The Agentic Shift in Marketing - Don Sklenka

0:00Don Sklenka:Hey, welcome to the CMO Whisper Show. I'm your host, Steve Olenski. Part marketing practitioner, part ad agency veteran, part journalist, I was a writer for Forbes for 10 years. I've had so many insightful conversations over the years with business leaders, to athletes, to celebrities, to of course, CMOs. The only difference now is instead of sharing those insights through written form, I'm doing it this way. My guest today is Don Sklenka, SVP of AI Optimization at Claritas, and one of the rare operators who's lived at the intersection of creative, media, and machine intelligence long before AI became a buzzword on every deck.

0:42Don Sklenka:I don't know if that means he's old. I don't think so. With nearly 25 years in digital marketing, Don built his foundation inside global agency powerhouses like Publis and Dentsuit, leading omni-channel strategies for Fortune 500 brands across retail, CPG, automotive, and healthcare. We're talking production, UX, media, dynamic creative optimization, performance. He's been in the trenches, not just those strategy rooms. Today at Claritas, Don is scaling patented AI solutions that turn data into decisions and campaigns into performance engines. Real-time optimization, predictive analytics, smarter creative, better outcomes, less noise, more signal.

1:28Don Sklenka:Finally, yes, he's a lifelong Cleveland sports fan, which tells you everything you need to know about his resilience and commitment to the long game. Don Slinka, welcome to the show. I feel like we could just stop the podcast there. You gave everything that everyone needed to hear about me, talked about the Cleveland sports fans. I think that's it. The podcast is over. We're good. We're good to go. Thanks for coming. Drive safely. That guy sounded great. I don't need to know anything more. He's your best guest ever, Steve. Yeah, sadly, Cleveland sports is an indication of loyalty and, like you said, resilience, because there's not much that we get to hang our hat on.

2:10But when we do, we really do.

2:13Don Sklenka:Oh, absolutely. Okay, I want to start with one of the topics that I'm interested in. That's agentic AI moving from demos, no shortage of those, to real deployment. So everybody's showing AI demos right now. But talk to us about who's actually using this in the real world and what does real deployment even look like? Yeah, I mean, I think, you know, first, like, you know, how do we define agentic and how is that affecting different businesses? So when we talk about agentic AI, it's you typing in a prompt and then having what is effectively a chatbot doing the work for you. So not just giving you a response, but actually going and executing that work.

2:54So part of the agentic process is setting up work streams and setting up APIs and integrations that when I want to buy a flight or something as simple as that, I can type in what I want to do, and then the agent will actually go and execute it. So I think what you're seeing more often than not today is the demo of that. So people or businesses or whatever you want to call that are demoing an agentic solution within a chat GPT or within a Gemini. And it looks great, But then the actual real life application is what is now the next level that we got to get to. So, you know, I think you're starting to see bits and pieces of that agentic process being integrated when we talk about advertising.

3:38So thinking about analysis, for instance. So analysis in a chat GBD is let me know what is your best ad, hit enter, and then it looks at data and gives you the answer. It's not really agentic. Whereas Agentic integrated into a dashboard would say, okay, we're going to literally look across every single data signal, look across every single creative. We are going to replicate a process of analysis, looking at, you know, that 3000 permutations. Then we're going to spit out not only the recommendations, but also what is the next best creative and use a generative AI solution to then produce that creative and then allow humans to approve it.

4:17Like that would be moving from, you know, an LLM to Agentic. so you know that that's where i think the the industry is moving towards more and more is away from the llms and more into the agentic it's just it's a bit of a you know it's a bit of an integration process because it's more than just simply oh i've got chat tbd let me add it i mean there's you know all the the hundreds and thousands of permutations that you've got to think about you know your system is different than the chat tbd system you know the different coding different language, QAing, you know, you don't want hallucinations, you know, so there's a lot that you have to do versus just simply a demo.

4:52So, you know, people will go on stage and show a demo, but, you know, getting that integrated is a big difference.

4:57Don Sklenka:Yeah. And you bring up a really good point about the evolution. Is that the right word from LLM to agentic? Yeah. I mean, I think there's, there's like the industry evolution and then like, how is AI evolving? You know, three years ago, no one really even heard agentic, you know, it was simply just a chatbot, now you're hearing agentic. So there's the industry evolution of simply AI. And then there's like spokes off of that. Like how are we as a Claratas company, you know, using that evolution to our vantage. So there's both of those things. So there's the industry evolution. And then there's like within each business, how's each business evolving.

5:31So that's a long winded response to that, to that comment.

5:34Don Sklenka:No, no, no. I, I know for a fact, or, you know, I talk to CMOs all the time. And my gut is telling me there's a, there's a reticence or reluctance or even an ignorance in the right context of what exactly is the difference and what should I be doing? Yeah, I use chat GPT. Isn't that enough? And what would you say to a CMO? It was like, my team uses chat. What else should I be doing? And I'm asking you to answer a very vague question, agnostic of industry or anything. But if a CMO is listening to this saying, what do I do when it comes to agentic what should be my first what should i start with i would start with what is your problem so what is it that you're trying to solve is it that your team is too slow is that you're paying too much is that you're not getting enough of a enough of an output you know those are three main main issues that i think ai is trying to solve for so speed efficiency and cost those are the first main main items and then agentic becomes you know do i need to actually integrate a solution into whatever that business is?

6:35Or can I continue to use external tools and get to the same point? So I'm not here saying that you've got to have a Gentic and a Gentic has got to be built in and integrated. I think you would have to weigh the downsides or upsides of both. And is it producing the output that you're expecting? So is your team using it efficiently? Are they taking the recommendations that you're getting into the chat and actually applying them? If the answer is yes, and that's positive. If the answer is no, then maybe an agentic approach where the issue is that humans are still slow. So you get all this output in the LLM, but if the humans are still slow to apply that output, then that's a bottleneck.

7:15So in this instance, you want to remove the bottleneck, which an agentic approach should remove bottlenecks that says, okay, this is the process. You apply it here. You analyze it here. You then make the adjustment here. And then a human in the loop is giving the approval versus an LLM is here's the analysis and the humans got to go. And what does that mean? And how does that, what do I do with that recommendation? And then I'll apply it. So I think it's a matter of, are you, are you seeing the bottlenecks? Is the bottleneck still a human in the loop? And then, you know, are you getting more efficient?

7:47Are you getting costs reduced? You know, are you getting that as output? So I'm not trying to use buzzwords just for the sake of it, but I really do think that that's where it's at. Like what problem are you trying to solve? And then, you know, are the humans in the loop still preventing that problem from being fully solved?

8:01Don Sklenka:Yeah, exactly. Listen, we could devote this whole time just to that discussion of agentic versus AI for sure. But I want to move on to something I know you're passionate about, and that's what you've called the real competitive battleground, which is identity, privacy, and measurement. And the real competitive background, as I think you've defined it. And I want to start with how is AI changing attribution and measurement compared to the old school way? It's basically, yeah, it's a good segue from where we just were at. So speed, efficiency, and cost. So what typically would take weeks to analyze hundreds of thousands of rows of data, that can now happen immediately within seconds.

8:38So, you know, that's the biggest output is simply, you know, getting the data, understanding what's working, what's not in seconds versus weeks. So that should, depending on your view, speed the cycle up or reduce significant cycles, depending on how you want to view that. And so who has now the fastest tool? But then the flip side of that is the most accurate tool. So AI should help you to become more accurate, meaning that in certain areas where you would have to have a more general approach to measurement, the size of the finite circle, if you will, should get smaller and smaller, which means that you're going to have more data to analyze.

9:16So the smaller circle, more data to analyze. And now AI should help you to analyze that quicker. And then ultimately, it should make you much more accurate in the approach. So I think that's part of it. And then you mentioned privacy. Privacy is also part of data. So who owns the data that you are analyzing? How much data is PII versus IP address versus user agent, things like that. So there are fine lines between what is PII and not. That will also impact. So the businesses or advertisers that own more of their data can then obviously do more analysis of the data that they own versus outside data providers may not own that data.

10:00So the PII line moves left or right because of that. So I think that's where AI is really coming into play is speed and efficiency, analyzing more data quickly or more accurate data in a short period of time.

10:16Don Sklenka:I'd be remiss if I didn't at least bring up the old school way, which of those of us like you and I who've done this for more than 30 seconds, there's always been that battle of attribution, right? Meaning, no, the social media team, we get credit for that sale. No, we get credit for that sale. Is it too much of an oversimplification to say AI is going to eliminate all of those battles? Oh, I definitely don't think it's going to eliminate the battles. It's probably going to make the lines blurred a bit more, which may actually increase some battles. I mean, exactly to your point, when you're working in the Google ecosystem and you're running campaigns that have Google Ads, I mean, it's obvious that they're going to weigh Google Ads more than you have an outside DSP.

10:59I mean, that's a very obvious element to it. Now, AI isn't necessarily going to solve that in particular, but if you're using AI as an independent measurement, it should be able to look across way more data signals and then provide, again, a more clear, independent view of measurement. But you're still going to have the same battles. You're still going to have people that are buying search ads versus meta ads versus display ads versus CTV ads. Everyone's going to try to say that their program is the best or their media campaign is the best. AI is going to make cases for everyone by looking at more data.

11:31But I think you're still going to have those battles for a while. I mean, the hope is that it does get a little bit more realistic. I'm not entirely sure AI is going to solve for it, but the idea that just look at the display space. The display space is very cheap inventory. You get a significant amount of impressions compared to like a CTV inventory. The idea is the same. You are likely not clicking on ads. 99 % of display ads are never clicked on anyways. CTV ads, you know, you're not really clicking on a CTV ad, a little bit of a controller. So they're kind of the same space. A CPM for CTV is$30.

12:05A CPM for display is$2,$3,$4.

12:07Don Sklenka:Right. So, you know, when you look at the display ad and the user scrolling on their article that they're reading and they scroll past the display ad, they've now viewed that display ad, similarly to viewing a CTV spot. But what often happens is that view in the display ad gets kind of thrown out. Well, you know, that display ad, I viewed it, but I didn't. And we can show that you converted on a view, but it gets thrown out for some reason. Whereas a CTV ad, it's everything is view. You view a CTV spot and then you go and convert. Well, that seems to be accepted. So, you know, I think it's, again, AI isn't necessarily going to solve for that.

12:40But the idea is if AI can help analyze more data, more signals, it can present a better and a fairer story for the different channels. That would be, you know, that's ideal. And that's what, you know, we are, you know, part of our solutions at Clartas is measurement related. You know, we are an independent measurement provider. We see that same story, but, you know, it's not really on us to make the decision. decision we present the data of which ai is helping but you know it's just it's just a matter of telling a more complete story than it is about you know removing the well social did this and i my claim that social did this or my claim that ctv did that right that will probably still exist for

13:17Don Sklenka:for at least a handful of years yeah i think it's too it's way too easy and quite honestly lazy in your analogy before of going well the display that's not working ctv's working i'll just put all my money in ctv well hold on right no the display ad may be working you just can't measure that per se yeah i mean it's i think i think it just like historically there's been a perception of okay well ctv is the new kid on the block it's a high cpm therefore it's gotta work so you almost like become pre-ordained i have nothing in ctv but anyways yeah pre-ordained as well you know it's worth that 30 cpm whereas i could get a hundred more impressions for the same dollar with with display and it's going to do the same thing you're going to see an ad and you're not going to click and that sort of thing so i think it's it's just because it's been around forever it kind of keeps getting pushed aside you know you have influencer marketing that's now probably ahead of ctv and that's the new everyone's going influencer but you know i think the idea is still the same like the cost of the media you know and the return you get on it everything is still up is still about the same so you know the return is a one percent conversion rate and then once you get to website it's another 1 % conversion rate to do what you need to do.

14:28So that conversion rate has kind of been tried and true regardless of channel that's seeing the ad.

14:33Don Sklenka:Right. Now CTV used to be just about reach, right? Now people are talking performance. That's right. What's happened? What changed? Yeah. So what's funny is that pre-COVID, there was this desire in the CTV space of using what is called a QR code. Oh my God, what's a QR code? That's craziness. We're never going to use a QR code and scan our phones watching an ad. Then COVID happened and it It became like the new thing because every dinner, every diner that you went to, you would have to use a QR code. So it became a natural thing with your phone. And so what benefited, interestingly enough, is the CTV area.

15:07So that when you're watching an ad, you now have a device or a mechanism of which you can then click through. You can then engage. You can then interact. So when you're watching a CTV ad, there's now direct interactions that it's not that they weren't possible pre-COVID. It's now that the end user is used to picking up their phone when they're on the couch and scanning that QR code. It just wasn't something that was top of mind or that's something that people wanted to do or used to doing. So I think that is part of it. That user experience in the CTV space has significantly changed. Therefore, in the CTV media, more and more advertisers are able to directly correlate not only reach of frequency, but also engagement.

15:48So that's part of it. And then the other part of it, again, is, you know, measurement companies, you know, being like a Claretas, you know, I have to pitch us a little bit on this, but, you know, measurement companies like us, we are able to understand the IP address or the device that the user is on when they're watching that CTV spot. And then, you know, if we have the end result pixeled, for instance, we're able to understand that this person not only saw the ad on CTV device, but then on their phone, because they're within the same network and converted, we can apply that conversion. So we are helping the CTV industry, which previously, you know, you didn't understand what happened after an ad was viewed.

16:27Now we're connecting dots to say, okay, after an ad is viewed, the user did X, Y, and Z. So that's improving the performance of that ad. And then, you know, that applies to all channels. But, you know, that's, I think between the two of those things, that's what has really helped the CTV space is interaction, user experience. and then the ability to measure beyond just reach and frequency. Those have really worked in parallel to each other.

16:51Don Sklenka:What about live events like sports? What should brands almost rethink when it comes to sports and CTV? I think it's that engagement factor. So you have the ability to take signals from the game itself or from the match or whatever it is that you're watching. Take those signals and have it impact the ad that you are seeing. So on our end, we take a variety of signals. And again, rather than a human trying to make a decision in a very linear decision tree, it is Monday at 9 a.m. Served as creative. The idea is actually letting AI identify hundreds of thousands of permutations of signals. So you are looking at a device in a certain market and you're watching a certain type of TV show and the context is ABC.

17:33Therefore, change the creative to XYZ. And then think about that over 40 to 50 million impressions. There's a significant amount of permutations. mutations. In the old world, we used that word earlier, in the old world, it wasn't possible to calculate and then have a decision with all of those impressions, even dynamic creative or DCO, it was a preordained decision. Whereas in the AI space, now you're able to do that. You're able to truly have a decision for every single impression based on hundreds and hundreds of different data signals. And then the output is going to be unique to that individual and it'll change.

18:06You may see one ad during an Eagles game, and it's going to be a completely different ad and experience when you're watching the Grammys. That's something that you just could not keep up with in the old world, whether that was a human decision or even dynamic creative. Now with AI, that is something that is now possible and we're able to do.

18:24Don Sklenka:Yeah. Let's put your future hat on when it comes to AI and creative. So for the most part, when we are talking about AI and creative, we are leveraging someone else's technology. So we're leveraging open AI, you're leveraging Gemini, you're leveraging Claude, something like that. So for the most part, 99 % of advertisers or agencies or companies that are talking about generative AI are using someone else's models. So we'll start there. It's very few where you are like there's companies that are truly building their own generative AI models when it comes to that. We've got to be inefficient. So starting at that point, if you looked at where the generative AI models were six months ago, or you look at where they were two years ago, the evolution within those models has been unbelievable, unbelievable.

19:09Even the biggest update was, I'm sure folks that are listening to this know what I'm talking about, where you would ask even ChatGPD or OpenAI to create an ad and the copy would never be in English. It would be like a garbled mess of things that look like they could be words, but they're not really words. Now you produce the same prompt or you type in the same prompt and the production is night and day. It's literally not only you're getting English words, but the words are within a font that is the same exact style guide that the initial brand has. And that was a six-month period of evolution.

19:44So the hard part about answering the question is that it goes back to perception. So the perception is hard to change when a brand engaged with generative AI a year ago, and you saw the output and it was a disaster. And so therefore, everything going forward is going to be the same disaster where it's not really true, it's an evolution. So the hard part is, is changing that perception. So we still on the Claretas side suffer, even though we're not generating ads in real time, we still suffer from that initial perception where generative AI was hallucinating and creating these ridiculous outputs.

20:20And so then all of a sudden legal and compliance at every brand, like everyone stood up and said, okay, if you have the word AI in your solution, we've got to go through this rigorous testing protocol and understanding all the output, even though it has nothing to do with gender. So like one, one nugget of a bad experience has created this perception in the industry. That's really hard to overcome. The reality is that it's getting significantly better. So like, again, if you if you put in a prompt that said, you know, generate an ad and give a style guide, some previous previous examples, it's going to be pretty close to what you're producing today, maybe even with some tweaks.

20:56Again, another long-winded answer to your question, where are we going? I think the models are going to be at a point here within the next six months or a year where it's going to be very, very, very difficult to tell the difference between an AI-produced flat ad. I'll keep it flat for now. So a static ads that could be a Facebook, meta, TikTok, Instagram display, whatever that may be. Tell the difference between a flat ad and that's human produced and a flat ad that's generative AI produced. So that's that that is happening now. I think again, within six to 12 months, the hard part is that perception.

21:34Will the industry start following and say, OK, yeah, we understand that it wasn't what it was at two years ago. Can you put the parameters to make sure that it doesn't hallucinate that sort of thing? So I think that's the next step is changing that perception. You're going to see, I think a lot of the output of the Superbowl ads that's video based. There's going to be a lot of AI uses in that. And I think the Superbowl always, it's the king of advertising, the pinnacle at, and from the Superbowl, then start, everyone starts like changing perception. So I think, you know, we'll see what happens after the Superbowl, but the idea that you're going to see more and more AI based ads is going to start loosening the brand and industry a bit when it comes to the perception of generative AI, which is then only going to increase the speed of which now more advertising is going to use it.