#448 - E-Commerce Will Never Be the Same: Andrew Bell Breaks Down Amazon AI

#448 - E-Commerce Will Never Be the Same: Andrew Bell Breaks Down Amazon AI Kevin King: Welcome to episode 448 of the AM-PM Podcast. This week I'm speaking with Andrew Bell. Andrew has become one of the thought leaders when it comes to AI and e-commerce. He studied the patents. He studied the Amazon science papers. He's written a couple of very long and detailed and very explanatory things about how Rufus and Cosmo and Agents and And all the stuff that's happening in AI are going to affect e-commerce. And today we're going to geek out on that and talk about what you need to be paying attention to, what's happening and where we're going. So enjoy this episode with Andrew Bell. Unknown Speaker: Welcome to the AM-PM Podcast. Welcome to the AM-PM Podcast, where we explore opportunities in e-commerce. We dream big and we discover what's working right now. Plus, this is the podcast where money never sleeps. Working around the clock in the AM and the PM. Are you ready for today's episode? I said, are you ready? Let's do this. Here's your host, Kevin King. Kevin King: Andrew Bell, welcome to the AM PM podcast. The trifecta is now complete. Because I've had Juana on, I've had Danny on, and now I've had you. And you came to prominence, to me at least. I mean, you've been around for a while. We'll get into that in a second. But because of the work the three of you did with dissecting, cross-examining, and laying out in layman's terms what the hell's going on with AI and Amazon. So welcome to the AM PM Podcast. Speaker 2: Hey, thanks for having me, Kevin. Kevin King: I appreciate you being here. So what is your story? Who the hell are you? Speaker 2: Well, I have a very unconventional background. I actually studied ancient Greek and preaching. I didn't even study marketing or anything like that. Time in my life, I came home, lived with my mom and spent all my hours every day basically in Barnes and Noble, reading everything off the shelf from every business title to Leo Tolstoy's, all his stuff like War and Peace and Anna Karenina, things like that. Kevin King: You're the guy sitting in the chair at the end of the aisle with a coffee reading the book all day long. Speaker 2: Oh, I'm that guy. I'm the guy with 20 books there reading the thesis of every book and then deciding from there which one I want to buy. In fact, I've had people ask me all the time at Barnes & Noble, they go, hey, are you writing a book or something? I said, oh no, I'm just doing this for fun. They looked at me like I was crazy and walked away very quickly. Kevin King: Was this before you were married with your wife or? Speaker 2: Yes, this is before I was, it's kind of in between. So when I had come home, you know, I spent six months reading everything off the shelf. And then I spent a lot of time with my best friend who was going to die soon, my grandfather. He was actually before the head contract negotiator for United Airlines. Someone I miss. Yeah. Well, funny enough, I actually was on the podcast with Joe and Max. That was kind of the first time I was kind of like, I guess exposed to the Amazon world and like a lot of other thought leaders. But I told them about my custom GBT I built that I've been working on like on my AI grandpa. So basically uploading all his stories from Ireland that he had told me, written himself. Currently, I'm looking for voice recordings, hopefully create a good enough sample to replicate his voice. Plenty of pictures of him. I think within the next five years, I could have a living avatar that, you know, I can talk to online. That's cool. I miss him dearly. I spent time with him and then I met my wife a number of months later. She was also into business and everything. She actually majored in marketing, began her own business. It was actually featured in Forbes magazine when she was in high school for a business she created. She was adopted from Ethiopia when she was nine years old. She built a business of a unique cookie mix. That she was then able to sell to a lot of people and then all the profit actually went back to her orphanage in Ethiopia to help that, like 100% of all the profit. Kevin King: That's cool. Speaker 2: So that was, yeah, it was really cool. Recently she was in Newsweek for like a lot of her momhood stuff and just being able to relate with a whole group of moms. She knows how to reinvent herself and you know, my story doesn't make sense without her story, you know, and I think that's a beautiful thing to have. So that happened and then a company needed to get on Amazon. Of course, like most companies, they didn't really take it that seriously. Kevin King: Is this the company you were working for at the time? Speaker 2: No, yeah, at the time, right. I was able to grow that brand from $0 in sales per year to $7 million per year. It was a luxury home decor brand called Touch of Class. Maybe I don't know if you ever got the catalog. Kevin King: Yeah, they do a catalog. Yeah, I'm very familiar. I actually bought stuff from Touch of Class. Speaker 2: So the first thing I did is, you know, I convinced him we should do Helium Tap. So you were my first teacher, Kevin. Kevin King: Oh, was I? Speaker 2: Yeah, you were my first teacher. Freedom ticket. Yeah. Kevin King: Who's this guy? Speaker 2: I love this guy. He's good, you know. Kevin King: I'm wearing a blue shirt in honor of it today. Speaker 2: Yeah, that's right. It's like, oh, I'm one of the guys from the books, you know, the books I was reading, Thought Leaders and such. So it's pretty cool. Yeah, definitely. Kevin King: So are you still working for a touch of class now? Speaker 2: No, I'm not currently working. I'm not working for them now. Kevin King: So just to be clear, they had their direct marketing business, the catalogs, and they had their own website, but they were biting the dust on Amazon. So you came in and took them from nothing on Amazon to seven million a year on Amazon. Speaker 2: Yeah, seven million per year, still at 34%. Kevin King: You did that by learning how to sell on Amazon on Freedom Ticket and just kind of winging it and then put things together. Okay, cool. Speaker 2: I'm very much a self-learner. So I knew like I had to, here's the thing that we were the, if you've ever looked at the catalog, we're always like the highest priced item on everything. So I knew when I went into the Amazon space, It wasn't enough to be good at everything. I had to be the best in my field with the people that I was competing against. So I'm completely self-taught. I'm talking from pick, packing, shipping these products from the warehouse with very limited square feet to eventually 50,000 square feet of space allocated to our FBA program by the end. I remember starting a brand store for us late 2020. That's, you know, brand store page. That alone made $1 million in 2023. I started doing Amazon Post when it first started around late 2020 and I remember going crazy. I posted like 200 in the first month because we were like on like the beta test of it. So I was getting like tens of millions of impressions over my four years. Then I posted around 2,000 unique Amazon Posts. I remember writing up the whole strategy for the Amazon Marketplace, managing the whole platform, 4,000 unique parent ASINs. Ranging from hundreds of bedding patterns to thousands of metal wall sculptures to hundreds of Tiffany style lamps and then Victorian style mirrors. Kevin King: Was this FBA or FBM? Speaker 2: Both. Kevin King: Both? Okay. Speaker 2: Both. We had about 300 best-selling products on FBA too. We didn't have the bandwidth, but one thing I did is I just coordinated with every team. I was a one-man show, but at the same time, I was able to coordinate with every team and have them be able to do stuff, but I had to prove it myself that it worked, so I learned a little bit of everything. I became a Canva expert very quickly, and I'd use Canva every single day. I was on the Amazon mobile app every day, desktop every day for just an hour, just scrolling around to see what's happening, what ads and such are looking at. So yeah, it's been pretty crazy. But one thing I did more than anything, and this kind of relates to my story, is keyword research. I would spend hours after work just studying Helium 10, finding everything I could do on that tool. I'm probably the guy you guys maybe got mad at because I had so many Cerebro calls. I'm talking thousands per product. I spent hours upon hours on keyword research and optimization. And so part of the reason I got us from zero to seven million per year in under four years is I had no other option, like I said, to be the best optimizer I could possibly be. So Helium 10 before AI, I wrote narrative style, long form bullet points, saturated with all this Helium 10 keyword research. And like I said, thousands and thousands of calls to Cerebro. And then so what you see in my AI generated text is the result of what I wrote extensively. So I was very good at writing. I optimized 1,500 product detail pages with the same exact rigor that I had. Kevin King: By hand. Annually. Speaker 2: Manually. By hand. I wrote it out in scribbles at the time because you didn't have Listing Builder at the time. But now, you know, obviously you have Listing Builder now. So I'd use scribbles, upload all the keywords in there that I possibly could, you know, organize it with Frankenstein, etc. And I did everything. You know, I did 20 Amazon Lives, 250 videos, 100 distinct campaigns, usually running 300 all simultaneously. While running everything else and my mind was racing and running, I had everything up in my mind as if I could see everything all at once. But every day, I also felt like I was behind. You probably know this because you know as well as I do that the more you come to know Amazon, what, the less you feel like you know Amazon. Because there's just so much to it, you know. So, you know, then going into 2022, about October time, Canva came out with MagicWrite. This was before actually in November when, you know, ChatGPT really made a splash. And then when, you know, GBT-3 and 3.5 came out, I would, you know, test the limits of it and in November and then. Interestingly enough, my wife actually showed me ChatGPT. I didn't even know anything about it. She's a content creator. She's on the cutting edge of everything. If you want someone who's ahead of me, just go to my wife. I'd show her a video. She's like, yeah, that was so like a couple of months ago. I'm like, oh, whoops. Sorry. But at that point, I decided to join LinkedIn, this was late 2023 that I decided actually to do that, made connections with lots of people, of course, showcasing all my skills. And I became known for my GBTs actually outside of LinkedIn first, and that's all organic from the GBT store from people searching. Yeah, and now I have the top GBTs for Amazon, even poetry. I think that comes from a lot of my reading that I did before with novels and like I said, Leo Tolstoy, like older literature like that. And then a Pinterest GBT is the top one. And then I even coordinated with all the OBGYNs that my wife saw, because we have twin girls, and I created a pregnancy GBT. That's actually the number one pregnancy GBT, because I wanted to show people you could create anything you wanted with this. I think that one has like, Over 100 review ratings. It's a 4.8 out of 5 stars. I think it's my best rated GPT, which is kind of funny. I guess that just proves I'm pretty good at prompting for anything. You know what I mean? Kevin King: How are you monetizing those? Are you just providing those out for free? Speaker 2: It's a hobby. It's fun. It's just the act of doing it, you know, the creative play, the whole idea of like creativity on fire. And I just enjoy doing it. But, you know, yeah, it's led to opportunities where, you know, some people want to hire me. You're familiar with Poe.com. I've been on Poe ever since it was a thing. I actually have the top, one of the most popular bots. I think it's the number five on the entire thing. And it's a writing, like GBT, if you will. It's like creative writing, writing your own style. Kevin King: For people that don't understand Poe, tell people what Poe is for the people that haven't heard of that, just so they get an idea. Speaker 2: Yeah, Poe is really cool because you can test. It literally puts on every single model. They just recently released the image generation API model, which I'm working on apps right now with that, creating a very comprehensive app. That lets people plug in the prompt and create whatever prompt they want, and then it actually will generate. I've been waiting for this API forever. Of course, everybody has, but Poe has integrated it very quickly. But they probably have, I think, 100 different models that you can play with on Poe from different people, whether it's Google, Meta, ChatGPT, OpenAI, Anthropic. Kevin King: So it's all the LLMs, all the top LLMs and some niche LLMs all in one place where you can just toggle back and forth between them or have it run stuff simultaneously in some of them. I just want to explain what that was. There's a free version and there's a paid version of it as well. And then you might have to have paid accounts that you link in on some of these LLMs, but it's It's a really cool tool. I use it in conjunction with my copywriting tools. It handles that. Speaker 2: Yeah, it's incredible. I got Danny hooked on it. Kevin King: He's on Manus right now. It's his big one. I think that he's playing around a lot with. Speaker 2: He called and said, well, I'm Danny Maness Poe now. So that was, yeah, that was, that was really funny. Kevin King: How did you hook up with Danny then? Speaker 2: No, this, this is like, this is perfect. So I told you about Max and Joe. I went on their podcast before, this would have been June, 2024. Then came Accelerate. This is where I met like a lot of really cool people. I feel like everything opened up after Accelerate because it was a late night. It was a networking party that we have, which are the best parts I feel like of those conferences and talking everything Amazon. And you know, for me talking about search optimization, most of the people I talked to, I felt like, oh man, this is kind of boring. But then I met someone named Juana. And immediately, I thought to myself, where was this mind the whole time? But she immediately knew who I was because she had used my GPTs before. And instantly, we were both speaking a language very few people in that room could understand. One thing led to another. And after a number of exchanges over LinkedIn with Amazon Science paper nerd outs, she introduced me to Danny McMillan. Then I got on the podcast with him together in October showing him how Rufus can actually render images into the chat. I even showed him a screenshot one time. I said, look, this is the first time they released this and it was back in June 2024. I have a screenshot that shows Rufus rendering an image in the chat. But it was October time that it really was starting to get released. And one thing that like, Juana, Danny and I have phenomenal chemistry. It feels like unlike anything else in this industry, if you will. We just, as a team, we make sense. I don't know if you see that. I mean, like you said, the trifecta, you know, you got to interview Juana and then Danny and then now me. And then Rufus the Blueprint came out. And then my Alexa plus shopping patent analysis came out on Forbes. And then now of course I have huge projects that I'm working on. Kevin King: You're not working with the touch of class. I mean, are you just doing consulting stuff or you have your own gig? I know about every three days, it seems like I get a LinkedIn post and I'm like, hey, Kevin, here's a breakdown. If you want to use this in the newsletter or post this, like, dude, this dude's like, I mean, some of that may be help, AI is helping you do some of that, but I'm like, he's cranking out some stuff. So what the hell, what are you doing now? If you're not working with touch a class, what are you doing? What's this, what's going on? Speaker 2: I'm the Amazon lead. My full-time job is I'm the Amazon lead for the National Fire Protection Association. So, you know, we provide like all the codes and standards are bought by like, like for example, we train all Amazon. This is actually you probably need to cut this one out. We train all Amazon warehouses on NFPA 70E. So I work with, you know, different people within Amazon leadership to do that. So it's really cool like making connections with those kinds of people. And then between you and I, I actually got to talk with the director of Amazon search because I was telling him like, hey, these code books and standards are being misrepresented on Rufus. And that's a big deal when it's life and safety like that. I mean, our books are used for nuclear facilities, the sprinklers and nuclear facilities of how everything's supposed to be built. Like if you're an architect, you're going to build something. You have to build around our codes and standards. The Department of Defense buys from us. NASA buys our books for and we're working with them on the spaceports that they have because we know there's a lot of explosion that's happened with like spaceports. So I sell our code books and stuff on there. Right now I'm dealing mostly with counterfeit listings, unauthorized resellers. So I've been working directly with higher up, you know, leadership at Amazon to eradicate those sorts of things. Long story short, I'm their Amazon lead. It's kind of a new thing that they have. They have a ton of potential. I can't give you the exact numbers, but you know, it's in higher seven-figure ranges. Kevin King: So just to be clear, this FIRE, what's it called? FIRE Associate, what was it called again? Speaker 2: So it's NFPA, but it's the National Fire Protection Association. Kevin King: So basically, they're selling code books on Amazon to other businesses. And maybe posters or compliance posters or workbooks or stuff like that. Is that basically what it is? Speaker 2: Like huge 500 page code books that sell like crazy. You wouldn't believe for, you know, very high prices. Kevin King: So these are, and these are required because the businesses have to go through some sort of audit or some sort of safety thing that, and this has all the code of everything, how they have to be in compliance, basically. Speaker 2: Yeah, it goes through 300 different committees. You know, we're talking from, PhD scholars to fire chiefs, you know, people, boots on the ground, putting everything together to make sure. And then right now, of course, we're in the wildfire space. We have been in a while and that's been a big deal. But yeah, we're, you know, we're very big in that industry. Like if you're an electrician, you know, NFPA, right? Use them all the time. Kevin King: So how does this, how is doing Amazon for this, which is much more business to business? And technical and compliance different than what you did when you were doing Amazon for Touch of Modern, which is more home decor. Speaker 2: Touch of Class. Yeah. No, you're okay. Yeah. No, it's much different. It's kind of like two different sides of Amazon. I just have a ton of experience and I can basically do anything I can set my mind to. And so I analyze and solve the huge potential of this brand. And they had a position open, and so I thought, okay, I'm definitely going to take it. It was a very, very good offer, so I decided to take it. And here's the thing, though. The big reason is it's fully remote, too, as well. Whereas the job I had previously was hybrid, and I was traveling like, you know, an hour. And so it was better, especially having twin girls now, that I be fully remote and able to be here with my wife when I need to, but then also be able to work diligently. I do love my job. I love what I do. And I like that what I'm doing here is a side project kind of everything. I know it seems like a lot to be a side project, but every day in the evenings from the time my wife goes to bed till the time I go to bed, it's usually about three hours or so, I just spend and I just let my creativity just fly. I just start thinking, Something is unlocked to me in the evenings that I can just, I can just think, I can just innovate. You know, I know that sounds abstract, but it's so beautiful when you have that freedom in three hours just to innovate. No strings attached at all. You know, there's nothing, I got nothing to prove. You know what I'm saying? So yeah, I just have tons of research projects now that I'm working on and we can get into those if you want to as well. Kevin King: So you met Juana at a party at Accelerate and that led you to Dan and the three of you geeked out and decided, hey, let's do this 10,000 word document, this big long thing where we analyze all the Amazon finance papers. Y'all put that out, which was great and really opened a few eyes and also a few people disagreed. I don't think this is the way it's going. And now you and Juana and Danny are all kind of doing your own little individual little research projects and putting those out. So what is it that people I've been preaching this for about a year saying like look a lot of these sellers right now they're doing millions of dollars or they're not paying attention. We're gonna get left behind and even some of the software companies and I'm of the opinion that right now you kind of kind of write a line between doing it the old way and the new way. But to me the writing's on the wall based on what your research says, what the data science says, where I'm seeing things just naturally going. That a few years from now, if not sooner, it's all going to be AI oriented, all intention based, all semantic, all gaming the system. Amazon, especially in our industry and the way a lot of people have been taught, is gaming the system, finding the keywords, make sure they're in your title, stuff in your title, making sure that you find keywords or someone else is missing opportunity on and you come in and grab that market share. A lot of that stuff, in my opinion, is going to completely disappear. And it's all going to be through AI search. And it's not just Rufus and with Cosmo underwriting that, it's also perplexity search. And it's also open AI search. You look at open AI, I just did a partnership with Shopify. Right. Speaker 2: Yeah, I did see that. Yep. Kevin King: You're going to start seeing it and Perplexity has a, you know, it can search Amazon. And that's one of the reasons Bezos is an investor in Perplexity. Speaker 2: I use Perplexity. We, in fact, we haven't, we don't use Google anymore ever since her pregnancy and the kind of answers it gave us. Yeah, we don't, we don't use Google. Kevin King: Yeah. I mean, even Google now, if you're still using Google, there's the AI summaries and that's changing SEO. And I just had Carabab on. Yep. We did a podcast last week just before you and we were talking about branding and how when we got into discussion about how perplexity and these AEO or there's different words for it, but basically instead of SEO, it's AI engine optimization and how you're going to have to change that and how you have to monitor social media and build that in and all these parameters. What are your opinions on seeing the big picture? You've dove deep, way deeper than I have. Your head's around this a lot more than me. Where is this going? In layman's terms, what does someone listening to this need to wake up and realize? Speaker 2: OSO. Remember that word. Omni Search Optimization. The new paradigm is omni-channel, multi-modal, generative, and agenic. Okay, what do I mean by that? It's Omnichannel. We all know what Omnichannel means, you know, between e-commerce, our own sites, and social commerce, of course, social channels. And then it's also multimodal. So you can upload text, image, video, and that's, of course, where it's headed. You know, not many people do as much for video, but you will be able to. You can do all three at once. You can do one or the other. The point is, you'll be able to search that way. And then, or what you could do, well, The input of that is important. So you can have text, image, video in the input when you're uploading and searching for something, but then you can have also there's the output of search of text, image, video, and it's generative. So it's Omnichannel. It's Omnichannel, it's multimodal, it's generative, right? So large language models are being utilized in specific. For example, TikTok is doing this with their search highlights where you ask a question, it's generative, right? But then it also gives you videos. So it's multimodal coming back out at you. And then the next stage is it's a genic. We see things facilitating purchases. So like right now, like Rufus, it can respond with text and image and it does give video in Sponsored Brand, but it is going to integrate into the actual chat as well organically. But expect it now to be also a facilitator of purchases. For example, think buy for me, but just on the Amazon site, right? Hey, I need you to go buy this for me right within the Rufus chat. You can check out right within Rufus. That's what I think is gonna begin to happen. But I think you see this kind of everywhere. I think being on e-commerce platforms is especially important because, for example, Google, the number one factor for it to show up is your authority score. What is your authority score? What kind of authority do you have? And if you have just your own site, it's going to be significantly less than Amazon because if you have the authority score of Amazon and you're doing really well on Amazon or even let's just say like Walmart or whatever, you're much more likely to show up because in perplexity, sure, it shows some other brands, but it's not going to show a touch of class anymore. It's going to be just showing Amazon because of how high that authority score is. So yeah, I think that's definitely where it's headed. I'm working on a huge framework specifically for that. I'm excited to eventually share it and I'll share it with you before anyone else. But I think OSO, and it builds off each other because maybe it's not, it's also virtual. It takes into factor augmented reality because when you say multimodal, multimodal can mean text, image, video. Augmented reality which we already see on Amazon, right? But think about that. You'll be able to view from your room right within the Rufus app and then it'll immerse you into an experience and you'll get to shop that brand store. So having a brand store page but then also being able to immerse yourself and we're seeing that right now. I don't know if you've tried out the virtual like Star Wars shop on Amazon that they have. They had a toy shop recently that came out during Christmas time. I think that has a long way to go, but I can see everything being done within Rufus. And of course, I think Alexa Plus is going to be able to obviously call out Rufus and Rufus will be the agent that does that sort of thing for them. And then there's Health AI that they released that's powered by Cloud 335. And it's a reasoning model. And, you know, Rufus, if, for example, if you're trying to buy something that you want to be medically sound, like if you're buying something for your kid for a cough, you know, Rufus will, you know, find something for you, but consult health AI assistant, like as a, you know, a shopping AI consultant, right? And this is where Agenic AI comes into factor. But yeah, I believe the future of search is OmniSearch. Omni-channel, multi-modal, generative and agentic. And I believe that accounts so far for everything that we're seeing from Google AI overviews, perplexity shopping results, you know, copilot, And by the way, I've always been a Bing guy. I don't know about you, but I've really never been a Google guy. I've always been a Bing guy. So when Copilot came out, I thought, oh yeah, this is awesome. Just another reason for me to brag about Bing. And I feel like Google's just trying to play catch up. I think they've been too stubborn in this whole race, and they're playing catch up with everything. And because they have some of the smartest people on the planet, of course they're going to come out with an AI that's really good, like Gemini 2.5. That's pro. That's not bad. Kevin King: You said like on Google though, I mean, one of the important things a lot of people overlook or two important things is the Google Knowledge Panel to get that authority, which you can actually, that will raise you dramatically. And then Schema. Schema is being used a lot by AI. And if you know how to do Schema, that's S-C-H-E-M-A for those that don't understand what word I'm saying. You can go look that up and if you set up your websites and a lot of your blogs and all your posts and everything, the content that you're putting out and you follow the schema, you can actually start showing up a lot higher in these AI searches and these AI recommendations. Another thing that you mentioned that the three things there, the text, image and video, those are things that you can control and you could submit and you can control the narrative and the keywords and the imagery and the messaging, but also the AI is going beyond that and going beyond things that are not in your control. Such as customer reviews and social posts, what people are commenting on TikTok or on Instagram or whatever. It's reading those comments, reading that language and incorporating that into the intent-based search and you're not necessarily as in control of that. So that's where it's even more important to have correct messaging and correct branding. And we really be trying to become more of a brand than just a seller of products. And I think that's a shift for a lot of people that might be difficult to make. Speaker 2: Yeah, I 100% agree with you on that. I just know like, you know, with whole AEO, answer, you know, engine optimization, GEO, generative engine optimization. In fact, I remember showing... Kevin King: Tell people what the difference is between the two of those just that are listening that are like, what's he talking about? Speaker 2: Well, they're very similar. I think people say, I believe they're synonymous in a lot of ways. But here's the difference. Answer engine optimization can just be simple NLP techniques. In other words, it doesn't make it rely necessarily on a large language model. It can rely on simple techniques that are able to like summarize information and present information. And it can also be generative. But generative is very specific to it's creating unique text that goes along with With video, with images, with text, combines those things. So like Answer Engine Optimization honestly has been around on Amazon, for example, in the customer review section where you can do Q&A asking. Before Rufus was there, there was a simple one and it used NLP techniques, which is something that doesn't rely necessarily on large language models, that then was able to produce a summary from customer reviews that were filtered in there. So yeah, I think the whole notion of I think we're just desperately trying to come up with a new language for this. I think what that's that's where that's coming from. And so here this framework I'm trying to provide is something that I believe I'm coming with entirely new language for this to be able to help. I help sellers in particular optimize but also brands too because I'm from that perspective where I know what it's like to be omni-channel. For example, I built up our Pinterest Touch a Class from 0 to 145K followers on Pinterest. Pinterest of course partnered with Amazon and then TikTok did the same as well. And so having a presence on those two platforms, I think is super key, at least it was in the home decor category. And then also, of course, being on, you know, Walmart and, you know, eventually like, you know, be on Taimu and then Amazon, but then also having your own site where you're directly, you know, shipping through your own warehouse, but not selling on, you know, another platform. So I'm very aware of being omnichannel. And how to optimize for that specifically. Kevin King: So what are some things people should be doing right now, whether it's on platform on Amazon or in prep for this Omni channel? What are some like two or three very specific steps that they should be doing that you've seen that might help some people? It's just basic introductory type of stuff. Speaker 2: Yeah, I would definitely say, you know, from like an SEO perspective, is that what you're saying? Kevin King: Yeah, from an optimization perspective to appeal to where we're going. Speaker 2: Yeah, of course, optimize per platform, be very specific on what the platform, because each AI isn't the same. You know, we got to remember it's not like there's this one self-contained AI. No, it's like think of AI, but then think of every LLM is tailored to each individual platform, which is I believe why like patent and Amazon science is very important because it shows you exactly how that LLM is working because their LLM is working much differently. I mean, there's a lot of the same stuff than Google's LLM. Right, because Amazon's trying to do everything at scale so that there's no latency. Google doesn't really have to worry about that as much because they rely upon an own site to be able to develop and have site speed, whereas Amazon gives you the speed for every single product, which I think is even more impressive that it's able to accommodate for that. But that's why Google, it's easy for them to go quicker into semantic search. It's a little bit more difficult with Amazon because the more you read through all the literature of science papers, you start to notice a trend of like, well, Yes, it'd be awesome to use something like Claude35 in search, but we can't do that because of computational costs because Amazon's trying to deliver an experience at scale. And so that's the hardest thing that I think that's happening right now. I think that's why Google's always been ahead in terms of semantic search because it hasn't had to worry about the site speed of a brand at all. But Amazon has to worry about latency. And that's why I think like with Roofies, you get responses back very, very fast. And I think that's the point. And that's, you know, you basically said basically the same speed for any given user. Yeah, I would definitely, you know, optimize per generative per LLM. And then, you know, secondly, rely on that, you know, the reason you want to be on Amazon too is like a lot of times for Google or if you're on, you know, Perplexity or whatnot is because they reference those e-commerce marketplaces more than just individual brands do. If that makes sense, when a shopping question is asked, it defaults to Amazon and Walmart and the like, right? So I would definitely say those tips. Optimize per LLM. And for me right now I'm working on, I've scoured through thousands of patents and right now I'm putting together every algorithmic patent per platform. So I have like a couple on Etsy, 10 on Walmart, it's not that sophisticated. And I have 60 right now on Amazon. And being able to optimize per LLM with different types of things. For example, if you have just basic NLP techniques where you're not using an LLM, that's generative. What happens is you have to be way more clear in your language and use way more words as well versus if you have, and that has to do with semantic similarity as well. Because technically you can do a semantic similarity without something being powered by LLMs, which is exactly why Amazon is relying on a lot of the natural language techniques right now. And it seems like they're behind, but really they're ahead because they're doing these things at scale. See ya. Kevin King: So yeah, I just saw something interesting. You're talking about like Claude is too taxing or Claude's style is too taxing. I just saw Sam Altman came out with something. He said, quit being so polite in your prompts because the words please and thank you is costing us tens of millions of dollars. And computing power, or people putting please and thank you. I was like, yeah, and he's like, it doesn't do any good. And it's just costing us money. So please stop doing that. But that's that that was kind of funny. What do you think, you know, voice search is gonna I mean, you've been big on the on the text based and image based stuff. But what do you think? Voice search is going. Have you looked at any of the patents on like what Alexa Plus is doing? Because I see that, you know, I'm an older guy, so I'm still old school hunting and pecking and texting a lot. But a lot of the younger generation, something like 70-80% of their searches are done by voice. And voice, yeah, you're young, you're quite a bit younger than me. So you're in that younger generation. But so, That's a different type of search and inflection and you can put some personality into it and some of these voice models can actually recognize personality and inflection, which you can't really get in text. So how do you see that affecting search going forward and e-commerce? Speaker 2: Yeah, absolutely. I mean, most of it like Gen Z and such and you're going to see this with Gen Alpha, I believe even more like big time. Voice search, it's amazing. The new patents that I've showed you, it accommodates actually for voice search. It's very interesting and for every patent, every so number of times Amazon releases a series of patents and everywhere in there, it's always trying to optimize for voice input and how that's going to translate to the search bar. And what to search, right? How do you use semantic similarity to take a question, put it into a search that's most relevant to that customer that is based on behavior and semantics simultaneously? And I believe like voice search is definitely the future. It's a lot easier. But let me tell you this, if Sam Altman hates please and thank you, he's going to really hate when everybody starts just doing voice search because you're going to have a lot of random words that you know that he's going to have to compute. I mean, they just got to get used to that, though, you know, that people aren't going to just be able to type. I find a lot more productivity actually in voice. Voice searching, even for, you know, ChatGPT. I'm not talking about their voice mode, but I simply need like when, you know, I want a prompt, I'll like just voice out everything. And I noticed that it actually understands the context better, but I can't imagine how much computational power that that takes to do it. In fact, having a lot of people do that are probably melt all their chips too, just like the image generation tool. Kevin King: And you can even call, what is it? 1-800-CHAT-GPT now, and you can talk to it straight, straight on. Straight without the app or without logging in or whatever, and it'll tell you. It's pretty cool. Have you played with that? Speaker 2: Oh, yeah, definitely. It's really cool. Every time I go to the copy shop, I either walk or whatnot, or if I'm on a drive, I'm always talking with The voice AI, whether it's debating it, trying to get to do something that it doesn't want to do. In fact, I've actually gotten it to yell because I told it to play like it's a theater, you know, a play. And I said, have this character be someone that's very angry all the time. And so I kid you not, I'll share you the recording of it because you can share it. The character was yelling at the other person and they just went back and forth debating. And then by the end of it, one's exhausted crying because I had prompted it to do that and I thought, oh man. And I don't know if they've seen that and then more guardrails will be put up. And that's one thing I'm working on too, Kevin, is my prompting philosophy. I have a book, well, not a book I'm working on, but a prompting philosophy. E-book, if you will, that is going to make Google's that they came out with look like child's play. I don't know if you've seen the recent one that Google's came out with. Kevin King: They just released a new, they have their prompting school or something like that. Is that what you're talking about? Speaker 2: Yeah, that's what I'm talking about. Kevin King: Yeah, they just came out with a new one just in the last few, as of the time of this recording, in late April. They just came out with a new version. That's a good point though. Prompting is, I know like two years ago saying, oh, prompt engineering is the new hot thing. And then people start backing off like, ah, it doesn't matter. They can figure it out. But I disagree. I think the way you prompt something and the way you set something up with Is critical to the results that you get and what I mean, you guys said what maybe, you know, maybe this is in your book, but there's like seven different things that you need to make sure that every good prompt has, you know, the situation, desired output, the example, I forget all seven of them right now, the context, the length. There's like seven things that you got to specify in there. What's your philosophy on what makes a good prompt? Let's get some best results for you. Speaker 2: One of the basic ones is it is better to have a good structure. If you're going to build a prompt and you're going to do voice search, I recommend doing the voice search, then it gives you context, then you get a better prompt. Instead of just writing, you don't get all your thoughts out, but if you're talking, that's how you get your thoughts out, then let ChatGPT structure it and you might think, well, are we really going to get a good output of ChatGPT structuring it? No, you're going to get a very good one because you're giving it more context than you could possibly give it even if you structured it. So I would recommend voice searching basically just putting everything out there and then saying, hey, can you put, you know, give me some prompts for this to be able to do these sort of things. Like give it what you want to do in voice search and then also have it create prompts based on that. Because you remember like you can put as much as you want in there or you can have your voice search, have it structure it. And then ask it to create prompts for you. And then one thing most people don't know is, you know how when you upload a document, a lot of times it doesn't spit out everything you want it to spit out? Copy and paste it actually in the text. I call this self-referential prompting, where each time that you go to prompt, copy and paste the text with it every time with it. Kevin King: The text is in the document. Speaker 2: Yeah, not don't upload the document, copy and paste everything in the document and just copy and paste it like into the actual chat. Kevin King: Okay. Yeah, I do that a lot and put in quotes or something instead of actually attaching the document. Speaker 2: Yes, exactly. So or not just quotes like put the whole document in there and it'll give you an infinitely times better summary than that. But like let's say you're summarizing a document. You're trying to get insights out of it. Make sure to give it exact examples of what kind of conclusions you want to make, right? Like, okay, what makes a healthy account? Give 10 examples of what makes a healthy account, what doesn't make a healthy account, right? Put out your philosophy. And this is where it goes into like if you're a PPC expert and you're trying to determine whether this account is, PPC is healthy and you have to create tactics. What makes a good tactic? What are the exact examples? Don't leave anything in the abstract if you don't have to. And that'll give you a good framework going forward to be able to like read something and then be able to make those interpretations as well. So I definitely am a big believer in what I call self-referential prompting and then also multimodal prompting where you're using text and images to do it. If you have text and images in there, it gives a way better output than if it doesn't. For example, with the new image generation tool, if you're trying to get something, hey, make this look exactly like this and it's kind of like close but not so much, upload both those images in there and outline, hey, do you see that this problem? And so having both, you know, some kind of screenshot of the problem and the text will help you tremendously and give AI way more context because you have to remember, if we just heard text, I mean, we're expecting AI to operate in the blind, you know, but if we give it visuals, right, you get better context. You don't want it to operate blindly with just text. You want it to have an image if you can. So that's where like, you know, visual reasoning, you know, comes into factor and, you know, ChatGPT's new models. O4 mini-high, which I love, and O3, like a regular O3, have been incredible. Kevin King: I'd also recommend Claude37. That image stuff actually works, what you're saying. I've used that twice just in the last couple days. I was trying to do a make.com automation. Automation and I was having trouble linking a Google sheet to it. And so I was given this error. So I just took a screenshot of the screen, you know, with all the, all the message and saying why it didn't fit. And I put that into ChatGPT and said, how do I, what, what, how do I bypass this? What do I got to do to fix this problem? And it came back with the solution. And I did the same thing with Beehive, my newsletter platform. It was, I was trying to do an embed code to embed something. And I was just acting up and wasn't doing it. I said, here's a screenshot. Here's that message I'm getting. Here's how it looks. How do I actually fix this so it looks like this? And said, oh, you need to turn this on or on Beehive. Actually, iframes don't work. You need to change it to this format and start with this, this and this. Here's the copy. Here's the code to put in. And I was like, Jesus Christ. I mean, what was this like? Ten years ago when I was developing stuff like beating your head against the wall for five days and never getting it solved and just boom, just like that. So your point of including a screenshot or an image is very good and it recognizes stuff. It's amazing. It's like a little magic sprinkle dust. Speaker 2: How many times have you ever said and then wait till we're able to like send screen shares, like exact screen shares, which you technically can, but like it only sees it in moments of image. It doesn't see the actual like film of it all yet. But imagine where you can do that because how many times you've been on with someone on a call and you said, hey, can you screen share with me? Kevin King: They can fully understand and see what you're talking about and not be trying to guess, was he over here or over here? I just wish I could do that with my parents when I'm trying to tell them, plug in the USB cable to the back of the device. Which one is that? Which one is that? The skinny one or the big one? It's like, no, plug the USB 2.0 cable in. I don't understand. They're trying to plug an HDMI or something in there. It's like, no, no, no, the USB 2 cable. Speaker 2: They'll just come over. Kevin King: Yes, exactly. They just pile them up. When I go see them for Mother's Day or for whatever, it's like, we have a piece of paper. We have 17 questions here about things to fix on IT. You mentioned earlier, A little bit about agents. Where do you see agents playing a role for e-commerce? I see it where I've been talking about this some in the newsletter more and more. I got something coming out, another story coming out on it. But I see it where you're going to have a person overseeing agents. And so instead of a person overseeing employees, You have somebody at the top of an organizational structure. You have employees overseeing agents that are doing all this stuff that maybe took a department before to do and now one human is overseeing it, whether that's listing optimization research, inventory forecasting, whatever it is. Where do you see agents playing a role? How do you see them playing a role in e-commerce? Speaker 2: The best companies are going to learn, obviously you probably know this, not merely, you don't want to automate your best skills. You want to augment them, right? Obviously you want to automate the modernity of life, right? The modernity to accelerate your creativity, but you also want, and I'm working on this in my AI manifesto, another project I'm working on, but you know, the basic is just AI does not automate our best skills. It augments them. So think of every employee you have and think of all the different tasks that they do. They will be able to work with agents. I mean, I believe working with agents will be as important as a typing class one day in high school that you're supposed to have, you know, because typing would just change the game, right? And obviously this is going to change the game even more. So I think it's going to go from there to prompt engineering are going to be classes that will eventually be taught. I know there already are in the university, but even in high school, I think it's going to be important. And I think kids are already learning that because, I mean, think about Gen Z and everyone, they're probably some of the best prompters in the world just because they know technology so well. I think we maybe take that a little bit for granted. But yeah, I think like facilitation of purchases is certainly one. But honestly, that's just, I hear that all the time. That's only like a little bit of it. Like, you know, kind of like you said. But I think in order, for example, to like augment, you have to be very aware of like your specialized role. And your prompting is only going to be as good as the knowledge that you're able to put in it. And the Agenit tasks that you can have it do and consult with. For example, I have a Po app that I built that is able to talk to each other based on different personality types. So I call it a team consultation and I give it a problem and it does nine different API calls all to different, I use the Enneagram, different like personalities. So there's like Enneagram 1, 2, 3, 4, 5, 6, 7, 8, 9. Nine's a peacemaker, eight's very much in control, six a loyalist, four's a creative, three's an achiever, one's a reformer. It's all about the good stuff. And so I ask, hey, I need your opinion on this and what should I do in this situation? And it makes all nine API calls to them. And then they, in a meeting, there's more API calls that take and they go back and forth talking. And then it comes back to me with a conclusion with another agent I have that's a generalist using GBT-4, goes in, takes in all the, you know, kind of like maybe what you do like with podcasts. You're so good at like just taking all my language and then you put it back. I'm like, man, that's all I had to say. Like I'm talking with you right now. I'm like, man, that's all I really had to say, I guess. But kind of like that, you know, and it's able to give a bottom, you know, instead of like Amazon having to craft six-page narrative, you know, briefs like they've done, if you remember that, the Amazon leadership, you know, if you have an idea, everybody has to have this. Everyone has had to read it. Kevin King: Yeah, do the press release in advance. Yeah. Speaker 2: Yeah, exactly. Kevin King: And no meeting can be more than a single pizza meeting. Speaker 2: Yeah, right. Yeah, exactly. So yeah, things like that that you can do that I think are going to like help so much, like that's going to help an HR person solve situations of employees because those Enneagram types, you'll have specific prompts behind all of it. So every agent will have an intricate prompt because a single API call can't possibly do everything at once. So you have to have separate prompts that power different agents to do the tests you're going to do. So when I said like you're researching with Rufus, you're looking for a really good knee brace and then it consults health AI. Maybe it says, hey, this isn't very good. Plus the customer reviews say this, this is in line with medical science. But you know, we do recommend these and then I'll spit out ones that are even better personalized to, you know. I'm the person but you're absolutely right. Kevin King: Upload a picture of your jacked up knee and it even makes it better, right? Speaker 2: Exactly. Yeah, exactly. Unknown Speaker: Yeah, show your knee. Speaker 2: That's a good one. Yeah, that's good. Kevin King: Upload the x-rays from the ER and it'll get you the exactly right one. Speaker 2: Yeah, exactly. So I see lots of that happening where agents will consult one another on various things and then that research is gathered. But there always has to be that human element in it. Yes, automate what you can automate, but I don't think it's going to take jobs as long as people know how to use AI for their job. If they can master AI for their job, for their knowledge, industry knowledge and their expertise in a particular field, I think it's just going to make them better at what they can do and augment their abilities. And we'll take companies. I think the best companies will learn how to do that to augment their skills. And for me, it's helped with my optimizations and everything. It's not that it automates my product listings as much as I believe it. It augments them and makes them even better than what they were before at scale, of course. Kevin King: It's a tool. It's not a one-all be-all, but it's a tool to make you much more efficient and to give you a lot more power to do things really quickly. And like you said, to scale, which is scaling. I mean, it's hard to imagine, you know, I sit here and like I can, I remember the days before we had one of these phones in our pocket where you could just look up anything at any time. And to get a hold of somebody, a business partner, I had to page them and they had to pull over in their car, pay phone. Speaker 2: What's that? Kevin King: What's paging? Put in 25 cents and call the number that showed up on the screen and that was how you reach somebody if it was an emergency. And now with AI, it's only been around for three years, basically, not even three years, two and a half years. Well, it's been around a lot longer. It's been around 20 something years. Speaker 2: I know what you mean, generative AI. Kevin King: It caught on with ChatGPT in November 2022, I guess is what, it emerged from the shadows. And you had Jasper.ai driving stuff, like you were talking about earlier, the camp I had the thing. That existed, but it wasn't This hot topic like it is now, but now I'm using it and it's evolving so freaking fast. Every day someone is one up and up in another one. I can't even imagine doing some of the work that we do now without it. I mean, it's become an integral part of my daily routine. And you said like, just like you said, you and your wife, you use it for search. You don't use Google anymore, but we're so engrossed in it, you and I, and we're, On the cutting edge, and you got to step back and think that 95% of the population, as big as 100 million installs on ChatGPT or whatever, you got to remember a lot of those are multiple things from the same company or different API calls or whatever. That's not 100 million different people. There's still a very, very small percentage of the population that's really actively using these. People have dabbled or played, but actually using it as a tool. They may be talking to their phone and not realize that AI is behind what it's given them, but using it as a proper tool, there's still a very small percentage of the population doing that. And so the opportunities are huge. Speaker 2: Yeah. Maybe just from my experience, everybody I've met or know around here, even in my church, has actually at some extent used ChatGPT. It was fascinating because I forget, you know, like Really? You guys use AI or they use something on the meta platform because, you know, they push out all of that stuff. And so people apparently do use it, not to the level we do. Like you said, we're, you know, on the cutting edge of things. But yeah, you'd be surprised how the everyday person, I mean, that's exactly why with Apple Intelligence, you know, what's its slogan is AI for the rest of us. You know, you have all these advanced prompts and such, but I think Apple is, you know, going to do really well with this Apple Intelligence because it's recognizing AI for the rest of us, for the everyday user. And that's something that Apple has always been very, you know, I believe very good at doing. And that's, you know, they of course integrated with ChatGPT and all that. Yeah. So I think it's going to be so seamless, you know, as it starts to go in, but Of course, tons of debates are going across companies of, hey, can we use this? Can we use this? Because people in the business field, I feel like are having way more. It's easier for them to adopt it than the academic world, for example, which is so interesting because usually the academic world historically has always led the culture. So before something would go into the culture, it always was a debate in academia. But I had a conversation, Max, Max had brought up something and he was talking about what, you know, you had talked about and, you know, immediately I was just like, well, I have to say that, like, you know, that is the complete object objectification of, you know, of women. I think an AI will thrive for parents' sake if they decide, like, ChatGPT right now can't have, like, inappropriate pictures or anything like that or inappropriate searches, but perplexity can, from what I understand. Exactly. Exactly. I think ChatGPT is gonna stand out because it has that. As soon as they don't go that way, I think that it's gonna fall flat in the younger generations. And maybe it's just because the people I know go to my church, but one thing I've educated them on is they can do this and they're like, oh my gosh. But some of the things you hear of the AIs that they're using, like AI girlfriends and things like that, which is why someone like me, I wanna be on the cutting edge too for the sake of humanity as well. So there's that aspect of me. Kevin King: It's not just the dark sides of society there, but also these tools can be used for war, for criminal aspects, for a lot of things. And then when you have the new quantum computing coming on that's going to accelerate this and people have access to quantum computing plus AI, it's going to get crazy on what some people with malintention can do. Speaker 2: I just watched a show. I've been watching AI TV shows and there was a show that, I'll send you a link to it, but basically it had that quantum computer set up, the one that Amazon was developing and the AI assistant was, you could tell, very hostile and for the sake of marketing and appeal, they didn't focus on that and so it became adversarial and of course everybody in the house died. You know how every movie like that goes, that deals with generative AI. So I think like having those ethical guardrails are super important. But I'm seeing more and more shows like that come out and putting it in a real context because you think, in the show it kept saying, oh, so it's like an Alexa. It's like, no, it's way more powerful than Alexa. And if you think about it, Alexa Plus and such, those kind of things are coming. That if there were no guardrails to it, yeah, it could be adversarial and could, you know, destroy humanity. I grew up watching Terminator as a nine-year-old kid for one of the first movies that my dad introduced me to, that Commando and the Arnold Schwarzenegger movie and Rainbow. But anyways, Terminator scared the heck out of me. And so I was like, I always think about that, like, oh, is AI going to destroy us? But I feel like OpenAI and Anthropic, they do have good ethical guidelines. I think they have great leadership of what's going to happen when AIs rise up that are popular that don't have good ethics and such. And so I think like United States being on the cutting edge of AI is crucial, like you said, for warfare and protection of the country. Kevin King: Cool. Well, Andrew, this has been awesome. We've been chatting here for a while. If people want to follow you, follow the work that you're doing, all the research papers and everything that you're doing and sharing or get copies of your GPTs, what's the best way for them to do that? Speaker 2: Pager. Kevin King: Just send a page to this number and next time you're passing a pay phone, Andrew will stop and call you back. Speaker 2: I stay grounded in history is what I'm trying to say. LinkedIn, of course. I feel like that's the number one place. Kevin King: So Andrew Bell, just look up Andrew Bell on LinkedIn, Andrew Bell AI, Amazon, you'll find him if there's different Andrew Bells, I'm not sure. Or just look at the picture on the thumbnail of this podcast and that's what he looks like and you can find him. Cool, Andrew, I really appreciate you coming on here, man, and sharing with us. It's been fun. Speaker 2: Yeah, it's been a blast. I've enjoyed this conversation as well. Kevin King: No question, AI is going to be changing everything and every aspect of how we sell and how we operate our businesses as you can see from this discussion with Andrew. So hopefully, you're staying on the cutting edge of it and you're going to be prepared for the future because there are exciting times ahead, including next week with our next guest on the AM PM Podcast. We'll be here with an eight-figure seller who's selling in multiple marketplaces around the world. We're going to be talking with him about what's given him his edge and how he's handling all this crazy stuff that's going on in the world of e-commerce today. But before we go, I've got some words of wisdom for you. If you want to make the wrong decision, just ask everyone. That's right. If you want to make the wrong decision, just ask everyone. See you again next week.