Beyond the hype: How AI is reshaping the media agency model

From the Palais stages to beachfront panels, artificial intelligence was everywhere at this year’s Cannes Lions International Festival of Creativity. Brands, agencies and platforms all claimed to be AI-powered. Everyone wanted to show off their latest tech — if not a working product, at least an innovative demo or sizzle reel teasing an AI-powered future.

A deep shift is certainly underway, especially for media agencies.

But just because everyone is building the next big AI thing doesn’t mean these tools are effective. New tech companies popped up offering amazing theoretical AI tools that couldn’t provide real-time results. And those demos? They simply involved clicking on a new tab without the components actually communicating with each other.

“In the agency world, there is a historic behavior of, ‘Let’s build something that looks like it’s real and click through it, and then we’ll think about building it afterwards, right?’” points out Graham Wilkinson, IPG’s Kinesso chief innovation officer and global head of AI. “When I’m assessing any kind of technology with regards to AI, there are a few things that I always think about.”

Navigating between hype and reality

Nine out of 10 ad agencies claim to be using AI, with about two-thirds of them utilizing generative AI, according to a report from Forrester. Clients and agencies are experimenting with AI agents using large language models (LLMs) like ChatGPT, Llama and Gemini across all their workflows, while media owners are exploring how AI can enhance the content supply chain to be faster and more cost-effective.

“Every corner of the industry is starting to move beyond experimentation to a much more disciplined understanding of how it should change or transform the way we do our work today, the way we deliver our service and the way we can create more interconnectivity across the ecosystem,” says Ralph Pardo, Omnicom Media Group North America CEO.

Still, “starting” is a key word.

“In many instances, businesses come forward and say, ‘We’re an AI business.’ What does that really mean?” Wilkinson questions. “Does it mean that everything is powered by AI? Does it just mean that your employees have access to some AI? Or does it mean that you put an agentic or a generative layer on top of your software — and your software is still doing the same stuff?”

Wilkinson believes the true test is whether the demos actually work. Real AI solutions need connectivity between systems, where the output of one process feeds directly into the next without manual copy-pasting or file uploads. In assessing the tech, the ability to input a theoretical client and instantly see results becomes more important than ever.

“Show me how the output of this thing becomes the input for this next thing — and not by copying and pasting it,” Wilkinson says. “I want to hit a button, and I want to see the output appear.”

Sometimes, you have to take AI tools for a test-drive to truly understand their value. Pardo emphasizes that success should be defined by measurable outcomes like faster campaign setup or improved media performance, not novelty. If a tool doesn’t show real business value at scale, it shouldn’t move forward, no matter how flashy the demo is.

“It’s about putting AI into real workflows and seeing if it actually helps, not just doing one-off pilots,” he explains.

When AI actually does work

That doesn’t mean that everything is just hype. For Omnicom Media Group North America, AI is already embedded in streamlining daily operations. AI agents, for example, automate campaign setup and measurement planning, reducing hours of manual work.

“We are focused on identifying the points of our business — not just in media but across the entirety of the marketing services ecosystem — where we can see tangible value today and where we can create some immediate benefits,” Omnicom’s Pardo says. “Those benefits are speed to insight, speed to outcome and the ability to orchestrate and connect the dots.”

At Cannes, Omnicom announced several partnerships backed by its AI-powered Influencer Discovery Agent, a tool integrated with Walmart purchase data that enables brands to evaluate whether an influencer’s followers are making purchases. A separate collaboration with Meta and its Live Partnerships Ads, which amplifies creator content, helps brands understand consumer behavior during livestreams.

Meanwhile, Kinesso is using AI to expedite the ideation process. For a recent campaign with a major sports league, Kinesso’s account group built five digital audience AI agents trained on survey and first-party data. These agents offered immediate feedback on creative concepts, which gave humans more time to improve those concepts. Content generators delivered tailored messaging for each segment, allowing employees to have more material to refine. It led to a 23% lift in performance compared to non-AI-generated work.

“That’s a process that would have typically taken three to four weeks for a team to conduct,” Wilkinson says. “This took a week, and what [the team was] able to take to the client was iterated on more, and therefore it was better. It had gone through more cycles already before it got to the client to review.”

But it’s not just about optimizing creative concepts. Part of the issue with brand safety is human error when inputting potential no-go topics, he adds. AI technology can provide an additional safeguard with rapid speed. At Cannes, Kinesso’s sister company, Acxiom, and Snowflake announced a new joint solution. This will add a layer of Snowflake’s built-in privacy and security features to IPG’s technologies, ensuring more compliance.

“Agents don’t stop working,” Wilkinson says. “They can talk to each other all the time. They can share all the same information. They can have access to it all.”

The key to both strategies is handing more operational tasks to AI and having humans oversee and fine-tune the work. AI is necessary for the speed of today’s business, but it’s not doomsday for agencies that use these tools to their benefit. Knowing how to prompt, train and interpret AI has become a new form of literacy.

“That advantage needs to be some degree of human idiosyncrasy or noise that kind of injects itself to try to really create something unique or give a leg up,” Pardo says. “I think we’re still going to be in that place for quite some time. I do not think the machines are taking over completely.”