Welcome to the Marketing AI SparkCast, where AI meets modern marketing. Hosted by Aby Varma, a global leader in marketing, AI, and innovation, each episode brings you real conversations with leaders putting AI to work in strategy and execution. This episode is brought to you by Spark Novus and the Marketing AI Pulse community. Here's your host, Aby Varma. Welcome everybody to the Marketing AI SparkCast. I am your host, Aby Varma. I am the founder and CEO of Spark Novus and the Marketing AI Pulse community. This episode is sponsored by the Marketing AI Pulse community. If you're not a member, you can join for free at marketingaipulse.com to connect with other marketing leaders who are navigating AI so that you don't have to do it alone. Speaking of which, you can mark your calendars for a really exciting session coming up. We have the Claude ecosystem training for marketers. It's a three part certificate series running from July sixteenth, twenty third, and thirtieth from twelve to 1PM eastern. We cover Claude chat, design, and Claude code. Podcast listeners get 20% off, and you can use the code spark 20. So with that out of the way, I wanna talk about today's episode. We are going to look at AI's impact on consumer insights, synthetic research, creative testing, audience segmentation with David Jacobson. David is the global senior director applied consumer science at the Coca-Cola Company. He brings twenty five years of marketing strategy and experience, including role of Publicis Sapient in PwC. Super excited to have him. David, welcome to the past. Great. Thank you, Aby. Great to be here. Appreciate it. We're recording this July per David, so thank you very much for for pending your day off with us. Do you have any plans for the long weekend? Yeah. So fourth of July weekend, think I might check out the annual fourth of July day parade in Dunwoody, but it's gonna be hot out there. So we'll see how it goes. Wanna bring my little puppy, but it might be too hot for her. And then I think tomorrow night, got Don McLean playing at City Springs in Sandy Springs, so I might check that out as well. Well, thank you again for making the time, David. Now we would love to know a little bit about your your background. I think you've been with Coca-Cola for about almost three years, but you've been in this industry for twenty five plus years. Tell us a little bit about your role at Coca-Cola, and maybe walk us backwards on how you landed at Coca-Cola in the world of marketing insights in in analytics. Absolutely. Start in traditional advertising in ad agencies at TBWA\Chiat\Day in Los Angeles and then eventually in New York. But as your traditional kind of brand strategist, researcher doing kind of traditional research and focus groups and understanding the consumer, the voice of the consumer, writing creative briefs, working with creative teams to develop campaigns. And then I was, I guess, lucky enough. This is the right place at the right time in terms of the evolution of technology and digital where in the late nineties around the dot... First.com rise and bubble and burst of being around how digital technologies started to shape the advertising industry. And we started building websites and building experiences and digital and just sort of was at the front row seat to how digital was starting to change, not just marketing, but research and everything around the marketing industry. Jumped into the deep end, really understanding digital and Slot really shaped the industry, not just in how we develop campaigns and then developed to say marketing campaign wasn't just how we include digital and websites and that, the entire notion of how we create experiences in marketing really changed. And then that shift continued as I went from from there. I founded my own ad agency in Miami where I'm originally from with a couple partners. So we started to do a little bit of everything, experiential marketing, website, traditional advertising. That led me to SIC, which was very much an experiential agency rooted in technology. It was there when we were acquired by Publicis. And in that world really started to change how I looked at marketing, the role of research and insights, because it really was about really not just understanding the consumer and developing a brief and a core creative idea or campaign, but really how we can engage consumers in all aspects of marketing across multiple touch points. The number of touch points increased dramatically exponentially in the last ten, fifteen years or so. And then I was moved from Sapient to PwC, similar consulting role in Experience Consulting. And... But one thing I always wanted to do was go to the brand side On the client side. For years in advertising and consulting and having moved here to Atlanta ten years ago, Coke was always one of those brands and companies here locally in Atlanta, obviously, but even globally that is well admired. Yeah. I got the opportunity to come to Coke from PwC, and it's been great so far. So at Coke, I sit on the global human insights team, and we do a lot of piloting of new research now a lot with AM audience segmentation, how we are applying new technologies into what we do in terms of understanding consumers. We really have a human centric approach, which I even learned at PwC really in terms of service design and human centered design. And we apply that to our marketing, Really, how can we understand consumers and why... Whether we're creating a campaign, piece of communication, or an experience like now, Cup is upon us and one of... Actually, it is the biggest activation we've ever done globally as one of the least launched here in The US and Mexico and Canada. So it's been great to see how we can really apply that to every aspect of the campaign, not just ads themselves, but FIFA fan zones at at watch parties and at stadiums, every little experience, the trophy tour. So it's just really taking a full lens and human centric approach to how we look at marketing. Amazing. Amazing. Yeah. That's quite a storied career. I would really like your perspective. Where is AI creating the most meaningful change when it comes to insights today? Yeah. I think we've... We were... I guess, a company, I was lucky enough when I started three years ago that AI was already starting to be embraced by the organization. We created enterprise or have enterprise versions of Copilot and ChatGPT that we can use that's integrated into our systems. We were encouraged to to use it. There's training webinars and sessions set up for everyone from basic AI understanding to prompting to using now AI to create images and stuff. So there's definitely a organizational there to get people using it and not be afraid of it and encouraging people to adopt it. So that's been great. In human insights and research and everything that we've been doing particular, it's been great to just see how AI, there's so much components of what we do in insights and audience segmentation and understanding consumers that AI can use. I guess the big question... One of the big questions for us is there are some components of understanding humans where fundamentally, it's probably better to understand humans directly talking to humans versus AI. There's this balance that we've been trying to strike between, yes, AI can be very efficient. It can do things a lot faster. They can come up with answers really quickly. The question is, are those answers... Are we confident in the answers? Are we confident that the answers AI gives us, whether it's a synthetic persona or testing creative with AI, is really being confident in that the answers we're getting are actually what humans are actually saying, experiencing, will react or engage with a piece of content or marketing the way that we typically would want. So we do still do a lot of human based research or traditional research with focus groups or diaries or surveys, ethnography studies, etcetera. So it's really a balance. Like, we do definitely go by the mantra that that there needs to be a human in the loop, even if we're using AI. And then we're also... We have an AI governance portal at Coke where every AI project needs to be submitted. Not necessarily to govern and say, this is how you should or shouldn't do things, but really to help people and guide people through the process. Also, we've done similar projects all across the organization, helping to scale that or not redo projects or spend money again when we've already done a similar project or share resources across the organization as well. So, I mean, sounds like you were steeped into it. And, I mean, I want... Maybe the first topic to me is... You mentioned that synthetic personas. Right? That is synthetic research Yeah. Overarching is just getting a lot of attention. Since you're right in the middle of David, maybe for our listeners, if you could talk a little bit about synthetic personas, what are they? And where can they help marketers, and how are people using them? Yeah. I'd say the basic definition is really just a synthetic version of a human consumer. Now that can take on many different ways or manifest itself in different ways. Like, you can have a digital twin, which is like a digital version of me or you that someone can interact with. You can create synthetic respondents in a quantitative survey and create... Use the LLM, create a purse... Like we would create a a regular persona in the traditional way. Here's a demographics. Here's a psych graphics. Here's their passions and lifestyle interests. You have all the different attributes that make up a persona that we would use in a traditional non AI way. Now we're just gonna program the LLM with those same inputs, all those attributes, certain age or show income, lifestyle interests, passion points, motivations, values, etcetera. So just like we would create a persona on a PowerPoint slide or whatnot, now we're gonna just use those same attributes to train the LLM model. Now survey based approach, you can say, want 60,000 respondents. So now I can say, do I want 5,000 each of a certain type of respondent in different age ranges, different attributes, etcetera, which is one way to say you have a very varied group of respondents. So it's easy to create 60,000 respondents in a very short amount of time versus in traditional research going out and recruiting 60,000 people is time consuming, often difficult to get the exact time... The exact type of person you're looking for, and also can be very expensive depending on also the market. We're typically doing research on a global basis for a global campaign like for Coke. So multiple markets create multiple different situations and challenges in recruiting those respondents. So synthetic makes it a lot easier. Now what we... Like I said earlier, we... One challenge is it doesn't replace human respondents in every situation. So we're really learning and piloting where is it best used for, what situations, what parts of the creative development or marketing development life cycle is it best used for because it's not gonna be great for every situation. So we're learning that. And there's also a lot of different companies out there. You can do a quick search and find tons of synthetic persona type of companies or that have that expertise or that that skill set or functionality capability. Because the other component is is there are lots of companies now that are aiding in getting human research respondents, but using AI and helping in that process or AI and going through the answers of both unstructured and structured data to help speed up the process. So still there are AIs, different flavors or levels of AI involvement from pure AI, you know, synthetic personas to using AI with human based respondents to help with the efficiency and speed of doing more traditional research. So has a full spectrum of use cases as well as the ability to to constantly apply it. Now you mentioned increasingly that synthetic personas are not a good fit in all situations. The general rule of thumb, is it like early research that synthetic personas are best for, or what is your overarching guidance? And the companion question to that is accuracy. I mean, it all boils down to that. Like, if your synthetic personal respondents provide a representative response to real human beings. Like, I I don't know what is the standard deviation that you guys see in general or if it's an industry standard. Yeah. It's a great question. Yeah. I think as a general rule, I would say yes. So like you said, using it for earlier stage research, we see a little bit more confidence in just say, for example, you have 10 different ideas that you're exploring for a campaign or for a a brand platform using early days synthetic personas or to help funnel it down to the top three or four ideas, building confidence in that ability to say yes. We can get similar responses a lot quicker, a lot more cost efficiently to get to the top three or four ideas. But then once we start to really hone in on those ideas and fine tune them, then as you get a little bit later stage or more towards the final development, we're still gonna do more traditional human based pretesting or copy testing or research and then validate that, especially for larger campaigns. So that's definitely one area more confident. We're generally seeing probably... If we can get the AI models to be 80 to 90% accurate, that's pretty good number. We do... The good thing about... At least for us, we have tons of historical research about our different audiences and markets. So obviously, we always want to know what are the latest trends, what's happening right now. But we do have a pretty good idea of, like, historically, what switching behaviors people have, what drinking occasion people have, what motivations, etcetera. So we're checking it against that to understand, does it make sense? Just have to do like a gut check based on most experienced and past research of does it make sense what we're saying. We've done some pilots with some synthetic persona AI vendors where it's easy to get back a very large report that looks very robust and chock full of insights and seems very thorough just like any one of us interacting with our AI model individually. It can give you an output that looks very professional, Gary put together, makes tons of sense. But are the insights truly insightful? And that's what we're starting to see a lot of It's easy to get a large volume output, a 100 page slide deck that was mostly put together through an AI model after the synthetic persona research. But then the question for us is are there insights there? So I have seen some instances and some examples where the the... Which can happen also in traditional research, I might say as well. Like, that happens. We didn't... The hypothesis you went in with, maybe it was validated, maybe it wasn't not, but you didn't get maybe the insights you thought you were. We just need to be careful that, like, the insights truly we think are insightful that it comes down to a core human tension or human need that we're uncovering or learning from. So that's just... That's the watch out. But, yeah, in general, it's feeling more and more confident, but in the right use cases would be the Now a lot of what you said is so heavily predicated on data quality. Like, what is the data that you're using to train the AI in order to leverage synthetic personas? Because if you don't get that right, it's garbage in, garbage out. That is such an essential part of that. But do you see do you see the results vary dramatically based on data quality? Because, obviously, when you're doing research all around the world, there may be data quality varies dramatically in some countries versus others. And does that have a bearing on the right use cases for synthetic personas? Yeah. It's a great point. 100% data quality in the old adage is definitely true with AI, garbage in, garbage out. You really need to be careful, like I was saying before, like, we have so much data and research just historically in in our datasets. So we are using that. I think we definitely, like, with our enterprise versions of the AI models where they have access to so a lot of our historical data as well or presentations, papers, whatnot. So we're constantly running it against that and using those to train the models. I think one thing we are encouraged not to do, which I think some people that use just your off the shelf AI models or the... A non enterprise version of ChatGPT or Claude or Gemini or whatnot. Like, we're... Need to be... The balance of it drawing upon general knowledge that's out there on the Internet or that the LLM is drawing from that's public versus our enterprise data and that kind of quality. So the data that we are feeding into the model to train a persona, for example, where it's coming from our internal datasets that also use our third party data partners, whether it's a Nielsen or or whatever. Like, that we're... We already have trusted Data there. Like, we're confident in that data for the most part. So that... Using that to train the models, we're very confident when we would not wanna say, let's go out and just comb the public domain or public Internet and have the models scrape data for that. There is a balance because there is... Like, we do wanna know, like, the latest trends, the latest things happening in a particular local market. If we're looking at media or a different activation in the local area, I also wanna know what's happening politically, what's happening with the economy, what's happening with culture, those local areas, what's happening with content creators. So there's an element of data quality and trusting the data that we're training the model from combined with... We also wanna understand culturally what's happening in that world or in that local market from a trend perspective or other areas, other data signals, should say, that we couldn't... That we don't have internally, that we could either get from a third party source or from from that local data source. Yeah. It's always fascinated me when thinking about synthetic persona is how do you, based on the analysis using synthetic persona, how would you gauge trends versus a mistake? Like, but at the end of the day, it's not real. But is... Does that is that a... Is is there ways to mitigate against that where you're using synthetic persona that you're seeing trends emerge, and you're like, wait. This data, is it a genuine trend or is it just skewed? Especially in a consumer, highly cultural consumer facing brand like Coca-Cola, which is so intrinsically involved with all of those external factors that you're talking about. What is... What's the approach there out of curiosity? Yeah. I think it is what we get an insight or a trend that's emerging or that we see from AI or from persona. I think just like in traditional sense, you just need to validate that. One data point is not a trend. If you start to see it multiple times in multiple places over a time period, then it becomes something that that is replicable, but also is actual... You start to feel that it's real. So it's not just an isolated data point. So you still need to validate that, which I would say, let's not use AI to validate AI generated insight. Let's see if we see that same point elsewhere. Talk to local experts. Talk to people in the local market. See if they're seeing the same thing. Again, that's a point where you may then say, let's check this with real humans. Let's get a... Whether it's a a survey or a focus group or talking to local experts or whatnot, but ask them if they're seeing the same thing. And if they are, to what extent? Is it isolated to a specific niche group or smaller segment? So, yeah, I think that's where there's a real need to keep that human in the loop to validate the insights or outputs from an AI and and validate that it's not just an isolated data point or something that's the AI maybe picked up on a signal. But the AI models in general, it's really good, which is where one of the things you brought up in the beginning was audience segmentation. Like, AI in general is really good at averaging tons and tons of data input and coming up with the next best thing that should... That either you're asking it to do or it's interpreting what could be next. So it's making that assumption. So if it's starting to see some little data signals come out there, it's gonna extrapolate that out and say, this is what I think could happen next, or this is what you're asking me to do, or the model's asking... I'm asking the model to do. It wants to give you the right answer, but it also wants to give you the answer that it's expecting to happen. So if there's a few little data points, like a trend versus something that's maybe not gonna grow into a full blown trend, but it's isolated. Like, it doesn't know that yet. So it could pick up a bunch of things that are happening and say, yes. I think this is gonna continue to grow and grow to additional audience segments or different age groups. But the the reality is in real life that it may just stop with a small group or in a small local area and not extend to become a full blown cultural trend. And that's where I look at global... Kantar Global Monitor or Morning Consult or all these different companies that also do trend reports, it's good to validate and say, are they seeing this with their models? And they also now are all... They have their AI attached to their trend reports or their reporting. So we can see if it's... If it is a local thing or if it is a global thing or if it is something that just happened over a three, six month period and then start to wane or die out. It's where some real human intuition and checking these to occur, I believe. Now that is that is really fascinating. So you mentioned audience segmentation. I'd like to unpack that a little bit. Like, how has AI changed audience segmentation in in in your world? I think audience segmentation, if you go back even before synthetic personas and even before a... AI became such a widely available tool to all of us through through the AI models, like, in segmentation and using technology or the earlier versions of that ChatGPT became so commercially available. Like, just that amount of data across multiple segments, that's what LLMs or just data analytics and automation was really built for or can... One of the key use cases. So you take lots of data from lots of different audience segments and find all the different commonalities between different groups of people. So that's inherently like what an AI model can be really good at or an LLM in general. So, like, that takes a human, like, just lots of time and effort and lots of Excel spreadsheets and combining different datasets to to say, here are all the common attributes of these group of people that share these traits. An LLM or AI model can do that really quickly and also now connect into different platforms. So understanding which audience segments share different different... Or share different attributes or traits, that's pretty straightforward to do. Now you can... Like, a lot of what's happening now is how you can use different datasets into it. So what became a very laborious, very kind of niche skill set for, think, very highly trained like data scientists and data analytics professionals to understand how to fuse different data sets together based on lots of common attributes. An AI model can do that much quicker and much more efficiently. So you're taking different data sets, not just in... Once you have your audience segmentation, now you can combine it with retail data, with first party data, with lots of different third party data sets. So that's what we're seeing now, not just internally at Coke, but with a lot of our partners like our agency partners. But we're looking to combine those all those different datasets along with the audience segmentation data to then determine, here are the niche of groups of people that have the attributes and characteristics that we're looking for, like consumption moments or purchase occasions that create those data signals to say, yes, we wanna match this person in this segment with this piece of creative and this media channel or platform like in Meta or TikTok or whatnot. So combining all those together to... I was talking about audience segmentation and personalization and optimization of creative ten, fifteen, and more years ago. But the reality was to do that at scale with tons of different segments and tons of different iterations of creative was very and also very costly. Now that's becoming a lot cost efficient and quicker to implement as well as scale. So everything I just said to understand how you can create not just an audience segmentation, but different sub segments of groups that are very important and exhibit the characteristics of what you want from a high performing segment. It may be a small number segment, but very high performing. In the past, you would wanna find the the efficiency of going after as many people with the qualities that you want to find from a high performing audience segment that have the performance drivers that you're looking for. But now with personalization becoming a lot more easier and attainable, you get... You can actually create customized creative from a lot more multiple segments much more cost efficiently. So you can small sub segment of people that may not have the volume you want to sense, but are still very valuable. Now you can create that creative or that offer or that promotion specifically for that group in a cost efficient way. So it's just allowing personalization at scale based on the audience segmentation that the AI can help with. And then combining those all those different data signals together give you that confidence as well as that that signal to say, this group can be very valuable. And now let's customize the offer promotion message, etcetera, for that group. Awesome. Well, that's super interesting. And you mentioned, David, creative testing. What are you seeing in the creative testing arena? Yeah. In the pre AI era, a lot of it was your... You do your traditional testing, whether it's focus groups or individual testing with your traditional companies like Kantar and Ipsos and whatnot that would work with. They put the creative in front of as many people as possible to see what they thought of the advertising. So AI is involved and has improved that in many different ways. Like the old school way twenty, thirty years ago was you have animatics. You have storyboards. You have different early stage creative ideas that you put in front of people to see if it resonated and engage with them. But the the stimuli they put in front of respondents was very rudimentary. It wasn't polished. It didn't look like a final ad. So there's always this challenge of, did people really understand what we're trying to communicate? Did they they get the idea? Over time, it... And the ability to create more polished creative became easier to put in front of people at least to understand whether they got the idea or they got the message, etcetera. But now you can use AI to generate stimuli and ads that look pretty polished. So that's eliminated that a little bit. But I think AI comes into play in a bunch of different areas. One is you can have AI in the sense that, like, synthetic personas like we've been talking about with the creator in front of a synthetic persona and ask the persona what they think of the advertising. Just like we were saying before, there are challenges you need to watch out. Early stage, probably fairly confident that we can get some feedback from early stage. But the more final stage creative, ready to run-in the air, run on specific platforms like Meta and TikTok. And I think you gotta be a little bit more wary of is the AI response giving you what you wanted to be in or say, or is it actually understanding the context of the platform or the situation that people are gonna experience the ad in or the the content? So there's an element of that. And... But it takes on... There are lots of different flavors of AI, obviously, just like with Persona. So there's the the... All the different traditional research companies now have AI models that they trained from thousands and thousands of ads that they've tested over the last decades. So they've taken actual human training their AI LLM models that essentially do the same kind of testing that you would perform on humans. So those are, you know, I've seen 90% confidence levels and higher with some of our partners and vendors. So pretty comfortable with using some of those approaches. Again, doesn't replace all of human testing, but it does enable the testing of a lot more creative stimuli as well as creative adaptations and variations like we're talking about with the personalization. So we can test a lot more versions, lot more cost efficiently. The other component is, yes, we're a global company. So there are going to be challenges like we're talking about recruiting respondents in certain markets. In Asia and Africa and smaller countries where it's harder to get to a big, you know, the population, it's much harder to see if their craves going to actually make an impact. On one hand, AI can start to replicate some of those, again, from past human responses and training. So they have direct AI models that test the AI or test the ad based on the training of the AI model directly, but also synthetic personas and digital twins and whatnot that you can put the creative in front of. So it also depends on the type of creative testing you're doing. We're still experimenting and I don't do a lot in the research or innovation area where we're testing out new drinks or new flavors and stuff like that. But that's another area we've been piloting and seeing how can we test new creative variants, new flavors, new areas of exploration, not just whether a message will resonate in that. So it runs the spectrum. So, yeah, the models are getting much more robust in terms of how we... The confidence level we have in the AI trained version of models, especially when it's been trained on hundreds of thousands of traditional human based research. And now the AI is just allowing us to do that similar research much quicker and cost efficient. Are you saying that this sort of resulting in an increased velocity of campaigns just because now you can both on the creative production side, but also on, on the testing side. So I don't know if there are a number of campaigns pre AI or x. Are you seeing them velocity increased to two x or more just because it is not possible to do that? Well, interesting question. I wouldn't say more campaigns necessarily. The the speed at which we can go to market and react. So the time lag between research, it... Actually, the insight created from the research and acting and leveraging that ease... That insight to create a marketing message or a campaign, that's... That I think has been condensed down, which gives us a lot more confidence that, yes, once you have an... Because typically, it could be a year or more sometimes from that initial research to the marketing campaign seeing the light of day. And things change in a year very quickly now. So was the trend or the insight that was originally based on, is it still valid a year later? Many times it is, but now that time is compressed. So we can get from the time from insight to brief to create development to on air or in market is a lot shorter, quicker, which is great. There are definitely cost efficiencies and just the creation of the campaign as well. And, yes, if you use AI to generate a campaign potentially, but I'm not talking about just replacing human production with AI production to do that. That's part of it. That's a whole separate conversation. But, yeah, I think you're seeing the ability to get to market quicker, more cost efficiently. So they... I think it... We think more confident in the ability to then localize those campaigns and how that they customize or tailored to individual local nuance, local data signals, local trends, local audience segments, etcetera. So the... We... And that's a shift to what we're seeing. Maybe another other global believer, Procter and Gamble, other global brands or companies probably as well or... Which is true in marketing. That's not true because of AI just in general. Like, the ability to act and activate a campaign on a local level that resonates with, locals' audience segments has always been true. So the ability to do that now is just just easier, more cost efficient, and quicker quicker to market. I hear stories and read articles about autonomous... The future is going to be autonomous and self correcting campaigns or campaigns adapting to near real time audience behavior or audience trends. Do you prescribe to that idea? Are you optimistic about it, or do you think it's all a lot of hype? No. I I prescribe to it. I do think you need to have, like, a healthy bit of skepticism and not just take everything at face value. But I do think it's definitely happening. I do think I see an acceleration. I see a use. I do think, obviously, reading some of the articles coming out of Cannes here, AI obviously was a big buzzword. But I saw an interesting interview with Martin Sorrell, and he was... And I agree with what he was saying, which is basically, yeah, there's a lot of hype around AI and data video with a... But the true value that's being created was his point, which is a lot of what we've been talking about is, like, the use of AI with just data and analytics and especially when it comes to marketing and campaigns, like the actual development of the strategy and media and activation of the campaign using AI to crunch all those numbers, crunch the data, be much more efficient and predictive in planning and buying the media, as well as where the right touch points are gonna be at different times and optimizing in real time and seeing more real time data signals, and AI can capture that flag both opportunities and some... Whether it's negative performance or positive performance and optimize accordingly. So a lot more... I think there's a lot more value that can be created by using AI in that component of the marketing life cycle versus just the research or the upfront because when we talk about audience segmentation, like audience segmentation is gonna be actually activated by the media team and the media plan to go into market on different platforms to push different creative to... At different times to different audiences. So how that is optimized? On one hand, for many years, we've been left up to the platforms themselves like Meta or Google to do a good amount of self optimizing or recommendations on their side with their algorithms. But now I think a little bit more control is coming back to the brands and the agencies by deploying like self made or tools that look at these data signals and AI can surface that really quickly, can look at tag and look at all the different elements of your ads that are performing well and easily extrapolate. The ads that are performing the best have these elements or have these messages or are talking to these audiences. Now let's optimize in real time really quickly and either reduce or cut off some of the low performing stuff and really optimize or amplify some of the stuff that's performing well. AI can do that extremely well and efficiently. I mean, speaking of Cannes, one of the things I read was a big travel brand was using real time weather patterns to end up marketing specific deals for those destinations, leveraging it delivered on the fly. I think, obviously, it's got the creative guardrails and the messaging and all that kind of stuff, but that's fascinating. That's as close to autonomous as one can get, and I think there's gonna be more and more of that. And I think one of the things that is enabling this is AI agents. So I wanna get into that a little bit. How are AI agents beginning to show up in the world of marketing analytics and insights? Yeah. It's a good one. A... Agents is definitely a hot topic right now for us. So it's basically any process that is repeated time and time again, which are a lot of... Definitely when it comes to marketing, a lot of performance, marketing, and data analytics processes and ways of working are done over and over again. Like, you're constantly doing the same campaign report, the same quarterly report, looking at the same metrics from a campaign to see whether it's performing or not. So those are things that are repeatable that agents can be really good at helping to make more efficient, not have to spend hours and hours consolidating the data and reports, but really focusing on the output of that report. So the agent can... Once you understand all the different data inputs, it can grab that on a regular basis. So the automation of campaign performance report can really be much more automated using an agent. So that's one area. Another is like I was mentioning with all these different datasets that we want to fuse together or combine whether it's first, second, third party data. So how can I have an agent that's into an individual date and those agents can now then talk to each other? They have one agent that's hyper focused on extrapolating data or different components of one day. Say it's an internal, like first party data and then another agent that works with, I know, a second party like retail media data from a partner. Now how those agents talk to each other to see how the data fits together or the common attributes and where we can develop some insights. So there's agent to agent communication starting to happen. But there is a watch out with that because, like, you're asking about data quality before, one piece of bad data from one agent that gets it from one day that hallucinates or gives a false positive or whatever, then that gets used by a second agent, and then it starts to expand and grow. And so that one little error or that one little, like, lack of confidence in the data can get extrapolated out and make the whole thing a bad day, essentially. So you have to be careful. But, yeah, I mean, anything we are using agents, not just in the use cases I just mentioned, but I know any base... Any repeatable process, I think I see people using it on individual basis in terms of if they have... If they're constantly diving in weekly into a specific date to get out a certain report, then, yeah, that that process of using the data from an Excel spreadsheet and extrapolating it out and putting the other report can be easily automated using agents. So seeing it more and more on an individual basis and then also an enterprise basis and larger data sets to communicate with one another. Yeah. When it comes to agent to agent communication, do you factor in data security and data privacy into the equation? Yes. 100%. Absolutely. I think that's a critical one just like in the data quality. That's actually one of the bigger challenges in these agent development of different agents and especially the agent agent communication is all the data privacy and sharing requirements need to be worked out ahead of time. So we're confident that no data gets shared outside of environment that we're not comfortable with and whether it's using a data clean room or it's using on our own internal processes or systems. So, yeah, all that gets worked out ahead of time. Cool. Alright. Looking ahead, David, before we wrap up, where should marketing leaders begin if they wanna use AI for insights and analytics? What's the typical approach for folks who are just getting started here? I guess, well, one one part of that is it depends on sort of how large or sophisticated an organization you are. I think anywhere from small business with, you know, a few people can start to use it, to large organizations as well. So I think it's applicable to all, you know, sizes and types of businesses. But, you know, let's just take a small business, for example. I mean, I think you can, you can go in and, you know, everything we're talking about with synthetic personas, and audience segmentation and whatnot, like, you can do that with a regular AI model, you don't need a bespoke, LLM trained just around synthetic personas to do that. Now, I say that knowing that if you do do the larger, like the fully trained synthetic persona with a company and partner that does that, I think the confidence level goes up in terms of not just how the model works, the inputs and outputs of the model, the interaction and user experience with the model, that all improves. But you can create a synthetic persona, somewhere rudimentary by just prompting your regular ChatGPT or Gemini or Copilot. So you can build the prompt. And this is where just understanding different, you know, create the prompt to have all the attributes and descriptions that you want of your ideal consumer profile or consumer persona. So even a small business who understands their ideal customer profile really well can create a persona in the right prompting to mimic that persona. So then you have to try trial and error a little bit like, you know, are you going to just ask questions? Are you going to put stimuli in front of it to see what their response, a certain promotion or a piece of creative? And then as you get more sophisticated or larger business, and you have partners that can do that with you or for you, then the model gets a little bit more sophisticated. So I think it's just I would just advocate, try it. Go in, jump in. All the things we've talked about in terms of having a little bit of skepticism, being able to have your own way of validating whether the response and the outputs are you're confident in for what your use cases are or your unique business situation, whether it's past business performance or validate with real humans or gut check with others. But just have a way to say that you feel confident that you can check and say, well the output, I feel confident because of X or Y. So I think it's just having your own sort of governance approach to how you are going to say, I feel confident in the output. But I'd go in, start with the prompting, create agents like all the models allow you to create agents and create the right prompts and even more sophisticated approaches to that. So that's how you know, I and a lot of others are kinda learning. You can take, you know, classes out there as well to learn how to do it better. I've done a little bit of that. Those are valid as well. Come to marketing AI pulse events. Learn from from other experts in the field. Yeah, don't be afraid. I think it's just like the early days of digital. Know, I think everyone was like, do I go in quick? Do I create a new website? What do I do? How do I do it? Do I do it on my own? Do I hire a big, you know, digital consulting firm or a web agency to build my web presence or my app or whatever? And yes, but over time, everything gets much more efficient for smaller businesses or wider use cases. And now digital is just pervasive in everything that we do. So I think AI is going to become that a lot quicker and exponentially than even digital would get infused into everyday work and life. But AI is going to happen, it already is happening much quicker than I think digital adoption twenty, thirty years ago. So now AI is even if you don't go in and play around with a model and create agents and personas yourself, even in the background, like you do a Google search, you're going get the AI results. So even in the background of everything that you're doing, you search for a trip on TripAdvisor, you're going to get AI recommendations. So AI is being infused in almost every use case of everything that we do, whether it's a consumer, individual or business application. So it's in the background and in the forefront. So one, don't be afraid. And two, just start experimenting and using yourself and seeing what works for your situation and your use cases. And three, learn and validate and ask others, like at the Marketing AI Pulse events or like with other experts in your company, others that are using it, share notes, see what works, what doesn't. That's the great thing about this is like we're all learning at the same time. Awesome. Thank you so much and appreciate the shout out to the Marketing AI Pulse community. Yes. There's a lot of very knowledgeable people both online and in person events. So we'll certainly encourage our listeners to to participate. Well, David, thank you so much. This was so insightful. Thank you, Aby. Appreciate it. And to our listeners, thank you very much. Have a good one. Thanks for listening. Big thanks to Spark Novus and the Marketing AI Pulse community. You can join the community at marketingaipulse.com.
