In this November 2024 interview for IRG’s research, Jon Halvorson describes an AI program that started with fourteen identified use cases across the commercial function and then narrowed hard to two: AI optimization of media, and AI creation and testing of content.
The reason for narrowing is the size of the impact, given the scale of investment behind media and content and its effect on the profit and loss account.
Key findings
- He argues marketing has a right to lead on AI, not because it will be most affected but because its use cases are clearer, its business case is sound and the risk feels more reasonable than in other disciplines.
- The program is large. He reports 48 different campaigns run to date, which he believes is close to the highest volume of AI-driven brand building anywhere.
- Failures are part of the record. Ideas were abandoned because of bias in the model, early attempts to generate humans produced six fingers and giraffe necks, and four or five campaigns were pulled at the eleventh hour.
- An AI governance board acts as the check, rejecting projects where AI is not needed and testing how bias is eliminated and how work lives up to responsible AI principles.
- Some proposals were rejected on the same grounds: one team wanted AI to generate recipes without testing them, and others wanted to use AI where no AI solution was needed but it was in vogue.
The three partnerships that make it work
The first is the technology office, for infrastructure, security and technical expertise, since data is what fuels the development of large language models.
The second is legal. A lawyer dedicated to advertising and marketing has been on the journey from the first discussion, because the work requires new contractual frameworks, confidentiality arrangements and decisions about what data the company owns and can train on.
The third is insights and analytics, which in his organization sits inside marketing. It matters for the role of data in training models and for building the measurement protocols that validate the business case.
He is candid about how comfortable each partner is. Technology is most comfortable because marketing gives them a vehicle for their modernization agenda, while legal is stepping into a world of undefined law that is yet to be settled.
The heaviest burden falls on marketing itself. In core marketing, he says, AI is not an augmentation tool but a disruptive one, and realizing its potential requires breaking everything, which is hard for the function that owns it.
He reframes the question of support as a personal one. It is less about whether the function gets support and more about whether the leader is ready to redesign almost every way of working.
How the C-suite was won over
The CFO conversation is unavoidable because the sums are large: well north of $20 million a year, and over $100 million across three years, which requires someone who has vision and believes the business case.
The CEO’s role in his organization is space and air cover rather than a top-down drumbeat, plus acting as a check that enthusiasm stays grounded and disciplined rather than following hype cycles and fads.
The thing that moved the room was a prototype. Once several C-suite members saw it, he says, they could not go back, because seeing a way of working makes you want it for your teams.
Context came before the solution. Rather than jumping to an answer, the team explained how AI changes the world, what the use cases are and what other companies are doing, then showed pilots already running in markets and asked to systematize, professionalize and mature them.
He was careful not to overclaim. The pitch was not that marketing would be most affected, but that marketing was clear on what it wanted to do, had a pilot, a use case and a plan, and therefore had the right to lead.
Sharing what he learned
Having led, he now helps other functions. He has spent time with the people, sales and R&D functions as they build their plans, telling them he was where they are six months ago.
He describes that as both a privilege and a responsibility: when a company lets you lead, you owe others help on the journey. The trait he sees in people leading in the space is generosity paired with humility.
That extends to competitors and peers. He has called L’Oreal, Coca-Cola and Heineken, including companies he knows are ahead of him, and says every one of those conversations expanded his view, including a use case he had dismissed because his own business is not direct to consumer.
With the people function, the intervention was to widen the aperture. They had narrowed onto one executional idea and had not mapped their use cases, so he showed them possibilities from compliance checking to internal chatbots that remove administrative work.
Capability is half the investment
Half the investment goes to people and change management rather than technology. His warning is that tools without capability do not scale, put as tanks that do not run without gasoline.
He describes this year as the first time he found a real capability gap when planning the year’s ambition, and one that will keep growing without a clear intervention.
The interim answer was an external speaker and expert coming in about five times a year, while the company ran a procurement process to find a partner to build the capability material, with a decision due in February and launch in the second half of the year.
He is explicit that others are further ahead, citing mandatory marketing training at L’Oreal and a larger HR effort at Coca-Cola. The internal system Mondelez is building is called Ada, standing for AI data and analytics.
The training itself has to break the usual format. Instead of an hour of lecture, he expects hands-on exercises, homework and tools, with the output judged on whether people got better.
Where he thinks this ends up
His first prediction is that as the cost of knowledge and of incremental assets falls toward zero, brands matter more than ever, because two people could start a biscuit company and have Super Bowl level creative the same afternoon.
That makes clarity a survival trait, because AI distorts whatever lack of clarity a brand already has. His image is a photocopy: a fuzzy picture gets fuzzier with each copy, while a clear one stays sharp.
His second prediction is a change in what wins. The last decade was won through agility and speed, with Oreo’s dunk in the dark as the emblem. Everyone will be fast now, so the next decade gets won by humanity and by the power of the insight.
He locates that humanity in time spent with consumers, citing watching people wash their hair with sachet shampoo in Asia as the most powerful experience of his career. Synthetic audiences can simulate things, but people create the real insights off them.
His third prediction is structural. He expects fewer permanent employees and more freelancers moving in and out, with a fracturing of job skills into pieces too small to justify full-time roles. On his own project there are more than 250 people on the agency side and a handful of contractors who move in and out performing client-side roles.
The race argument
His strongest claim is about first mover advantage. The drivers are the Size of your data, computing power and success signal everyone has access to same. Computing power is equally available and success signal is broadly similar within a category, which leaves who starts first in a world of compounding impact.
He contrasts that with earlier digital shifts, where laggards did catch up. The top search results and the most liked brands ten years ago are different today, which tells you there was no permanent first mover advantage there. He does not see how the same catch-up works here.
The conclusion he draws is that this is a race, and that sitting on the sidelines is a decision that will age badly. Waiting looks reasonable now, but he asks how those choices will look five to ten years from now.
He attributes the hesitation to fog rather than laziness. The period reminds him of 2000 to 2004 in digital, and confidence about who the most valuable partners or media companies will be in ten years is, in his estimate, at an all time low.
His answer to the fog is to keep moving through it. He tells his teams that it is unclear, but they are driving, because when the day clears anyone who stopped will be too far behind to catch up.
The last thing he wants measured is partnership. He believes the winners will be good at partnering because the answers do not live within any one company’s four walls, and predicts that internal and external partnership strength will correlate strongly with success.
Scope and limitations
This is an automated transcript of a single conversation, with the speaker’s own estimates rather than measured results, and the figures he gives for investment are described in round terms.
Source
This page summarises Transcript: Jon Halvorson (Mondelez) on AI in Marketing, by Jon Halvorson, Mariana Peneva, November 26, 2024.