The real conversation at Cannes was about the bill

The real conversation at Cannes was about the bill

Fifty-seven per cent of large UK organisations plan to spend more on AI this year. Twenty-three per cent can point to clear, measurable returns from what they have already spent. That figure comes from a Snowflake and YouGov survey of 500 UK executives, published earlier this year, and it tells you something useful about the shape of the AI moment: the money is moving fast and the proof is moving a great deal slower. Cannes Lions last month was a very expensive version of that same gap, played out in public.

After the panels

Ask anyone on the Croisette to summarise what they'd heard about transforming the industry and you got pretty much the same words back, delivered with the same confidence, regardless of who was talking. AI. Efficiency. Transformation. Scale. But Digiday's Krystal Scanlon and Seb Joseph found a different mood at the dinners, in the conversations that happen once the panel has finished. The same people were doing sums they had not expected to be doing yet, on bills that had not behaved the way the projections promised.

Ian Maxwell, who runs the ad tech firm Converge Digital, told them that throwing an entire codebase at a model can send costs through the roof, and that with token prices where they are it can work out dearer than simply paying engineers. The productivity gain he can actually measure inside his own operation is around three times. Larger multiples circulate freely on the conference circuit, and they get quoted a great deal more often than they get evidenced, which is a different question from whether they are achievable.

A senior marketer, speaking without attribution, was more direct: cost has already become part of the conversation with media agencies. Gartner's Jess Dervyn noted that clients now expect agencies to be open about where AI is being used, while considerably less gets said about what it costs to run. Havas Media Network's global CEO, Peter Mears, put it, probably, most plainly of all: new technologies accelerate what is possible, but they also accelerate the expectation of getting more for less.

The shift is not really about whether AI works. Almost no one at Cannes disputed that. It is about what it costs to make it work, and about who can show their working when finance asks. The bill only starts to matter once a use case proves out. At that point token cost stops being an abstraction and becomes a line item like any other.

The verification lag

The UK data makes the same point without the rosé. The Snowflake and YouGov survey found that the biggest barriers UK executives named were not the models or the vendors but their own data quality, internal culture and leadership clarity: the unglamorous plumbing that turns spend into something you can show a board. A global CMO survey from Comviva, published in June 2026, landed in much the same place. Ninety per cent of organisations had raised their AI marketing investment over two years, twelve per cent could prove it had worked, and sixteen per cent of marketing leaders felt confident defending that spend with hard business evidence.

MIT's NANDA initiative^[a] found that 95% of enterprise generative AI pilots return no measurable business impact at all. Different sample, different sector, same shape.

One industry over

If the pattern looks familiar to us, it is because we have already counted it next door. Buzz Radar transcribed six years of pharmaceutical conference dialogue end to end, 475 sessions running to 1.73 million words, and read the record for what the industry actually spends its time on. It talks about AI roughly a hundred times more often than it asks whether AI is working. The enthusiasm turns up everywhere. The checking barely turns up at all.

We saw the same thing at Cannes, slightly better dressed: an industry sprinting toward agentic systems while the question of whether they earn their keep trails a long way behind. For a regulated marketer that gap is not academic. Spend has to clear medical and legal review, sit inside a compliance regime and still show a return, which means proof is a condition of doing the work rather than a nice thing to have afterwards.

It is why we built BRIANN, our audience intelligence for pharma, around the measurement layer rather than the production one: tracking what health professionals and patients are actually saying about a piece of spend rather than what gets claimed for it under hot lights, with every insight sourced back to the conversation it came from and checked by a human pharma specialist before it reaches anyone.

Cannes will get to the bill eventually. It always does. The conversations happening over dinner this year are the ones that turn up on the main stage next June, by which point somebody will have built a panel around them. We would rather count the gap now than applaud it later.

Admin Published on July 16, 2026 1:46 pm