C (2:20)

say because AI is all the rage.

A (2:22)

Maybe I'll start with that. So, you know, Emma, my. My wife Janie works in the industry as well. And one time we were having dinner and I had all these conversations with clients about geo and I just felt like there were some things that like teams were focusing on the wrong things or, you know, there were all these areas where there was a lack of clarity. And then as we started talking, it just went into like tangents of like, oh my God, this is going to open up this and it's going to lead to that. I started kind of like writing notes on my phone of we probably need to talk about this. And originally I was going to do a LinkedIn post and it just kind of snowballed into like 12 pages and I was like, I need another way to do this. So I got some people together, some very smart people to think through this and how we wanted to say it. But yeah, so overall I'd say there's kind of a few key themes that I wanted to make sure we got across. One was just what is changing? I think generally everyone knows that things are changing and I think that there's enough chatter now that everyone is focused on it. But what exactly does that change mean? What could we realistically impact? What can't we impact today and what's going to probably move? I think at this point we can see kind of like what is the direction of travel. So understanding what is that direction of travel going to be and how we prepare for it. But then also what is the next phase and the next phase of this? Because I think as exciting as where we are today in terms of generative search, it still feels like there's so many areas where we're just scratching the surface and if we start to think about advertising platforms and agentic shopping and what this means for personalization, there's just so much, so many things that brands need to be thinking about today to stay ahead in 12, 24 months. So it felt like the right time to talk about it and it was honestly a lot of fun to write. I still feel like I'm having these conversations at least once every week with someone about what's going on in the space.

C (3:56)

I love it. On your first kind of theme that you mentioned, and that's the what is changing and what does it really mean? I'm going to flip that back to you and ask you to tell me in your opinion, like what is changing and what does that mean in terms of the AI search revolution? We use that term. What does it mean to you?

A (4:12)

Yeah, I think there's some more obvious things and some less obvious things. So like from a more obvious standpoint, right, we're seeding some of our decision to a machine in areas where we don't feel like we have all the information. So we just use the example with like golfing. Right. If there's something that I don't, I don't feel like I know a ton about golf, maybe normally I would ask a friend for a recommendation. Now I have this option to ask something that has this massive database of information to help me answer that. And there are going to be purchase decisions. We tie this to a commerce world where one the of when I'm doing discovery about products that's going to be really relevant information. Maybe I would have gone to Amazon reviews previously, maybe I would have asked friends before. But now I have a new option and overwhelmingly consumers who are trying that option, like that option, that's something that we're seeing. This behavior is increasing, not decreasing. Retailers that don't have this behavior are picking up options for it because they're seeing the consumers have an interest. So what we're seeing in certain categories is consumers just being okay with seeding some of that decision making process. I'll also say with this is like browse behavior is changing rapidly and in some ways going away. There are some situations where historically the path is I am looking for a product, I want to see a few options, browse through whether this is a purchase decision or a discovery decision, what my options are, click into something, learn about it, maybe I pop back out, look at something else. That funnel is different now. Now it's I ask a question, the machine gives me an answer and I either say I like that answer, I trust that answer, or I ask a different question or I ask the question again. So it's a little bit of a different path that doesn't have as much browsing and puts a little more trust in the broader machine to answer it for me.

C (5:52)

In the paper we have this stat from research that says that, you know, consumers are kind of trusting AI search results at like unprecedented rates. I'm curious, do you have any thoughts and opinions on like why are people so willing to trust that? Even though, I mean I know in my, I get a lot of hallucinations and so that has put me on the. I don't trust you entirely. I like, you got me somewhere. But just curious, why are people trusting it the way that they are?

A (6:15)

Yeah, like, I think that's a healthy amount of skepticism. And I know like even our own teams, when we use AI from our acceleration team, we give that direction, right? You need healthy skepticism in this world right now. But a lot of these models, right, if you're on Google, you're getting the source information. If you're on Amazon, they know some things about your purchase history. There's a little bit of a belief there. There's a, it's a very intelligent decision. Intelligence is right there in the name to the degree that people feel like, okay, I feel like there's enough backend and information here that I'm getting a pretty good recommendation. And the fact that we can click into sources and some of that allows us to get a little bit more confident there. The models are also getting a lot better. And if you look at some of like the modeling over time and hallucinations, they're improving dramatically.

B (6:54)

Okay, now we spent a good bit of time talking about how AI has changed the consumer experience and shopping behavior. Let's get into what this means for brands and what brands need to be doing because, hey, that's probably why you clicked on this episode. So I know that brands need to optimize for geo or generative engine optimization, but what does that actually mean? How does that differ from SEO? Is one more important than the other?

C (7:16)

This is overwhelming.

A (7:17)

Let's start on why this is so scary for brands. Maybe so it's like if I'm a brand, an SEO world is very clear on how I can influence that, right? There's keywords and I can identify what those keywords are. I can target those keywords either in paid or non paid environments and I can make the change. And then it just, I can see in real time, look, I added this keyword and now I'm more relevant on that keyword. The measurement exists, the ad platforms exist, the data exists. In the LLM world, none of that exists. Right. There's very limited areas in that space. So we talk about the black box. I think we use that phrasing in the paper. It's like, yeah, it does feel like a black box. And yet if you're a brand, I go on and I search for the know the best manual shaver and it's bringing up a competitor of mine. That's pretty scary for a brand because you're like, what do I even do about this? So I understand the concern with brands and because it's pulling from so many different locations, it can be really overwhelming. We're not, if we look back, you know, five, six years ago, all these brands are having conversations about separating out our commerce content strategy from our Google content strategy and having like these centers of excellence. Well now it's all kind of getting aligned again, right? That these tools are pulling from a lot of different locations. So it also kind of makes brands rethink how they're going to approach their own content strategy. And I think that can be very overwhelming. Kicking back into your original question, which I think is a good like tail off of that, which is so what actually are we talking about with generative search? Like what is happening here? If we think about the major providers, right? OpenAI with ChatGPT or Google, whether you're on a Gemini app or you're on their AI assisted search, what we're seeing here is a variety of a few different activities. There's a large language model which is taking tons, tons, trillions of parameters together and bringing them into a single location to basically say how is this enabled search working? Right. So if we look at OpenAI in a ChatGPT model or if we look at Google with either a Gemini app or like a Google assisted AI assisted search, what you're getting here is a combination of a few different things. You're getting a large language model which is trained on just a mass quantity of data and information, historical information. It's trained to avoid certain things, focus on certain things, to think as closely as possible to the human brain in terms of how it identifies and segments information, shares information. There's also a process of rag. There's also a process of going through web browsing, right? We're pulling information from live things that are happening across the web there. There's different values of recency and how much we're going to target recent information, user generated content. And something that I think is kind of a hidden piece within this area too is not every one of these models is going to be thinking the same way either. Some of them are going to weigh certain things a little heavier than others. So the challenge is how do we influence those things? And I think the reality is the pre trained historical LLM models are very difficult to influence. That shouldn't be the focus. There's some opportunities for web browsing and if we look at something like Amazon's Rufus, we, we can get some insight from their Cosmo framework to understand how they're informing Rufus and reverse engineer that piece. So we Almost want to start thinking about these different models as individual focus areas and targeting them accordingly to understand exactly how we're going to win in that space.

C (10:27)

Kind of sounds very similar to how like if you are just optimizing your pdps like you always have been, you need different things for different retailers and different retailers require different kind of data inputs and whatnot to go into the PDPs. And it's been tough historically to really optimize across every single retailer ever, which I know there's a lot of technology that helps with that and we do a lot of that, but it kind of, it reminds me of that sort of thing where it's whether it's retailer by retailer or model by model, it's understanding it from the core and then building from there.

A (10:57)

Yeah. And we're probably in the nascency of that. Right. Like these companies are still finding ways to utilize data. If you look at a partnership like OpenAI and Walmart, great example. Right. Where that was actually something that when we were writing this paper pre that announcement, we kind of said we assume this is going to happen. We assume that OpenAI is going to partner with retailers because that gives them access to different data they can use as context to serve an answer. That data isn't integrated yet. It's not something that is like aligned. But as more data becomes available, we're going to see each one of these models probably continue to deviate, have specialties, have focuses and utilize the data they have to gain a competitive advantage.

C (11:35)

Let's get specific.

B (11:37)

Mike mentioned that some models are harder to optimize for versus others, but when it comes to say, Amazon's Rufus, we have a better understanding of the framework that informs what Rufus outputs. With that, I asked him to give us a step by step process for brands who would like to start optimizing for a model like Rufus and they don't know where to start.

A (11:53)

I live in the commerce world, so Amazon Rufus is easiest for me and I understand the most there. So if I'm thinking about that model, for example, you definitely can't go wrong with consistency. And I think that having a tight PIM strategy, having tight product experience management across all of your retail platforms has never been more important. These models are looking across different sources and if we're giving them a consistent answer, especially across product detail pages, that's going to hold some weight. It's going to tell us, hey, I can trust this answer more because I'm seeing it in multiple places. It's very validating. So we Want to make sure we have some consistency. Certainly the way we look at a model like Cosmo, which again, is kind of what informs Rufus, it's almost, I would think of it almost like question and answer focused more so than keyword focused relative to regular SEO. So there's a little bit more weight on trying to answer what the consumer is potentially looking for. And that matches the conversational search environment in which they're searching. Right. You're usually searching with some sort of a question that you're looking to be answered. So looking at things like, what are those Q and A environment situations within my page? What types of typical FAQs would consumers be having about my product? And also looking at contextual relations. So things like, what is this product used with? What is this product used for? What else would this product be? What's the environment of this product to be used? Those types of relations also feed in and we think about those things as well. So there's many different elements of that. As I'm explaining it, you're probably like, how on earth would I do that at scale? And to be honest, we had to build an agent to do it. It's like Thanos, I did build the stones to destroy the stones. Right. I had to build with the AI to solve for the AI. So we essentially, at scale, now have a tool that allows us to look at all of these different relational aspects, look at the product page and say, this is where the gaps exist. And we're using that to fuel and make recommendations around where we feel like those things, where there's opportunity to potentially make ourselves more relevant. For the Rufus engine.

C (13:47)

I'm curious if we're looking at, like, organizational structure of a brand and their. We'll just call it like their commerce function, if they have one. Who do you think should be, like, responsible for this type of work? Should this go to a traditional content team? Should brands be building out people who are specialized in this and they are just like AI product specialists? How would you recommend a brand approach from that regard? Like, who do you assign to do this As a first step?

A (14:16)

Yeah. So in the short term, I think it makes sense to have somebody upskill on this and really understand it. But in the long term. Right. I think search is going to be searching. The best experience for me as a consumer is not having a Rufus agent down here and a search bar up here. If I'm looking for paper towels, I probably don't need Rufus. But if I'm looking for a golf club, I might. So it's like we probably want and we will probably see traditional and let's call it generative search options, kind of like blend together over time. You're already starting to see the major providers in the AI world create, you know, browser interfaces and they're building their own browsers, right? They're building the ability for us to search and integrate AI and traditional search together. So I think in the short term it makes sense to have someone upskill and really understand this. But in the long term, the person who is managing content, the person who is thinking about search, the person who is targeting advertising, is probably going to need to be aware of both of these things. Because if we were to totally scrap SEO and just focus on geo, that would be a massive mistake and would set us on the wrong course as well. So to answer your question, I think that there's a short term need to really go deep, but I think longer term this will align into just a search strategy.

B (15:25)

How do you, even if you're a

C (15:26)

brand and you're actively working on optimizing for geo, how do you actually measure the success of that? Is there a way to measure it or do you just do it and hope that as people adopt this type of search behavior going forward in greater quantities that you sell more products? How do you measure anything?

A (15:43)

Yeah, it's a really good question. And I think that this is an area where brands really struggle today because OpenAI and Amazon and Google are not giving that prompt information, right? They're not sharing with you what are people prompting. They're starting to open up advertising capabilities and that will give us some very clear views into what's performing, what's not from an advertising standpoint. So very excited for that. And we're already doing some testing there across a few different areas. But that I think will illuminate a lot of this. In the meantime, one thing that some brands are taking advantage of would be like mass prompting. So going across and saying, I will send X number of prompts across these different models, reviewing what those sources are that come back and using that as like a sort of a guide to say generally where a source is coming from, what types of responses are coming back. There are some limitations to that, right? You don't get full context, you're kind of assuming some of the prompts, but it's within the environment that exists today. That's one of the limitations is, you know, that's kind of what's available. The other thing that I would say is total views, total conversions, total clicks, those things always matter, right? Like shares, sales views. We can still look at the high level metrics and the high level KPIs to determine if these new strategies are making sense. And if that means that this kind of falls a little bit into the 50% of my marketing works, I don't know which half. While we get some of these advertising capabilities on that, that might be a necessary evil. But I don't think brands should stray away from looking at some of Those higher level KPIs is a way to see if what we're doing is working or not working.

B (17:08)

All right, let's start talking about the future. Where is all of this going? Can we say maybe that's too broad of a question here. I can narrow it down a bit. There's a line in the white paper that the scalability of AI will soon enable unprecedented mass personalization, forever redefining the consumer experience. So I of course asked Mike to tell me what that means.

A (17:28)

Means I think AI is going to bring forward this, this era personalization. I think more people are talking about this now. I think when this was initially written it was a little more forward looking. And now I think this idea is catching on. If I'm an advertiser, right, My ultimate goal is to build a campaign that meets your specific needs. Like I want to build a campaign for Emma and I want to build a campaign for Mike. But there's like all these limiting factors that get in the way. So like originally it was like, you know, in a era of like TV advertising, there's really not a great way for me to segment out unless I'm like focusing on who watches, which shows or which channels. Then we got into like digital media and all of a sudden you kind of can tie back to a user and we could target users more directly with advertising. And it got a little bit more surgical in that world. But it was a lot more difficult to understand, like how those impressions then led to someone actually buying. So you wound up with some metrics that were a little less straightforward. Then you get something like retail media coming along. And now, okay, cool, I can see what someone buys. And that allows me to be a little more specific and a little more targeted so I can target those consumers based on the purchase behaviors. But there were still like some limiting factors around how do I segment out those audiences, how do I align those audiences and how do I build creative at scale that differs for you or for me? Because if I have, remind me, Emma, what kind, what breed dog you have?

C (18:42)

I have a husky and a chihuahua. So take Your pick.

A (18:44)

Okay. You got a husky and a chihuahua. Awesome. I have a great pair of knees, right? Like if I'm showing you an ad with a dog food bag and that ad has a chihuahua on it, and the one that I see has a Great Pyrenees on it, will probably be more likely to buy that because we'll be like, holy cow, no one's ever shown me a Great Pyrenees before. So how do we get to that? Right. That that would be like the dream state to be able to really connect with the user. And AI is kind of unlocking that at a really crucial time when clean rooms are also opening up some of this data. Right. So it's like with a clean room, I can start to bring in some external data. Let's say I'm that dog food company and I have from my vet database what dog breed. These different. I can align through like a pseudonymized ID dog breed to user because I can link up an email address or a phone number or something like that. Right. In a very privacy protected way. I can start to build out audiences around that. Now the challenge becomes how do I build creative at scale that can do those things? And obviously AI creates some opportunities for us to scale creative, to try different messages, to give consumers different experiences through that. And we're starting to get to a point, we're working on this now where it's like we're building out these campaigns by audience and we're actually showing those audiences totally different experiences, or we're building four different similar experiences and testing and learning with an audience. We're showing them different experiences, creative experiences, taglines, to see what is actually going to drive you to convert. And seeing that on the back end, to be able to measure it, which is really cool and exciting and something that I don't think was really plausible or doable at that scale before, you

B (20:15)

know, I was talking a lot about some familiar themes like personalized reach and messaging at scale, audience focused advertising. Kind of reminding me of what we've been talking about this whole year via our other white paper, the Big Shift.

A (20:29)

Yeah, Huge fan of the Big Shift, I think. I mean, this to me is a very clear indicator of the big shift. What we're seeing here is the ability to be more audience led with our targeting. It comes from better data, better data analysis. It comes from being able to personalize experiences. But it's like, what would I rather do? Would I rather have a commerce strategy and a media strategy and a TV strategy, or would I rather have an audience strategy? And I follow that audience wherever they go. I think we can all agree that it'd be more impactful to take that consumer, understand what makes them tick and serve them an audience up and down the funnel. Take the learnings from, you know, what's actually driving conversion of different media experiences. If we can see in real time a sale links to a different advertising creative, that's a learning for that audience that I can carry back up the funnel. I can reverse the funnel and say, okay, well when I target that same person on Google, I should probably give them the same experience because this is working really well. And I think we'll start to see that permeate on and offline. I think we'll start to see just more ability for us to focus on the path to purchase and less limitation on having to lock ourselves into channel dynamics and stuff like that. So 100%, this is not only in line with that, but I think a very clear indicator that that is where we are moving as a, as an industry.

C (21:46)

My last question for you is, is there anything you can tell us about what we're currently cooking up when it comes to AI at Flywheel? Anything you can hint at or things that already exist that maybe people don't know about?

A (21:59)

We could do a whole podcast on this. There's so many fun things we're doing with AI right now. I would say like, you know, we've essentially built an agentic structure across not just Flywheel but all of our Omnicom agencies. So all of our teams have access to every major LLM at this point, I think it's over 40 different LLMs. We're using it for generative content, we're using it to create agents to share those agents. We can even share chats now, which is like wild. You can have like AI enabled chats with multiple people. That's kind of like just wider how we're using the tool as a company. But a couple things that I'm pretty excited that we're launching for manufacturers that we work with, we're implementing an AI orchestration across our platform. So you'll be able to call the data that's in our Flywheel commerce cloud platform in a conversational way. You can merge media and retail insights and like find really cool aligned insights. You can create agents that you can create repeatable tasks as can your team. Right. So your team is now powered with this technology. You as a manufacturer power with this technology. So it's something we're really excited about and it's really innovative. And I'll add too you know, when we think about our roadmap, one of the key things our team does as well is not just upskilling our team and teaching them how to use these tools, but it's also working with our product team to make sure that we're implementing AI enabled tools into the roadmap. And that's yielded AI bid management, AI bid planning. A lot of the tools that we're bringing forward that we're using for content now are AI enabled. So it's really pervasive. I mean, it sounds hokey, but it's like there is nothing that's coming forward from a transformational side for us that isn't AI enabled in some way right now. And now it's just how do we get every single person using it as, as effectively as possible? That's. That's my job.

C (23:36)

Okay. I do actually, I have one final question for you. What's something that's in your cart that you just won't actually purchase and why?

A (23:43)

Yeah, I mean, right now, honestly, it's just like mountains of baby clothes. I. I have a one year old and we have a little girl and dressing her up is way more fun than I expected. And so like for Christmas, I like, it's like Christmas is one day and maybe there's like three or four days that you see family and stuff, but it's like she's gonna have like 15 different outfits. I don't. We're just gonna be picking between them.

C (24:04)

Oh yeah. I can imagine Janie being all over the fits for this child and probably the best child, best dressed child and maybe the country.

A (24:15)

She is, she is quite the wardrobe. I won't take credit for that, but I will say for the holidays and so stuff, I go all out. That is super fun.

B (24:25)

And that wraps up another episode of Ecommerce Collective podcast. Thank you to Mike for his insights. And if you want to learn more about this aforementioned white paper from Powerless to proactive, how brands can own an AI search revolution, I've linked it for you in the episode description. You should check it out as a fast follow. You'll walk away with actionable strategies and even more context into what all is changing in this crazy world of commerce. All right, I've been your host, Emma Irwin, and I will see you next time.