"Our collective Blade Runner moment really concerns me"

m&k: Oded Lavie, you are one of Publicis’s most renowned AI experts. We’ve all been talking about this topic constantly for over a year now, but most of the discussions tend to remain rather superficial. To what extent is AI really changing our industry?

Oded Lavie: First of all, I’d like to say that, on a personal level, I find it difficult to be called an AI expert. I think the true experts are those who develop this technology and create models and platforms—people who have been doing this for years. I’m just someone who loves technology and its connections to our business. As for the broader discourse: Before November 2022, discussions about generative AI were already taking place in various small circles, but they were often dismissed by companies and agencies. ChatGPT 3.0 brought about the shift: it evolved at lightning speed from a fringe phenomenon to the mainstream. Now we’re in the midst of a very broad discussion—but not yet in the implementation phase.

 

Why is that?

In my opinion, the reason lies in the far-reaching implications of the technology. It truly represents a paradigm shift, much like in the early days of the World Wide Web. This poses a challenge for all areas of business: data, legal, copyright, ethics, talent management, efficiency, processes, the silos we work in, marketing, IT, and much more. The implications and challenges of actually implementing generative AI are enormous, and brands want to get it right. That means learning to crawl before walking and running. To talk first, before applying it. Smart companies viewed 2024 as a learning phase, testing several pilot projects with various partners to identify where the value drivers lie—and are now planning to scale up in 2025.

 

Are there any other reasons?

Another reason why we’re still in the discussion phase is the amount of misinformation out there about generative AI and what it can actually be used for. For example, some say it can reduce the effort required for creative services by 50 percent; others talk about complete production automation and how cost-effective that would be. Those are all fairy tales, if you ask me. We’re not there yet. And the gap between what you read and hear—what’s supposedly possible—and what’s actually possible can lead to disappointment and reluctance to adopt the technology. That’s why the pilot and experimentation phase is so important: you learn, understand, and continue to evolve.

 

Pascal Winkler, you work with Oded on various projects for Swiss clients and help implement AI technology at Publicis Groupe here in Switzerland. What is the current status here, and in which areas has Publicis already implemented AI?

Pascal Winkler: At our Swiss headquarters, our journey with AI began much like that of others in the industry—with initial experiments as various tools began to emerge. It was a bit like the «Wild West,» where everyone rushed into this exciting, unknown territory with great enthusiasm and wonder, but without much guidance or rules…

 

So what happened next?

We have moved beyond this phase to the actual implementation of AI in workflows and projects. In addition to the «visible» aspects of campaigns that are AI-generated, there are already many steps in the campaign development process that are carried out with the help of AI: AI-assisted insight generation, AI-assisted briefings, AI-assisted early idea visualizations or animated storyboards, AI-assisted voice-overs, and so on.

 

So the «Wild West» isn’t quite as wild as it used to be.

To stick with the analogy: While we discover new areas every day, we have now staked out several clearly defined territories in which we operate. We’ve defined clear rules within these territories and, over the course of the year, implemented a comprehensive training program aimed at familiarizing various stakeholders with AI tools and methodologies. So our teams know how to use them.

 

And is that how all Publicis agencies operate?

Over the years, I’ve worked at all kinds of agencies, and of course, each model has its advantages. But in the case of a technological revolution like this one, I can confirm the significant advantage that comes with being part of a global network. All of our agencies in Switzerland benefit immensely from the clear global roadmap and gain access to a suite of vetted tools «on a silver platter.» The image tools we use, for example, generate outputs that our clients can use legally, ensuring both quality and compliance in our AI-powered output.

 

Oded, can you tell us more about the global strategy and the process of identifying and implementing tools?

Lavie: Essentially, our goal is to create a «human-plus-machine» work environment and provide our talent with the best tools, platforms, and processes. This will enable them to do their work in close collaboration with AI. This is intended to enable efficiency and growth opportunities for our customers and our company, while taking into account the risks of AI, the pace of change, and the need for agility and flexibility.

 

What does it take to do that?

We gain insight into challenges and needs from various sources: trends we observe, investors and accelerators we collaborate with, and feedback from our 120,000 employees worldwide… and as soon as we identify a solution that addresses a need, a challenge, or a desire, we begin a due diligence process. This may include the quality of the output, potential use cases, ethics and security, economic feasibility, and the integration of the technology into our processes. If the basic thresholds are not met, we will face challenges in implementing a tool.

 

And you have established this process within your global Gen-AI partnership team.

Yes, I'm very proud of that. This process allows us to move quickly and meet the needs of our talent. It took some time to develop this process, and we're constantly looking for ways to improve and optimize it.

 

Since these are still rare: Are there "Best Practices", can you name any?

Case studies are still actually quite rare, but I’d be happy to share three with you. The first is a consumer experience initiative in which we put generative AI in the hands of consumers. A few months ago, we trained an image-generation model to understand a grocery retail brand and its merchandising vision. We then made this engine available to consumers so they could experiment with it and generate anything they wanted—in the brand’s style. We were amazed by the engagement and the results—we had over a million generations in less than five days, with an average of more than 1,500 requests per minute. In the end, one participant won a prize, and the item actually went into production.

 

And what about the second case?

The second case study comes from the automotive industry, where we decided to conduct a «virtual» photo shoot at a ski resort instead of a real one. We faced several challenges in the process—such as the fact that generative AI cannot create an authentic product, which is crucial in the automotive and other industries; we had to handle reflections and transparency, integrate the wheels into the ski environment, and much more. Ultimately, we published three images in an Instagram carousel, achieving about 85 percent cost efficiency and about 90 percent time efficiency—which is already impressive—but the impact of these efficiencies on growth, segmentation, localization, and personalization is even more exciting.

 

Let's move on to the last example.

In the last case I’d like to mention, we launched a new product within three months—including branding, packaging, marketing, advertising, and assets. This process usually takes months, but with our «human-plus-machine» approach, it took about ten weeks.

 

AI-generated or AI-assisted advertising has repeatedly drawn public criticism, such as Coca-Cola's Christmas ad or Jaguar's new campaign. Was this to be expected? Does the public first have to get used to the "state-of-the-art" gewmoan?

Winkler: The mixed reaction to AI-generated advertising, such as Coca-Cola’s Christmas campaign, likely stems from a feeling that it’s «good, but not great.» And that’s simply not what people expect from a brand like Coca-Cola, which is known for its exceptional advertising. AI-generated videos quickly and efficiently deliver 80 percent of the quality we’re used to, but the remaining 20 percent is missing. And that’s noticeable. Furthermore, it can feel like an emotional disconnect—or even a betrayal—when the audience knows they’re reacting to something created by a machine. Something bothers us humans when we feel that a machine is trying to evoke our most human trait: our emotions.

 

What is your personal opinion on this?

I don't think this commercial will be remembered as Coca-Cola's best work. My guess is that, at this stage, it isn’t even their most cost-effective «good» ad. Nevertheless, the lessons learned from producing such a campaign are invaluable and likely the main benefit for Coca-Cola and its agencies.

 

Douglas Rushkoff argues the following: When new technology emerges, everyone is eager to jump on the bandwagon. But if everyone does, in the end no one has a competitive advantage. Companies and agencies can all purchase the same tools. So will it ultimately come down to who is quickest to empower their employees to use them? Is that where the real advantage lies?

Lavie: The real advantage will come from three different, interconnected areas: talent, data, and technology. When it comes to talent, it’s «shit in, shit out»—pardon my French (laughs). If you don’t have talented people to manage the generative models using the right process, you’ll get mediocre results. Talented people are crucial for generative AI, just as they are for everything else. Today, you can start a podcast in two seconds—but how many podcasts do you actually listen to? Twenty years ago, anyone could start a blog—but how many blogs have you actually read? The tools themselves don’t guarantee quality; it’s the talent behind them that does.

 

And what about the data level?

If everyone uses the same models and tools, the question is how data can influence those tools—can stakeholders train a model? What data do they have to train that model? How do your data sources differ from those of the competition? As with talent, good data will improve results and become a competitive advantage—bad data… well, you know. Thanks to our transformation over the past few years, we at Publicis Groupe have very good data sets, both in terms of quality and quantity.

 

Let's talk about the technology.

Technology will become a commodity. So it’s not just about being first; it’s about implementing it the right way to support the process and being flexible and agile when it comes to adoption, change, and testing. All three of these things must happen—or be in place—to remain competitive.

 

How does Publicis plan to stay ahead of the competition—while, of course, remaining within the bounds of ethical standards?

Winkler: As Oded just described, the key is ensuring that our employees are trained faster and more effectively, that the right technology is in place, and that the data is available to feed and train the algorithms. But another important aspect is bringing our clients along on this journey. No one should use AI behind closed doors—and ethical considerations are absolutely central. It’s essential to engage your teams and clients in transparent discussions about the limits of AI use and, for example, how AI-generated content can be clearly identified by end users.

 

Are there any other aspects to consider?

Yes, and that means making AI easy for everyone to use. An important step we took a few months ago was the implementation of «GenCanvas,» a proprietary platform that integrates all of our shared AI tools into a collaborative workspace, similar to Miro. This gives all employees access to eleven different tools—including ChatGPT, Google Gemini, DALL·E, Bria, RunwayML, and ElevenLabs—in one place and makes collaborative creation easier. In 2025, we will extend this collaborative approach to our clients and support them in improving their skills and navigating the evolving landscape of AI technology in marketing.

 

Do you have a vision for the ideal future use of AI in advertising? Are there areas where you’d like to see it used, and others where you’d prefer that people continue to handle most of the complex thinking…?

Lavie: Wow, that's a big question (laughs). In Hebrew, we have a proverb that can be translated as «Prophecies are for fools,» so I try not to predict too much and focus on the here and now. But if I were forced to make a prediction, it would be that we’re entering a new phase of creativity. We’ll be able to do things we couldn’t do before due to time or resource constraints. We’ll be able to operate in a more personalized way than ever before. We’ll be able to gain more and deeper insights. We’ll have—compared to today—superpowers.

 

What does this mean for «human» creativity?

I think that once the initial hype dies down, we’ll know exactly where AI works well and where humans do. The balance will also shift: as AI improves, and the more we «delegate» to AI, the more we’ll continue to evolve and be able to create new kinds of advertising and experiences.

 

Maybe this is taking it too far—but don't you sometimes wonder if, somewhere in the intricacies of technology, there's a kind of "Spark" ...even though you know that's impossible?

Winkler: Oded, you're in charge of philosophy (laughs).

Lavie: Our collective *Blade Runner* moment really has me thinking. I think the answer depends on how we define that «spark.» Will AI be able to «match» the achievements of Newton or Galileo? Probably. Will it be able to do so under the same conditions as they did? No. AI and generative AI require massive amounts of data, significant computing power, and electricity, and they will determine things based on statistical relationships. That’s not a «spark» as implied by your question. What we’ll see is an acceleration of «sparks» from people who know how to work with AI.

 

Because of my philosophy studies, I'm finding it increasingly difficult to distance myself from the emotions or thoughts that AI evokes. As I find myself being almost fooled more and more often—does that happen to someone like you, too?

I started out studying math and philosophy and then earned a degree in psychology. I know that in some conversations with AI, I fall for an illusion, and yet it still happens to me. It will probably happen more and more often. AI makes decisions for us without our knowledge, and generative AI will guide us toward certain decisions. In some cases we’ll welcome this, and in others we won’t, and it will be up to us to decide which is which—and where we should invest more effort in understanding and ensuring that we aren’t deceived.