#408 - Innovating E-Commerce Through AI with Max Sinclair

#408 - Innovating E-Commerce Through AI with Max Sinclair Speaker 1: Welcome to episode 408 of the AM-PM Podcast. This week my guest is Max Sinclair. Max used to work for Amazon for about six years in the UK and Singapore, all over the place. And then he went out and started an AI company to help Amazon sellers as this new AI revolution is taking place. He's actually a couple months before ChatGPT actually launched. He launched his service and it's really, really, really cool. So we talk everything AI, what you need to be thinking about as an e-com or Amazon seller when it comes to AI. What his tool does and just the general direction everything is going. Enjoy this episode with Max. 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. Let's do this. Here's your host, Kevin King. Speaker 1: Hey, what's up, everybody? Kevin King here. You know, one of the number one questions I get is, how can you connect to me? How can I, Kevin, get some advice or speak with you or learn more from you? The best way is with Helium 10 Elite. If you go to h10.me forward slash elite, you can get all the information, sign up for Helium 10 Elite. Every month, I lead advanced training where I do 7 Ninja Hacks. We also have Live masterminds every single week. One of those weeks I jump on for a couple hours and we talk shop, we talk business, do in-person events. Helium 10 Elite is where you want to be. It's only $99 extra on your Helium 10 membership. It's h10.me.me forward slash elite. Go check it out and I hope to see you there. Welcome to the AM-PM Podcast. Max Sinclair, how are you doing, man? Speaker 2: I'm good Kevin, good to see you, how are you? Speaker 1: I'm alive and kicking, fighting a little cold here, you can probably tell by my voice, but money never sleeps, so always gotta be working, right? Speaker 2: Yeah, the hustle, exactly. Speaker 1: You're traveling around right now, you said you're in Vancouver, But you're originally based in the UK, right? Speaker 2: I'm from London. Yeah, I'm from London, England. That's where I was born. But I've been traveling around these last two years. Speaker 1: Now, your background is you worked for Amazon for a while, right? Speaker 2: Yeah, I was there for six years. Speaker 1: What did you do in the London office or where? Unknown Speaker: Again, it was pretty global. Speaker 2: So I started my career as an account manager in 2016. So I was helping to launch sellers on Amazon for the business, kind of walk them through their first FBA shipments, first sponsored products, all of that kind of good stuff. I kind of transitioned from being a strategic account manager, so I was managing the top 10 sellers across Europe. So I had Anker, for example, I had Sun Valley Tech, I had some household names that you may recognize. Speaker 1: You were their SaaS rep or you were just assigned by Amazon to look after them? Speaker 2: I was their internal Amazon account manager. So I was their POC for all things Amazon. I mean, we were great in that role as a strategic account manager. It was less metrics-driven, I'd say, than some of the other roles, like we're trying to increase revenue, but more just trying to keep the big sellers happy, basically, and help them to grow in a strategic way. To be honest, with people like Ankur, I would have been 20, Like 22, 23 at the time and I was learning a lot from them, to be honest. They really were pushing the boundaries of product innovation and their Amazon strategies were far beyond what I knew. I've been in the company a year, year and a half, so I kind of knew some stuff, but these guys were the experts, but I was helping them manage everything internally. I mean, you know how it is, right? There's lots of different departments that a seller of that side needs to coordinate with and it can be a bit of a challenge. So, yeah, that's my role and also helping them, you know, product launchers and this kind of stuff. Speaker 1: How did you actually learn Amazon? Did you see a job posting and go and take it and they just kind of train you that this is how Seller Central works? This is how you open an account and here's the SOP and just regurgitate this to somebody or what's that process like? Speaker 2: Literally, it was my first job out of university. So, I had two offers. I had a consulting offer. for like a big consulting firm and I had Amazon. People were pushing me at the time to go to the consulting, but Amazon actually paid better. So I was like, this sounds cooler and it pays better, so I'm off here. So I kind of was a strategic account manager. Just to finish off, then I was part of the team that launched Amazon Business in the UK, so the B2B side. I then went out to Singapore. I launched Amazon in Singapore, which was just the most amazing experience, and there I was in charge of browse and catalog quality, so getting deep into the A9 team, setting up the browse. Browse notes and this kind of stuff for the Singapore marketplace. And then I came back to the EU and was in charge of the 3P grocery business. So I was helping supermarkets to sell as 3P sellers. So like, maybe you won't know them yourself, but like in Europe, it would be like Monoprix, Dia, Morrisons in the UK, Iceland, so big supermarkets in the EU, helping to launch them on the 3P marketplace. And that's where I left. Speaker 1: What are some interesting things you learned working at Amazon about e-commerce? What were some eye-opening things that you learned? Speaker 2: The most eye-opening thing in my perspective would be the culture. Amazon, at least when I started in 2016, had an extremely strong culture. And I could probably recite all of their kind of leadership principles off the bat, but I'm not going to. But you know, they had like ownership and dive deep and trust and that kind of culture was really how the company would be organized. And you'd have Jeff Bezos at the time would be doing Kind of quarterly all hands. So he would be, you know, very present and at the time he wasn't the world's richest man. So he was, he wasn't a household, he became a household name, let's say in like 2019 when he became the world's richest man, but before then he was kind of this nerdy looking kind of You know, Tech Brew, right? And he was an extreme visionary. And you could see that just from the way he talked about customer obsession and thinking big. And he kind of set the tone of the company in these kind of quarterly all hands. And that really When I joined, Amazon was 350,000 people, so a lot of people. When I left, it was over 2 million, so it grew enormously during my time. But that culture was super strong, and that's something that I think was my biggest takeaway. And I saw the culture kind of change and shift. Especially when Jeff left was a major shift but the culture changed and diluted a bit for the worse in my opinion sadly. And the mission that individuals felt in the business kind of subsided a bit with that. But that was my main takeaway, I think. Speaker 1: Did you ever sell on Amazon? Did they encourage you to try selling? Or did you ever try selling? Speaker 2: So one of my... At one stage, I was in a team of two when we were launching Amazon Business in the UK. And the other person in my team was someone called Sam Horby, who went on to set up Old Sam and is one of the biggest aggregators around. I remember at the time he's kind of sitting there doing his Running a notebook business. And in the end, I think he sold that to someone else who I was there at the time with, who went on to build Sutrana, the agency in the UK. But personally, I never sold on Amazon. I regret it massively now. I remember they didn't encourage it, but at the same time, they didn't discourage it. So when Sam was doing this, he was kind of At the beginning, he was kind of sharing these interesting insights with the team like, oh, I'm becoming a seller and this is a challenge. And it was helpful, right? Because working there, you'd see what was happening, people facing the pains firsthand and reporting that each week in a meeting. But I never did it myself personally, which was a big regret of mine. I guess one of the things I did do, which did help me is there was tons of educational resources. So I did, there's something called the Amazon Machine Learning University. So I graduated in history, but I did all of, you know, I taught myself how to code at Amazon. So I'm not, I don't code content, like I'm not a coder, but like, I have a very rudimentary understanding of it. And I, yeah, I kind of did this, you know, my side project, let's say instead of Instead of becoming a seller, I was actually kind of leveling up on machine learning and kind of just going through and understanding all of this, which helped me to get roles that I got later down the road in Amazon. So like, you know, being, you know, working with the A9 teams and the browse teams and all this stuff required machine learning, a grasp of machine learning. So that, yeah, I mean, I did that kind of on the side of, you know, whilst at the company. Speaker 1: So you did this for six years from 2016 to 2022. Yeah. And what changed in your mind? You said you wanted to be an entrepreneur. Did you see this opportunity emerging in AI or is it because of the machine learning that you were studying? You started to see some patterns? Yeah. Start a company around AI and Amazon or was it just I just want to do something different. I'm not sure what I want to do. Let me figure it out. What was that process like? Speaker 2: I guess I reached a seniority level at Amazon where I felt that I kind of hit, let's say, five years in. I'd hit where I wanted to get to. When I first joined on the graduate scheme, I'd always wanted to basically be a head of a department, have a team that has direct reports and indirect reports into the team. So I wanted to kind of have a mini organization. And I kind of got to that stage and I was like, the next stage up is such a big jump. It could take five years of hard work and there's no guarantee. That I'd achieve, that I'd even get there, right? It's such a tight pyramid that like, I could slave away for five years and see nothing on the back end. So I had started experimenting with different business ideas. So, I mean, I won't go into them now. You asked if I was a seller. I have been an online seller, not on Amazon, but I, you know, I did a bit of kind of depop and kind of, you know, like soft line stuff, like, you know, buying, rebuying and selling. So I did a bit of that. I explored some different business ideas, but I felt I felt that while you're managing a full-time job and you've got a lot of responsibility, I was dedicating one or two hours every other evening to the side hustles and I just couldn't give my full attention to it. And as a result, nothing was happening. And then really this AI A wave came and my co-founder, my now co-founder, showed me the first release of Stable Diffusion. And this is kind of, I'm sure many people listening are familiar with this, it's like mid-journey, right? You put in the text, you generate an image. And he showed me this and I said to him, could you put a product in that? Could you put a product in that image generation? Rather than generating a random suitcase, could you put in a specific suitcase and generate that in a lifestyle image? He built over the weekend a scrappy version of a demo and it had all the problems of like the hallucinations and the distortions and it kind of looked a bit weird. But from basically that weekend, you could see how transformational this was going to be and give more time and energy and effort that this was going to change content on marketplaces. And it really felt to me like this was a once-in-a-generation opportunity. It required my Full intention, like my full attention. So, I mean, we hadn't raised any money at this point, but I left, you know, to go full time on this because I just believe that, you know, this was basically it. And if I didn't do it now, I would regret it for the rest of my life. So, yeah, we went full time on the content at that point. So that was in September 2022. And in November 2022, everyone will remember this because ChatGPT launched and it kind of changed, you know, suddenly what I thought was going to be Transformational, but have a longer, you know, take longer than it for people to understand what AI was. And I remember with our, you know, our very first customers, we were trying to get them to prompt and they would have never prompt, they literally never prompted, they'd never seen a technology like this. So you're saying to someone here, like now type in a prompt and create yourself an image and they didn't know they, you know, They kind of didn't know really how to handle it. Suddenly, everybody knew how to use ChatGPT within a month, two months. Fast forward to 2023 and everybody knew what was happening and it suddenly was just an incredibly hot space very quickly, but also a very exciting space because people kind of could see the solution. And I guess one of the insights we had then, which I think will come to bear now, is you could see with ChatGPT, Not just the image side, but you could see how much better a conversational experience was for discovery. It was just so much better to talk to the... if you're looking to understand something or buy something, to have a conversational-based dialogue rather than type in a search and then scroll down and try and find what you're looking for. You know, from from chat GPT launching, it really became obvious to us that All of these major marketplaces, Amazon, eBay, everyone, they're going to have to go this way because if they don't go this way, someone else is going to go this way and everyone's going to flock to using it because it's going to be just so much better. So we've been working on those two principles, I guess, since we started. Let's help people to create content with GNI, number one, but also let's optimize for what we firmly believe is going to be the future of search, which is going to be LLM-based search. Speaker 1: So with your software, you've expanded it quite a bit since the first launch, but one of its features is you can, like you said, take a suitcase, your suitcase, just take a picture with your phone and stick it in a scene and create the entire scene around it that looks real. And are you tying into an API of one of the big image creation LLMs out there to do that or how are you doing that? Speaker 2: Yeah, so we have actually filed a patent on our technology and the answer is we don't train the base model. So we're using, you know, there's various different base models out there like Stable Diffusion, DALI, Minstrel, OpenAI. There's a lot of open source models and our bet early on as a company was we believe that open source is going to improve and therefore we're just going to build fine tuning, which is basically The technology will guess the best pixel around your product. So you put in the suitcase and it's going to guess the best pixel around that product. Now, which model we use to guess that pixel generation, we can swap in and out. It doesn't matter to us. And our bet was that over time, these models are going to get better and better and better. And therefore, that image quality is going to go from You know, being AI, you know, obviously AI generated, let's say in 2022 to now looking like hyper realistic and soon probably better than better than better, you know, if it's possible better than a photograph. So yeah, we don't know. Yeah, that's how it works on the image side. Speaker 1: What do you think is the best out there? For just playing around with standalone images right now, is it Dolly? Is it Mid Journey? What is the best one for creating images outside of like your software, if you just want to dabble? Speaker 2: I think Mid Journey, what they have done is amazing because they are about as old as we are in terms of a company and they're obviously far more successful, so it's not really comparison, but they first launched, I think, I want to say mid-2022 was the first mid-journey model and they kind of built, they never raised a single dollar in venture capital and now I think they're kind of, I don't know how many millions they're turning over, but they're incredibly well-known and have the best quality and are super focused in just that element. So I would say, yeah, I think mid-journey kind of the standard barriers, but I think We're going to reach this point very quickly where every model can hit a very high quality and actually is relatively indistinguishable. I think the costs are going to come down. There's so many people working on this that, yeah, I think if people to listen to this, let's say mid-2025, basically everyone's got a photorealistic model and there's like 10 options to choose from. Speaker 1: The big issue seems to be text right now with a lot of these. I mean, I know there's some specific tools that help in that way, but they have a hard time generating text. I can even tell MidJourney or I can tell Dolly, create me an image of an Amazon truck and put the Amazon logo on the side of the truck. Instead of putting the Amazon logo, it puts like, instead of spelling Amazon, A-M-A-Z-O-N, it spells it. AM, OM or something like that. Yeah. What's the challenge there? Do you think that'll get sorted out? Speaker 2: I have no doubt to be sorted out. And I think the challenge is that the models we have at the moment are not, they have no intelligence in any real way. So like the way I I would like to describe this. It's the largest LLM that we have on the market right now, GPT-4. A four-year-old child has seen 50 times as much data. So that's where we are right now, right? These things are absolutely... Speaker 1: Wait a minute, a four-year-old child has experienced 50 times as much data as GPT-4? Yes. Really? Speaker 2: And not just 50 times as much data, but... They've obviously had visionary data, they have auditory data, they have touch, they have interaction from the physical world, whereas the LLM has just seen text. So the four-year-old child has not only seen 50 times as much data, but they've seen, you know, I'm not going to say an infinite, but maybe like 15 different types of data, whereas LLM has seen text and it's seen, you know, less. But these models, like if you look at these graphs, which I'm sure you've seen of like how much bigger they're getting, that's not going to last a lot, right? So where we are at the moment with these models is they are not They have no intelligence. They're basically just predicting the most likely next pixel based on your prompt, based on the training data they've seen, and they don't understand the concept really of like, it's an Amazon truck, and you want the Amazon on the side of the truck. Now, that isn't going to last for long, right? We're going to have models which understand these things conceptually, and Sora was a big leap forward towards this, as you may have seen the video model, because what they're saying about Sora is that it has some understanding of physics, and therefore the videos are better. So these leaps forward that are going to be made, where the models Have a deeper understanding of the stuff they're producing rather than just numerical prediction of probabilities, which is where everything is now, is going to mean that you're going to be pretty much photorealistic and this is not going to take long. I mean, this will be 2025, right? Speaker 1: How are we going to determine in the future what's real and what's not? I mean, it's going to get so, I mean, like you said earlier, the earlier models, when you put a suitcase in the scene, you could kind of tell, oh, that's AI generated. Now you can still, on a lot of stuff, it kind of has that feeling of looking like it's AI, but it's getting sophisticated enough in some cases where you can't actually tell. And some of these tools where you can change, you know, someone's head, on another body and have them talk and move off of a still photo. And like you said, Sora, you can do what, 60 second, I think it is, clips right now that look pretty realistic. How are we going to deal with this, you know, five or 10 years from now, knowing what's real and what's not real? I mean, obviously Meta and all these guys are saying we're going to watermark what's created with our tools. Not everybody does that. Speaker 2: So I'm a huge technophobe in the sense that I'm very positive about this stuff. And what I, What I would say back to people is we've had LLMs live on the market for a year and a half, right? Can you name one serious... One big issue that's actually happened from these things, like everyone said in theory, oh it's going to be terrible because people are going to be able to like make bombs from like, chat teams can tell you how to make bombs and people are going to have fake news and I mean like, let's be, you know, it's been a year and a half, like I don't think you could, I'm sure you could find some, but like realistically, the doomsday scenarios, of this content hasn't come, right? People know they trust sources. I think what will happen is what's happened now. People look at the source, there's sources where they trust and therefore if it's You know, an established newspaper, they trust the source. If it's some random person on Twitter, they don't trust the source. I mean, we've had like, you know, there's many wars going on at the moment. You have these things where people share clips, which is actually from, you know, one war, and they're saying it's from another war, right? So it's fake content, you know, it's not AI generated, but it's fake. And yeah, AI, to some extent, is going to increase that. But I don't think the issue that society faces, to be honest, with AI is like, oh, can people recognize this? There's going to be gatekeepers, as there is now, of this content. And people can choose to believe. People will say, if this person is putting their reputation on the line to promote this, then I believe it. And if this is just some random But that's on the web, then I shouldn't believe it. But I don't think it's going to massively exacerbate anything more. Speaker 1: I mean, you could frame people, especially in politics or public figures, very easily. You know, someone gets divorced and they have a crazy ex-wife or something. Yeah. And that wife, you know, all this Me Too. And she says, oh, 10 years ago, he I don't know, he beat me or something. And she could have AI create something that looks grainy, that looks like it came from 2009 or something. And it looks like it's some video footage from the security camera of her husband doing something and say, look, I found the footage. It was lost on my hard drive. Here's the evidence. And then go after him. That kind of stuff is out there. Or the AI looks so good, That when you get the product, like in your case, it doesn't even look anything, it doesn't look that good. So there's these sides of TikTok. Speaker 2: I think I agree. I agree with the, you know, there's obviously going to, it's going to be easier for anybody to generate fake content of politicians and whatnot. I can see that. I think it's not, it's not hard to manipulate stuff. Like, it's not really hard to manipulate stuff now. We've already seen examples like in the UK, there's this fake recording of the opposition leader bullying someone. So they, you know, they use his voice, they put it through an 11 labs or something like this, and it went viral on Twitter. And yeah, like is in one sense is damaging. In the other sense, like any sensible person, I think what's interesting is people will believe what they want to believe, right? So you see this in these kind of wars nowadays, like on both sides, people will take something and they don't actually care if it's true or not, they will share it if it supports their view. So I think, yeah, I accept there will be like increased polarization, but I think like a sensible person will be able to quickly say like, this has come from a bad source. I think on the product side, I hope it's the case that your content customers, you know, feel like something really exciting and brand new and shiny is coming to them. So yeah, I guess like, I kind of feel like, you know, the genie's out the bottle in this stuff. And like my, I mean, I'm, you know, what am I worried about in AI is kind of like mass, Mass kind of unemployment quickly because people get replaced, you know, we build this super intelligence in the next few years and suddenly jobs are, you know, only 20% of people need to work. So I think I'm much more worried about kind of that kind of societal impact than the content side. Speaker 1: I mean, that happens in any industry though. I mean, anytime there's a big breakthrough, you know, you look at the industrial revolution, you look at, you know, airplanes put trains out of business, a lot of trains out of business. You know, when the airplane started being a way of travel and put cruise ships like the Titanic that were meant for crossing the ocean, you know, basically out of business, not tourist cruise ships. I think AI is a really good thing, but the thing that I see these stats, as popular as it is, as much as you and I are involved with it, and some other people that are listening, most of the population out there has no clue what AI is. They just know it's something artificial and it's something that you can do some fancy stuff, some special effects or something with. They haven't even dabbled with it. I mean, you look at the numbers that have played with it, it's pretty small. Speaker 2: I would say the only difference is the timeline to your example, right? So if you think it took thousands of years to go from a wheel to a car, and then 200 years or whatever it was to go from a car to a plane, and then 60 years from the first Wright Brothers plane to landing on the moon, I think now we're going from, it's going to be like six years, from the first LLM to an intelligent So I have a friend who works in a startup which automates SDR outreach. So they use AI to define your target customer. Write them intelligently hyper-personalized emails and then send that to them. And there's another AI company I know from the YC batch who do phone calls in human voices. So you can follow that up with a phone call. So I think the speed at which this is going to move is not like anything we've seen before. So I agree. Technology happens and it's good but I do think like how fast is this all going to move compared to how fast can people retrain themselves I think is the only difference. Speaker 1: Well I think you said earlier that you believe that on the big LLMs they're adding more and more tokens and more and more data but you think there's a cap to that right? And that's because there's a point of diminishing returns. Speaker 2: That's true. But what I hear from, you know, listening to kind of the leaders of OpenAI is that right now, they're pretty amazed that, you know, there's a direct relationship between how much bigger they make the model and how much more intelligent they are. And I think this kind of makes sense because these models are kind of modeled on like how the human brain works with kind of like neural networks, right? This is the basis of the technology. And at the moment, if you compare the model to the size of a human brain, it's like a tiny fraction of it, right? Like as we said, it's like 50 times smaller than a four-year-old child's brain. So the I think that the element of just adding different training data in terms of images and video and soon like these AI are going to be around in the world in like figure and like in robots. So they're going to have like sensory data as well. Adding that data to the ability of our, you know, Moore's law of just bigger, bigger, more, more compute. I think there's a long way to go. In how much more intelligent these are going to get just by doing what we're doing right now. Speaker 1: Do you think they're going to specialize though? Instead, there's going to be a shakeout and there'll be a few big ones that are kind of like a A know-all, be-all and then there's going to be ones that they're going to narrow down and like there's going to be an LLM that's only knows e-commerce and it's the expert on anything e-commerce but it doesn't know anything about how to take care of your baby or something. But then there's another, you think you're going to see the specialization of all these AIs and where there's very targeted, very specific use ones? Speaker 2: This is very interesting because as I alluded to at the beginning, when we founded e-content, me and my co-founder basically had a choice. We're like, are we going to train The e-commerce model. We're just going to build like the e-commerce model, e-content is going to be like the e-commerce LLM. Or are we going to say, you know what, we don't have the funding and the ability, you know, the team of the kind of general source ones and we're just going to Work on fine tuning and stuff like rag basically meaning rag retrieval went to generation basically stopping hallucinations and making it more reliable and this kind of technical stuff rather than like actually building models and. It's very clear right now that the general models are winning out. And I think this is because the AI works in unexpected ways. So if you basically give a lot of people, these kind of extremely smart people, lots of compute and say, hey, here's more compute, here's more data, like go try a new model, like see where you come up, but you're going to come up with like Sora and it's going to be And, you know, these are totally generalized and it's moved at the moment like all the big serious companies who are building these foundational models like OpenAI and Minstrel and, you know, Claude's parent company, I forget their name. All these guys are building just generalised... Anthropic, yes. They're building generalised models. Could I see a world where it becomes specialised? Yes, but I do believe that these generalised models just have so much more space to go in terms of the data they're using, in terms of the data... and the size that they can get to. And I really believe we are so at the beginning of this still that give it a few more years of building these large, all-encompassing models, and we're going to have something which is just so incredible. And actually, do you need to build it? Does e-content need to build its own base model? I see a world where we enable customers to switch to different models eventually. So we say to a customer, hey, you want to have something which is Claude. Claude has three different models at all different prices. They have different qualities. More human sounding, whatever, like, basically say, you know, you go on, you pick your model, we'll switch different models and you can generate with the model you want and you pay the different price because they all have different, they all price slightly differently. So that's how I kind of see us going, which is you, you know, at the moment, we're making the decisions for the customer. We're choosing the model that we think is best. It's not always going to be best in every situation, but I think easily trading between different ones is our future. In the long term, yeah, I can imagine fine. But it's a different thing between building a base model and fine-tuning, right? I think people already fine-tuned. We fine-tuned for e-commerce. We fine-tuned for photorealistic rather than cartoons or anime or any other type of image generation. And we fine-tuned for conversion, what works best, the most exciting images you can generate. So I think that's more of a question of fine-tuning. Speaker 1: If I go to MidJourney though, and I go into Discord, open up a chat, and I know how to do prompting, and I say I want a photorealistic photo, and I want it slightly angled up to give this perspective of a size, and I put in my prompts some of the psychological things of marketing, And I tell it to create an image. How is that different than what your tool does? Because if your tool is sitting on top of a model, and are you basically hand-holding them? So you're assuming that people don't know how to prompt it, right? So you're adding in these safety rails and these safeguards that actually will generate that? Or what is the true, why should I use your software versus just if I'm a good prompter going and using one of the other tools that's out there? Just directly. Speaker 2: I think yeah. I think what we discovered in the last two years is that our real value to customers is speed and scale. So with e-content, you can generate the entire listing. This is text, infographics and images. You can do it basically all in one click and you can integrate directly into Amazon and you can publish to Amazon. And soon it's not going to be just Amazon, but you're going to be able to publish to eBay, Walmart. You're basically going to have one input. Speaker 1: How does that work? So I have my new product. I have my new suitcase. I just take a picture of it or my factory sends me a picture. And then I load it and say, go make me a listing, what do I got to do? Speaker 2: You connect your Amazon account, so you connect via our API. So then we have all your Amazon account, we have all your data in your brand analytics, we know all your keywords, so we kind of have... Speaker 1: But I haven't sold this product before. Speaker 2: At the moment, you'd have to create the new listing on Amazon. So the base listing has to be there on Amazon. We are launching, to time stamp this, we will have launched by now a kind of a CSV input. So you can start with a CSV. Speaker 1: So you've got to have a basic shell of a listing of some sort. Yeah, so you could bring in a CSV from your manufacturer on Amazon. Speaker 2: Exactly. And then you upload that to your content and then we generate, optimize the whole listing. Speaker 1: Little graphics and everything. Speaker 2: And you can use e-content to, like the first time you may have tried it, you can still use e-content to prompt and generate an image. And the difference between us and Midjourney in that respect is that our models take your input image, let's say like the suitcase, and it's going to predict the next pixel around your specific product. You know, interesting story in this, like when we first launched, we were training models for each product. So you would have, you know, we would have like a suitcase model and we would train the model to always replicate that suitcase and then you could type in something and generate the suitcase in any scenario. We found that And this may change in the next one or two years, but we found that the hallucinations were still not, like it still had hallucinations in these models when you did that. So therefore now, we are simply predicting their pixels around, we're painting the image from the starting point of your suitcase, which is not what Midjourney is doing. Midjourney is kind of creating a, like it's creating from white noise, right? It's not creating from anything. Speaker 1: You know what would be cool is, and maybe this is something you guys are working on, is if I could take, if I want to get in the suitcase business, And i'm like okay i think i want to suitcase business but i'm not sure how i'm gonna differentiate what i need to do here's a picture of a couple suitcases i have ideas on. AI, go out there and read all the other suitcases on Amazon, read all the reviews, look at all the pictures and create me 10 prototypes of something that would be unique and different with a sample listing for each one. And then let me use my human brain to go, that looks cool. I think that will work. Do you see something like that coming when it comes to product research and product development? That's a very simplistic way of saying it, but it's pretty cool. Speaker 2: There's a company called Pietra who you can do text-to-product. So you can literally type in, I want a suitcase, blah, blah, blah, and it will AI generate a sample product for you, and then you can actually manufacture that product. So it's a really interesting space. I guess in the long, long term, as we look to expand, I think it could be cool that we incorporate something like that. Speaker 1: What about people that lean too heavily on AI? English is not their first language and so they assume that everything is correct on a listing, everything is written in a good way versus you and I that are native English speakers can look at it and go, that doesn't quite sound right or look right, but they might not catch that because they're not native speakers. How do you deal with those kinds of things? It's not really a hallucination, it's just, you just know it's not right. Speaker 2: So I think there's like two steps we take beyond using a ChatGPT to create a listing and there's lots of amazing listing like GPTs, which if you have like a subscription, you can get like all these listing GPTs for free and you know, some of them are created by people in the community and are great. But we kind of, we take two things beyond that, which is we've done fine-tuning and we've done RAC. So fine-tuning means that we've trained the model on high-performing Amazon listings, like it's not just You know, putting in a prompt a few examples of what a good listing looks like, but it's actually like a machine learning process to fine-tune their model to... Speaker 1: What defines a high-performing listing on Amazon? What do you look for? Speaker 2: Something which has got, you know, well, something which is on page one, has good conversion rates, a listing which is performing well in terms of, you know, click-through and purchase and that kind of stuff. Speaker 1: How do you know that? Speaker 2: Because more times than not, they're on the page one of Amazon and they're selling well. I mean, you can see data. Speaker 1: How do you know what their conversion rate is and what they're, that they're not, that's not just outside traffic that they're pushing from TikTok and that's why they're there? Speaker 2: It's a fair question. Like we kind of took a, you know, you have a number of these tools which you can rely, you know, you can get certain estimated metrics from which we used. You take you know these listings and you fine-tune the model and therefore it kind of it understands the The phrasing and the stylistic notes of what looks good. And then you do something called RAG, which is Retrieval Augmented Generation. And basically, this is where you query an external database to fetch relevant data. So the relevant data would be keywords, right, from your brand analytics. The relevant data is Amazon's prohibited lists. Stuff that you just can't say that's going to get you blocked. So you have fine-tuning and then you also have the reg step. And these are kind of two additional steps which help AI to create good listings. I would warn people against using If they're good at prompting, you can get something good out of ChatGPT, but the problem is, as you say, you need to be good at prompting because it's very balanced in everything it says. Any question that you give to ChatGPT, it's always going to give you a watery two-sided corporate BS answer. That's just the nature of the model that they've trained. In your prompting, be able to peel that away, which takes, you know, like, it doesn't take much experience, but you need to know that you need to do that, because if you don't do that, it's just going to sound kind of quite blur and dull. And also, it will do all those weird things AI does, which it goes, that's when this comes in, and it kind of, you know, opens, has this weird phrasing where it starts, ends a sentence of a question, and then it answers the next one, which no human would normally do if they're writing something. Speaker 1: How do you feel about the e-commerce sellers that are not really paying attention to AI right now? They're like, eh, I don't got time for that. I don't think this is going to be around. This is just this shiny object and I've seen some stuff it does. It looks like crap. So I'm just going to keep doing my thing. I'm making good money. So why change? Speaker 2: I think it's so transformational in where it's going. And I know it's a lot of hyperbole and everyone's going, oh, this is crazy. Fundamentally, this technology is unlike anything we've developed before. And people say, oh, AI is a tool. It's not a tool, because a tool is under the influence of a human, like a hammer. A hammer doesn't do anything that a human doesn't want it to do. AI is Unpredictable, it's evolving, it's seen far more data than any one individual has ever seen. It's read all of the data on the open internet. I posted back in April an interesting TED talk that the Microsoft AI CEO put out I was saying that AI is a new species, and that's what it is, right? That's his argument, that's not my words. I don't know if I agree with that to that extent, but I definitely agree that this is not some phase or some like, this is a fundamental shift in human history. And to me, it's like, you know, when humans create a language and like that completely transformed our evolutionary trajectory, I think this is another, like AI, in my view, is is on that level, right? So to just ignore it and say, oh, this is just a fad is really misunderstanding the point. And I completely accept we're at the infancy stage. I remember showing you e-content, let's say, a year and a half ago, whenever we were in person in Silicon. And yeah, it was OK. It wasn't great. But the speed that this is moving is just insane. And you compare I take MidJourney again. What MidJourney v1 looks like versus MidJourney v6, these were two years apart, right? So yeah, these are two years apart and literally you kind of see a jumbled mess of colors for the same prompt that now gives you like an ultimately photorealistic Picture, and then that's two years, and then you fast forward six months, you're going to not be a picture, but you're going to be a hyper photorealistic video from Sora. And you fast forward three months, and then one month, so the speed at which this is going is wild. So I would say, not even from an e-commerce point of view, but just from a humanity point of view, this is not something to kind of Ignore, you know. Speaker 1: What should e-commerce sellers be doing right now when it comes to AI? Should they be experimenting with tools like yours or with software like yours and trying to improve their listing? What should they be doing? Speaker 2: I think the future is for the entrepreneurs and I think it's really exciting and we're really starting to see this with You know, the layoffs in Amazon and like all of these, Google, everyone's doing layoffs and consistent amount of layoffs. In my mind, this is going to accelerate because you can get so much more done with a smart AI system and I think the future of e-commerce is fantastic because you're going to be the CEO or leader of your company and you're going to be able to have an extremely smart assistant to generate your listings across e-content. You're going to have another one which is going to be able to help you to do all your product development and everything else. Like all that stuff and you're going to have is and maybe the AI is actually in the factory so robots are there helping to design the product as well. I think we're going to be moving to a world where we all have like a chief of staff, which is like an AI bot, like an inflection AI, which is kind of helping us to manage our calendar and our time and is helping us to write all of our key emails and we're discussing with this AI our strategy because it knows us better than anyone else and it's trained on everything that you've seen and it's 20 times or 100 times as intelligent as you are. So I think I think it's going to be so much more than just helping you to create good listings on Amazon. We're at the early stages, so I think e-content is a good place to start. Other kind of AI companies doing stuff. I think the challenge that we've had over the two years is working out how to apply what we have now. There's a huge exciting future, which we want to build and be part of building, but also we want to deliver value to people now. And that's kind of the line that we've been working, exploring, and now I think delivering on. But I would say, yeah, if you're not If you're an e-commerce alien, you're not going to have this incredible AI assistant, and your competition is, and that AI assistant has been trained in all of Amazon, and it understands how to do seller queries, you're going to be out of business, right? So I think it's definitely something to be throwing yourself into. I think understanding how these models work is so important, the basics of it. I think that honestly, to me, is the best place to start. What even is this technology, which is It's going to be defining all of our lifetimes, right? What is it, right? I think that's where I'd start. Speaker 1: There's tons of software that comes out every day. There's an AI for that. So you have a lot of competitors in your space that are doing similar things. A lot of them are a lot worse than what you're doing. Some of them may have a feature that you don't have. So how do you stay on the cutting edge and how are you going to stay as a leader when it comes to content AI assistance for e-commerce sellers? Speaker 2: I think our competitive advantage, well, first of all, we have a patent. Second of all, we've raised a bit of money now. I think there's a whole host of people who are creating a stable diffusion or whatever. They're kind of plugging in a beautiful UX in front of an open source model. And we are, you know, we're two years deeper than that and we've got a team and we've got, so like, I'm almost like, from a marketing perspective, like, maybe people try this stuff, but fundamentally, like, I don't think we're going to lose any customers to like that. I think when you go to the more established companies, and especially where we want to be going, like, you know, we're going after the established companies, right? Like, we're not, you know, the big boys in the space. I think, you know, we like, I talk, I try and talk to customer every day, at least one or two, we really want to understand the The pains that they face, what value they're looking from us, like why did they sign up to us? What were they hoping for? How can we deliver that with what is available now? And that's really what we'll focus on. And I think, circling back to the start of the interview with Jeff Bezos, the customer obsession was that's something that I really learned from working at Amazon, like really putting the customer first and understanding it from their perspective, and that is kind of the ethos that we have. And we don't, you know, I get it. I don't get it. Annoyed is the wrong word, but I'm still on my team when they come up with random ideas for something which isn't driven by a customer interview or customer discussion. I don't want that in the company. I want customers who are using it actively to be like, oh, this is something which would be great, and that's where we focus the resource. Yeah, I hope that's our strategy and I hope it plays out. Speaker 1: Imkan, how does it work? Do I pay a per usage fee or is it a monthly subscription? Or how does it work? Speaker 2: It's an instant question. I think in the long term, all of these AI tools are going to move to a kind of a service fee for the provider, let's say e-content, for basically the UX and the painted technology, whatever they provide. And then on top of that, people are going to be paying for the tokens on OpenAI or So I think that's a future business model that people will get, not just at us, but everyone is going to have to get used to because these models are expensive and they're variable in price. And I think people are going to have to get used to that pay-for-usage. But right now, as of today, we are a simple subscription. We give people unlimited Yeah, we're a subscription model. For a $165 a month package, you get 100 SKUs a month, unlimited generations on that. For $465 a month, you get 2,000 unlimited generations on that, like images, text, infographics. And we'll stay at that level as long as possible. I don't want to be I don't want to be innovating in, let's say, the pricing space. I don't want to be innovating on the e-commerce side, but my prediction is that in order to make the economics work, that will be how the world looks like in the future, just because right now, you've got venture money, you can burn money, but at some point, it needs to become a sustainable business. Speaker 1: Do you have a quick example of someone that was doing pretty good, started using your service, and changed up their listings, and now they're crushing? Speaker 2: We have a seller called Silly Slick. They're a knife seller and using your content, we got them onto page one for the keyword of, what is it? Titanium knives or something like this. We've got them on page one. They've generated a ton of beautiful lifestyle images and infographics and the rest of it on our tool. So if you go to the case studies on the website, you'll see the one I'm referring to. But we have six or seven case studies there and I'm always trying to build more in-depth ones out. So we have some testimonials, some other examples that click back to the listing. Speaker 1: Where do their sales go from or their conversion? What's some hard numbers? Speaker 2: So the hard numbers is we did some A-B testing on, I mean, the challenge of conversion is that it can be driven by lots of factors, including price, price of their competitors, time of year. But the hard numbers are we did A-B testing on product opinion of their previous listing and our listing, AI-generated, and 88% of their target audience preferred our listing. And that was, you know, high earners, prime members, you know, all the rest of it. So I think for now, We are working on dashboard. Well, probably by the time this is out, we'll have dashboards. We have them internally of what's happening with the conversion and that's how we monitor the AI and we want to publish that to the customers so they can see that as well. But having worked at Amazon, I know the challenges of attribution very well and I don't want to We've got one case study which says we increased conversion by 30%. So there's a number, but we know that 30% can be driven by a lot of factors. So I want to be very scientific with these. I guess the split testing is the easiest and fairest way. Some of our own customers, some of our enterprise customers are doing are doing their own testing and I really hope they let us share it. We're under NDA so maybe they won't but they've been using us to generate the content and they're doing this research as part of the PAC so I'm hoping we can publish that because that will be the best case study actually because it's come internally from them rather than us. Speaker 1: Well Max if someone wants to actually check out your tool, can you spell that for them and tell them how to go check it out? Speaker 2: So it's a very unimaginatively named e-content from e-commerce content is the idea behind that. So it's E-C-O-M-T-E-N-T dot AI. But yeah, I'm quite active on LinkedIn and Twitter, Max Sinclair, so you can find me there and you can message me. Tweet me, whatever, and I'm happy to jump on a demo. I love talking about this stuff. One of the reasons I love talking to customers is, I know you say, Kevin, that a lot of people aren't using AI. All the sellers I talk to love AI. It's very strange for me that... I always hear Joe, my podcast co-host, saying, sellers don't like AI. But from my perspective, I must talk to 10, 15 sellers a week and they love it. I'm certainly not seeing that side of it personally, but I believe you. You know a lot better than me when you say them. Speaker 1: The ones that are using it love it and the ones that aren't just haven't done their head around it yet. It's been a great talk to you. I can sit here and talk AI for hours as well and go down all kinds of rabbit holes, but I appreciate you coming on on the AM PM Podcast. Speaker 2: Thank you. Thanks so much for having me, Kevin. Speaker 1: If you're paying attention to what's happening in AI, it's going to dramatically affect everything we do from the employees we hire, to the way we create our products, to the way we create our listings, to the way we sell, to the way we interact with customers. And Max's company is just one aspect of that. But as you can see from our discussion, AI is world changing in what it's going to be doing and it's quickly, quickly evolving. So I hope you're staying on top of what's happening in AI. If not, you can get a piece of it from my newsletter, BillionDollarSellers, BillionDollarSellers.com. Every Monday and Thursday a new issue comes out, but I cover some of the latest when it comes to AI and e-commerce in there as well. So check that out if you're not already subscribed. And don't forget to check out my upcoming Market Masters. It's limited to like eight people. It's going to be happening September 12th to the 16th in Austin. You can actually check it out at BillionDollarSellerSummit.com. BillionDollarSellerSummit.com. Choose the option for Market Masters and you might be able to spend about three hours in front of the Dream 100 where they actually zero in on your specific business and your specific growth pain points for what you're doing in your e-com business and help you almost one-on-one with some of the most talented people in the world. It's called Market Masters. It's happening in September in Austin. We'll be back again next week with another awesome episode. But in the meantime, remember, whatever you tune into is what you turn into. Whatever you tune into is what you turn into. See you again next week.