Transcription
Humans have to spend their time on the high-quality, high-value tasks, and we have to automate and push, you know, the more factory-based—call it—work to AI. Because frankly, humans shouldn’t be adapting 10,000 variants of an ad, you know, when a machine can do that in 20 seconds.
The average consumer receives between 3 and 10,000 commercial messages every single day. These messages can influence us to buy products we need or solve problems we didn’t even realize we had. Still, most of these messages are quickly forgotten. What if artificial intelligence could help ensure that an ad would catch your attention? It’s a custom-trained AI model can predict whether an ad will work, why it will work or not work, and what you need to do to improve it. Could new technology mean never-before-seen volumes of personalized content? You can see there are 13,000 variations here. I literally have to scroll to page 1300 to get to the end of this. And is there a future where we might not need humans to create the content at all? This idea to train models to help you make decisions, to inform ideas—we’re only getting started.
I am Raja Rajamanar, chief marketing and communications officer of Mastercard and a best-selling author writing about our industry’s future. Marketing as we practice today was founded on theories and principles from more than six decades ago, long before the advent of the internet, social media, or artificial intelligence. I have had a front-row seat as new technologies, cultural shifts, and an explosion of data have disrupted the ways of traditional marketing. To connect with the savvy consumers of tomorrow, marketing as we know it must take a quantum leap forward. This is quantum marketing. Absolutely like A1. Like unbelievable.
Yeah, I sat down with Stefan Pletorius, the creator behind a fully AI-driven system with the potential to transform the ad creation process.
Hi Stefan.
Hello Raja. How you doing?
How are you? Very good, thank you, Raja. Your—I don’t know if it matters—but your head is slightly cut off on this side.
No, it’s okay. My head is cut off from the… So far my head is quite intact, Stefan, thank you.
There we go. That’s better.
That’s better. Okay, great. So welcome, Stefan.
Absolute pleasure to have you here. And I sincerely appreciate you calling in from London. I’m in New York. And just to kick off for our audience, it might be a good idea if you can just give a little bit of background to yourself and to the organization that you work at.
I think I’ve got the best job in advertising. So as chief technology officer for WPP, I’m responsible for the group’s technology strategy, which is really—um—simplistically, the application of technology to the practice of marketing and advertising.
When it comes to new technology, most companies are thinking the same thing, and it is just two impactful letters: AI. The term itself has become a buzzword, but incorporating artificial intelligence meaningfully at scale is the real challenge. Companies and brands often have a CMO like myself and an in-house marketing team that might collaborate with ad agencies, which often operate with larger networks. In this chart, WPP sits right here as one of the world’s largest marketing and communications services organizations, with a vantage point for these changes and challenges in our industry. You know, obviously marketing is going through a mega transformation.
Yeah. Absolutely unprecedented, driven either by technology or cultural shifts or a data revolution that’s happening. What I wanted to ask you is what trends are you picking up in the world of marketing?
Look, I think the one trend that’s that’s—that’s already well on its way is the is media proliferation. So increasingly consumers—um—you know, consume more and more different channels and sources of media, and therefore it becomes increasingly complex for brands to reach them effectively because there’s so much fragmentation in the marketplace. If you combine media fragmentation with content, you know, volume plus audience targeting, you have this kind of incredible explosion of content that you need to make. You also have the reality of the consumer attention span shrinking by the day. And I’m told that now it is less than 8 seconds. So you have got this attention span of human beings, and then here we’re talking about creating an extraordinary amount of content across a multiplicity of channels, which are also proliferating by the day. What a paradoxical situation.
A very big trend right now is that people are beginning to realize that marketing is almost the last enterprise business function that hasn’t properly digitized or modernized. Right, but marketing to some degree is still a very, very ineffective, manual, difficult-to-manage process with many different subcontractors and vendors and agencies. Almost all our large clients are grappling with what is the right transformation they need to go through that sets them up for a future that is more streamlined, more efficient, more automated, ultimately because people are beginning to realize that if you don’t do that work now, you’re not going to be ready for AI. And so, you know, in an interesting way, it’s almost like the future is putting pressure on the present, if you know what I mean.
Totally agree with you. I think everyone is seized of this complexity, and everyone is trying to sort of figure out what should be the best playbook for themselves, and every day you are trying to think about it, you are already feeling that you’re obsolete because three more things have been launched while you are sleeping, and that has to really, you know, force you to rethink your entire strategic approach. You know, the magic of AI happens in business happens when you combine—um—you know, AI technology with data, proprietary data with unique domain knowledge. And I think the biggest fallacy on the data side is that people have been telling marketers that if you know who someone is, you can market to them better, right? So if you have personal data about consumers, then you can market more effectively to them. And frankly, that is that is not really true. That is not consistently true. Very often, non-personal data, contextual data about where you are, what’s the weather like, you know, what event is happening around you, um, what’s the time of the year, you know, these kind of things are far more useful in terms of predicting behavior than who you are.
And there is always a lot of debate saying that, look, AI cannot generate original, create, you know, creative stuff. It can adapt. It can put things together in more creative ways, but it cannot be a substitute for creativity by itself. How do you look at AI and creativity in that context?
It’s a great question, and and you know I’ve I’ve been debating with my team on will our machines—um—you know, be creative at any point in time for for years now. And and the conclusion I’ve come to is that it really depends on what you define as creativity. True innovation, truly new ideas, truly new connectivity of things that that people do is actually quite rare. So on the one hand, you have got commercial creativity, i.e., creativity that drives business results in a measurable fashion. Yeah. Then you have got the pure creativity, which is more for the self-actualization or self-expression. It’s more artistic. It’s more creativity for arts as opposed to creativity for commerce. And in the world of marketing, we are not creativity for creativity’s sake. It’s partly a definitional problem. But then, you know, more—more maybe pedantically—is AI useful in creating content? I would say enormously, enormously useful in creating content because if you think about brand building, brand building is a lot about consistency, right? It’s about differentiation, but it’s also around consistency, and and you know using consistent, you know, brand sort of, you know, sort of elements and design styles and aesthetics is a key to building consistently good brands, right, on a global basis, and machines are very good at being trained on that. So you can train a machine to use a specific, you know, to develop content within a particular aesthetic style. You can train it to represent a product completely accurately. So you know machines are actually very good at—um—you know, being used for marketing content. I think the more interesting question, Raja, is is what what future do we want? Right? It’s not it’s not a question of what machines could potentially do, but it’s for me it’s a question about what future do we want, and I’m not interested in a future where machines are creating all the content and humans are just sort of floating around on on rafts like in WALL-E, right? I mean, that’s not a future that’s interesting to me, and and to some degree it’s a choice you’ve got to, right? You’ve got to say are we building black boxes that that do the work that humans have always done, or are we going to build technology, technology or software that makes us more intelligent? And I’m far more interested in the in the latter.
When a brand decides it needs an ad for a product, it sets off a long and expensive creative process. By the time the ad is served to a potential customer, months have passed, hundreds of people have been involved, and often millions of dollars have been spent. Stefan and his team have developed an AI software that can replace some of the most tedious and time-consuming parts of that process and likely save companies a pretty big chunk of change. So Stefan, can you just give a demo? I think that’s something which will make the concepts extremely clear and it’ll bring it to life very effectively.
Absolutely. I’m happy to do that, Raja. So—um—what you’re looking at here is a configuration of our WPP open platform. It’s our AI-driven marketing operating system, and we’ve set up a fictitious brand here. It’s called Zia. It’s a perfume brand, so that we’re not disclosing any confidential secrets of our client. So all the data in here is fictitious. The brand is fictitious. The imagery is all fictitious as well, but it illustrates the point. Within this proprietary software, marketers can augment their process with various AI brains. For example, brand brains. These are custom AI models trained on data, brand assets, and details like tone of voice for any given client. To take it a step further, users can apply the brand brain to create a persona. This is an entirely artificial stand-in for the target audience that can give feedback on what will resonate with consumers. So let me take you through this. You have your persona. The next thing we’re going to do is we’re going to use that persona to create a synthetic—um—focus group. I can take, for instance, this handwash idea here and I can say copy it in there, and I say here’s my idea. What does—um—that eco-conscious Gen Z consumer say about it? And and now what you’re doing is you’re taking a human-generated idea and testing it against a synthetic focus group that’s been based on actual marketing and research data. And and so you get a quick response, but you can also then speak to this persona and say, “That’s cool. Um, you know what should the the packaging be?” Um, and you can then effectively have a conversation with the synthetic focus group. And so you can see, Raja, this is this is stuff that we always were able to do in real life in marketing, but never at this scale, never at this speed, and never at this incredibly low cost, right? Because you know what this would would have taken in the real world to to build something like this, organize the focus group, you know, organize the events, etc., etc. What does that look and feel like to you? Is that quantum marketing?
It does. In fact, you know, one thing which is very fascinating is even as we are sitting here, the speed with which it is generating and the kind of variables that you can play with. It’s pretty rich. But have you validated that this is how real-life focus groups are also actually responding, so that I can depend as a marketer on this data very with credibility? I say, “Yep, this is something which I can believe.”
Uh, it’s a fantastic question, and and and I wish I had a greater answer for you. I think I think the—um—you know, more traditional, you know, ways of running focus groups are definitely going to be influenced by some of these techniques. I think practically some of what we’re doing here is going to be impossible to be replicated because you know you can’t train a person on a a thousand-page, you know, marketing best practice playbook. Um, right. Um, so I I think a lot the the way that that traditional focus groups have been have been used has been really relying far more on people’s intuition and their existing knowledge. We’re creating something here which is effectively superhuman, right? We we’re creating something that has knowledge beyond any kind of that any one individual consumer could have in their minds at that that single moment in time, which is actually what makes it so interesting.
Right.
Got it. Got it. Okay.
Artificial intelligence has come a long way in recent years, but a combination of new technologies might be the answer for marketers and brands seeking precision. Let me show you one one other additional example. So we’re going to say I want to create some content for the beginning of an ideation session, and—um—I’m going to create some some images and some some taglines, you know, just to get, you know, as thoughts started. And for each intersection of the the matrix, you now get a suggested headline, right? Elevate your hand care ritual with Zia, sustainable symphony for the senses. You get a call to action, join the ritual. And you get a set of sample images that have all been generated based on the brand guidelines, the brand brain guidelines for this for this product. Right, so these are not consistent and and sort of aesthetically similar by chance. It is because we had we had the custom AI model, the brand brain, um, sitting behind this for the generation.
Great, Ron. These visuals, Stefan, firstly, it’s very, very impressive. Let me start by saying that, but just one question is on visuals. Are they totally synthetic, i.e., composed by AI end-to-end?
It’s a great question. Um, these are entirely text-image generated, but it’s a custom-trained model, but the product imaging—images that you see in here are not AI generated. These are 3D renders that are composited with a generative AI background. And the reason why we do that is because you want the the bottle in this case to be absolutely picture perfect. You’ll see that the lighting and the shadows and so on is different by by image, but the actual product shot has to be perfect. And so this is why we use a hybrid of 3D rendering and generative AI. So here we have a matrix of content, and you can see there are 13,000 variations here. Right, so I literally have to scroll to page 1300 to get to the end of this. It becomes overwhelming as to where I should focus and what I should zoom into. So does the engine sort of assign some kind of probability and say, you know what, these are my top five recommendations, or these are the top three recommendations? Is there anything like that?
Yeah, exactly. So what we really need to to do, and this is where the the skill in using these tools comes in, is we need to build models that help us to understand whether this content is going to be effective. And so what we have here is we have all these different variations of the ad combining the product shots, the—um—the logo, the kind of the background image, the headlines. And you can see here that—um—you know, we can see that there’s a predictive score against each of them. Now this predictive score is based on a performance brain. So this is our our AI model that we’ve built to predict whether these these content elements will actually lead to either brand lift or—or or direct response or sales based on the data that we have. And and so what what’s really, really powerful here is that you can start seeing very, very visibly and and kind of quite intuitively, you know, how different content elements will perform. And again, this is not just about automating some production process. You know, by putting an ad live that has a 30%, 36% prediction score of being successful against the paid media impression, you are sometimes wasting, you know, tens of thousands if not hundreds of thousands of dollars of of media spend against something that’s not going to resonate with the audience. So you know this is a great example of of, you know, showing how we can use AI to create more efficiency not only in the production process but also in the media allocation process. Um, you know, uh, using these technologies, increasing use of AI is looming across many industries, and big impacts are already here. Decisions have to be made about how best to incorporate AI and the inevitable tradeoffs, the amount of output and the amount of relevant, very pertinent output with such a vast variety and in such a quick time. Does it have any negative implications or any implications, positive or negative, uh, on the agency staffing?
Look, the the the impact of technology in any knowledge work—um—and this has been the case for for decades now is that every time that you bring technology into knowledge work that automates mundane tasks, it pushes people to become, you know, more intelligent and to to apply their minds to higher-order problems. And so having someone sit in Photoshop all day and create 13,000 variants of a particular ad manually, you know, one by one, is not really a very stimulating task, right? And so these kind of automations are built so that you you free people up from doing menial work and really think about the design or to think about the messaging or to do more research about the consumer audience or to aestheticize the the product shots better. Right? So there’s the it it shifts work, you know, and I think this is the key thing. I I really don’t believe that AI destroys jobs. I think AI shifts tasks, right? So AI means that you have to work differently, and people do do need to learn new tools and new technologies. Um, but I I think it’s a I think it’s a net positive, right? I mean, people are not meant to be assembly workers, and a lot of what we we’ve done in the past in in digital, you know, kind of production work is really kind of just, you know, quite menial assembly type work.
No matter which way you look at it, we are in a new dawn at the intersection of AI and jobs. With these rapid advancements, companies must evolve quickly to have any hope of keeping up. So based on your extensive experience and what you are seeing and what you’re experimenting and playing with, what will be your advice to marketers?
I I think there’s only there’s only one important piece of advice, and that is—um—start doing things right now. You know, don’t don’t pontificate about it. Don’t wait. Don’t watch other people doing it. Get stuck in, and and you know start learning by doing. Um, AI is a very interesting technology. It’s a It’s the kind of technology that you know seems—um—obscure and maybe a little bit kind of intimidating at first. And as you start working with it, you start changing the way that you think. You become more aware of how you think. And and as soon as you you start realizing that for yourself, you reflect back on the way that you do marketing much in the same way that you’re talking about quantum marketing and and your positioning around that, you you start questioning the traditional ways that you’ve been doing marketing or you start thinking about am I using the best data or the right data, you know, or the right process to come up with this optimization or this decision? And so I I think it’s a it’s an incredibly powerful—um—process to go through, but you cannot you cannot do it theoretically; you have to do it practically, and I would say that applies to the entire organization. This is not something you don’t become an AI-driven company by having an AI department. You have to do this across the entire organization, from the CEO, not just the CM, CMO or the COO or the CTO, from the CEO all the way down, and that’s when you will have the big the biggest impact from that. So if if you’re in a in an organization that’s led by people that are naturally curious and naturally technologically—um—you know, sort of comfortable, then that’s you can do that easily. I I know some organizations have have more trouble with that. Um, but and so maybe we’ll find it more uncomfortable, but there’s there’s no alternative. You cannot outsource this process.
[Music]
[Applause]
[Music]
[Applause]
[Music]
[Applause]
[Applause]
[Music]
[Applause]
[Music]
Hey, hey, hey.
[Music]