Cannes Briefing: Cannes is selling AI transformation. Off stage the talk is about the bill

For him, the point is that AI augments his engineers in a supervised and managed manner. Or as he put it: “It means we can still get three times the productivity — real world of course, not the 10 times claims that float around in the ether — which only really exist for super specific use cases without exposing everything to the AI, and thus avoiding those massive costs.”

“It’s already becoming a conversation in the conversations we’re having with media agencies,” said a senior marketer who was not authorized to speak to Digiday on the record. 

The shift, when you get people to talk about it honestly, isn’t really about whether AI works. Almost no one disputes that anymore. It’s about what it costs to make it work — and that conversation has changed register over the past year, from theoretical to felt. A year ago the talk was almost entirely upside: faster production, leaner teams and productivity gains. Now the same people are doing arithmetic they didn’t expect to be doing yet, on bills that didn’t behave the way the projections said they would. The hype hasn’t disappeared — Cannes still runs on it — but underneath it sits a much plainer, much less glamorous conversation about line items. 

“The introduction or the arrival of new technologies accelerates what is possible, but also accelerates the expectations required to want more for less, ” said Peter Mears, Global CEO of Havas Media Network. 

There’s truth in that but it’s not the whole picture. For all the industry’s habit of reaching for hyperbole on AI some of this silence is more mundane than intentional — it’s that most people genuinely aren’t there yet. They’re not withholding the cost conversation so much as they haven’t worked out how to have it. The business model of the future is still being built in real time, from understanding what CMOs actually want to pay for to to ratifying how agencies are remunerating for delivering it.  

That dynamic showed up almost word for word in a conversation overheard yesterday between two ad execs who were working through the same logic out loud. The problem, as one of them put it, is that early in the AI build-out, the cost-benefit math is still incomplete, and given how transformational the technology is supposed to be, not experimenting feels riskier than overspending on it. So businesses build first and ask what it costs later, designing around a vision rather than current unit economics. The bill only starts to matter once a use case actually proves out — at which point token cost stops being an abstraction and becomes a line item like any other: power, people, premises.

“Clients are expecting a lot of transparency,” said Jess Dervyn, an analyst in Gartner’s marketing practice, “They want to understand where AI is being used, so agencies have to provide that. They have to be honest and say, ‘this is where we’re using AI.’ But they’re not as transparent about how much it costs. It’s changing everything.”

Every use case that survives this experimental phase eventually gets weighed against what it costs to keep running — and once it does, AI stops being a transformation story and starts being a procurement one. The conversations happening quietly now, at dinner and between panels, are a preview of the ones that will happen on stage next year, because the variable that’s currently invisible — what AI actually costs to sustain at scale — is also the variable every other cost in this industry eventually gets measured against. 

Whether it’s efficiency, agentic or orchestration, it all has to clear that bar eventually. Cannes just hasn’t gotten there yet. It will.

AI isn’t controversial. It’s just early

The advertising industry doesn’t have a principled position on generative AI and creative. It has a lagging indicator. What’s controversial today is just what hasn’t hit the scale yet.

That’s one of the prevailing conversations actually happening on the Croisette this week, across several unrelated exchanges that keep circling back to the same place.

“I don’t think going into these kinds of discussions [thinking] anything is sacrosanct is a mistake, right?,” said Jim Mollica, CMO of Bose. “But that being said, I think that there are certain things that AI is particularly good at, and it’s a lot of routine, fundamental stuff.”

When it comes to nonlinear thinking, though, it’s poor, he said. AI, in other words, isn’t very good at being creative. That’s not the surprising part. The surprising part is the suggestion that it doesn’t need to be. One agency founder had watched this play out directly in rolling out AI production tools for a global client. They watched the polish drop as a result, and yet what stuck with them wasn’t the quality. It was the question of who’s actually close enough to notice. At the speed a feed moves, the label on a bottle is the kind of thing a marketer clocks. A consumer, maybe, just scrolls past.

“What I didn’t expect was how quickly the narrative became more balanced,” said Abby Laursen, vp of product marketing at Snap. “At first it was all about the possibilities — endless content production, endless creation. But as that content became so proliferative, so ubiquitous across the industry, the conversation shifted fast into: is this really always what’s right for a brand?”

Because that balance was never really about trust. It was always about fit. Once “is this what we want to be” stopped having one answer, it stopped being a question the industry could settle once and moved on to one every brand has to keep asking on its own. The shift isn’t principle arriving. It’s the same lag, just running on a thousand separate clocks instead of one.

“Brands need to consider what’s the best fit for them, what they’re most comfortable with. That differs between companies, between realities,” said Laursenhat.

A thousand separate clocks is also, in effect, a thousand separate definitions of “controversial”, which is where Debra Aho Williamson’s read on disclosure gets interesting. The founder and chief analyst at Sonata Insights has been tracking the debate over whether brands should flag when an ad was made with AI. There’s no consensus yet, she said, but the open question underneath it is more basic than disclosure itself.

“Does disclosure matter?” Williamson said. “Or is it something that becomes second nature to how you create advertising — like CGI, or graphic design, or any Photoshop?” Nobody discloses retouching anymore. The debate over whether to disclose AI use, by her account, isn’t heading toward a resolution so much as an expiration date.

Which is really just Occam’s razor on a sliding scale. 

Voices from the Croisette

“More so than ever, the conversations have been more about how work gets done than what the work is. And I think that’s a true testament to what orchestration is and why orchestration is becoming the topic du jour here at Cannes. I also think orchestrating means bringing together lots of different players, and capabilities — which is really around creating a composable way of working. For the most part, this is around thinking through not just what, but the ways in which you get work done. Also, what’s your op model? What’s your process? Do you really need briefing anymore? Or is that just built into the way in which you work? So, a lot of those concepts come into question. And the concept of orchestrating is not just doing work with agents, but actually, and fundamentally working differently than the way you have before with your agencies and partners.” — Mark Singer, CMO at Deloitte Digital

Digiday 2026 Video Studio