes of data itself.

Omni AI

Our model has always been to create very, very open systems like Omni. This is an extension; it’s Omni AI. So we give everybody most of the models that we think are deemed fit for production, and in the back office, we are testing probably 15 to 20 models at any given time. Maybe 10 of them are on the desktop.

WTF is agentic AI?

Say you’re booking plane tickets. [The AI agent] knows that I like to fly in business class, aisle seat, I don’t like to take red eyes, I like to order a certain kind of meal. Here’s my frequent flier number; book the ticket. Instead of doing each of those manually, the agent will just remember what you like, and it will do it for you.

SEO for AI agents

SEO was always essentially marketing to an algorithm at Google to try to get your content well-placed. Essentially, we were marketing to a version of a robot. The idea was you weren’t really talking to a person directly. You were talking to an intermediary that was a computer, who in turn would talk to a consumer. I think you’re going to see the same idea times a trillion.