Latest / Talk Commerce / ChatGPT Shopping and Agentic Commerce for Multi-Channel Sellers with Jorrit Steinz of ChannelEngine
Transcript
- Brent Peterson (00:00.962) Welcome to this episode of Talk Commerce. Today I have Jarrett Steincy as the CEO of Channel Engine. And by the way, the best stroopwafels you can get at any conference. Jarrett, go ahead and give us a better introduction. Tell us your day-to-day role and one of your passions. Jorrit Steinz (00:18.957) Yes, thanks a lot. I'm the founder and CEO of Channel Engine. Marketplace integration, nowadays multi-channel integration solution and management suite. So we connect more than 1300 marketplaces worldwide. Think of Amazon, Walmart, also Zolando and many others. We connect TikTok shop, but also that's going to be one of the topics of this podcast. Agentic commerce is coming. So one of my passions e-commerce for long time otherwise I would have quit. I you must love it otherwise you're gonna hate it. And one of the other passions is outdoor sports whether it's skydiving, snowboarding, mountain biking just to clear the mind and enjoy nature. Brent Peterson (01:01.91) That's awesome. Yeah, you're my second kite surfing, think, in this in at least two weeks. So I haven't tried kite surfing. might be a little bit too old for it now, but I should never say that, right? been I've been trying to. Yes, it is definitely a signal. I've been trying to learn how to surf. So that's that's that's a lot of challenge. Just just that part. So. Jorrit Steinz (01:12.965) No you're not. It's like this is a signal. Jorrit Steinz (01:22.159) That's a good start. Brent Peterson (01:27.406) That's great. All right, so before we do get started and we're going to talk about agentic and I'm really looking forward to talking about that today. But I do this thing called the free joke project. So I'm going to tell you a joke. You just give me a rating eight through 13 and I kind of wish I would have brought I've found a strew waffle joke, but I don't have one. But anyway, here we go. One spelling mistake can ruin your marriage. I accidentally texted my wife. I'm having a great time. I wish you were her. Jorrit Steinz (01:43.621) Thanks. Jorrit Steinz (01:55.301) you Yeah, politically correct that would be an 8, but I think it's funny, that's a 12. Brent Peterson (02:02.702) Good. Yeah. mean, nobody talks about politically correct. In fact, I just wrote an art. I do a lot of vibe coding with Claude and, you know, I've been to so many conferences in Europe and the Dutch and the Germans are always so liberal and using language that maybe the US or Americans wouldn't use in their vernacular. I think it's super fun and interesting to talk about what is politically correct in speaking and what does it mean when you're speaking English. But that's another complete topic. Let's talk a little bit about how people connect and how important it is now to connect all these different channels into one spot. And then let's kind of loop in each agentic on that kind of give us the 10,000 foot view of channel agent and how it solves a lot of problems. Jorrit Steinz (02:49.935) Yeah, so I think over the past five, six, seven years, and even before that, marketplace came up with Amazon being first and eBay popping up, opening up for typically smaller sellers that came on board or retailers started selling on those marketplaces as a third party seller. But it shifted more and more to brands that start selling directly on the marketplaces to either backfill or augment their collection where Amazon didn't buy everything from their assortment. And they have the opportunity to sell directly to the consumers and that started expanding Because of a lot of retailers are now also opening up Marketplace as well. You see a lot of retailers didn't want to have that many products on the balance sheet So they're opening up a marketplace and that means an opportunity for a brand but also a challenge because Not one marketplace and your own DTC shop, but it's two or three or five We have customers doing 60 or 70 marketplace with a small team And that means you need to start synchronizing. Get the product information out there and they have different requirements of titles, descriptions, images, attributes. And that's where the journey starts because then the marketplace needs to know how many items you have in stock. When you sell something on Walmart, you need to take it off of Amazon. You need to take it off your Shopify store. then orders needs to be shipped. So all that synchronization and management is done in Channel Engine. And we have customers using us worldwide. North America, Latin America, Europe, Middle East, South East Asia at scale. So it's getting more more important. And then you see all social commerce becoming transactional. And then the AI agents are coming up. How do you sell there and make sure if a transaction takes place that also is fulfilled through your own data center or warehouse. Brent Peterson (04:39.0) Yeah, and it really is becoming then every agent is potentially a purchaser or a channel in which you could buy. And I know that I've seen perplexity has cards already that would allow you to go through. I think JetGPT has big aspirations of just letting you buy in app. And I'm sure that you can buy an app in a lot of these things already. talk about how the channels now are really Maybe not taking away from your website, allowing you to have more opportunities to find clients, right? Jorrit Steinz (05:13.465) Yes, I think especially in the discovery phase I'm using it as well. In the discovery phase comparing certain products you can use natural language and that helps. In Chachiopee, Perplexity, Gemini you will get recommendations including reviews and you can talk to the agent and then ideally it also knows the best price, which shop is selling it, which brand is selling it and do they have inventory and can they ship to your location. It's all coming together. Now most of these LLM agents are scraping the internet. Sometimes it's outdated information is getting better and better at every time. But it's all opening up now. Replexly was the first but ChetGPT did it at scale with the Agenda Commerce Protocol. Where can actually feed the information into ChetGPT with a combination of data feeds. And then there's a connection to the APIs of the seller. So we are connected to... these integrations for a chat GPT, co-pilot is going to launch and others will follow suit where you actually transact within the chat interface. Then a payment is created in Stripe or a Gen or PayPal and the authentication token can go to us and we can verify that it's actually paid and then it's just like a marketplace where it can be shipped. I think there's one distinct difference between a marketplace like Amazon and the how the chat agents approach it. is that the payment is going on Amazon is going to Amazon themselves and they pay it out without their fees and with the Chetjipati co-pilot etc they are using the merchant's payment account it's a Stripe, a Gen or other payments account so the money directly goes to the merchant and then they get an invoice for the referral fee Brent Peterson (07:02.958) You did. So one of things you mentioned is that kind of feed, right? And you do both sides. You pull in the information, but you also push it, right? And I think the concept of how you're going to push data into, say, chat GPT is it's not complete yet. Give us some background on that. Is there a fee that you can already push into chat GPT that merchants should already be looking at? Jorrit Steinz (07:25.581) Yes, it's ready on our site, but JGPT needs to open it up. So the direct checkout, they are still in their pilot phase with Etsy. I know the first shops on Shopify are integrated, but they open sourced the integration protocol. So we know exactly what the definitions are. And a lot of people forget typically data feed, the product feed is title images attributes, but there's way more context that any platform needs is do you have it in stock, where can you ship it, what is the price and which currency for that region. So more debt awareness is also fed into the data feeds and then there needs to be the connection to the payment provider to complete the circle. And then once this is all live these chat interfaces need to compete for stock versus all the other platforms. So if you have inventory I don't know 100 items keywords in stock. You can sell it on your own DTC shop, on Amazon, on Walmart, on Nordstrom, Target, well as ChetGPT, as well as Copilot. And then depending on where somebody orders, we need to update the stock on all these different platforms otherwise you're gonna oversell. So that's something to keep in mind. And then also strategically, if you only have one or two items left, are you gonna reserve it for the LLMs or are you gonna reserve it just for your own DTC site? So where do you expect the consumer to be? The new consumer that you want to win, that you want to get into your post purchase funnel to be able to retarget. So it's a strategic bet, but very much an operational thing as well. Brent Peterson (09:06.892) Yeah, and the complexity just has gone through the roof, right? Just in making sure that you're managing all those different, even inventory levels. I interviewed Sharon Gee from BigCommerce or from Commerce, they call it nowadays. And maybe you should, if you should consider just renaming your company channel or engine, right? But that's a different topic. One thing she said is for LLMs to discover you, you have to have so much more data and context around your product. Jorrit Steinz (09:09.859) Yeah. Jorrit Steinz (09:19.535) Yep. Yep. Jorrit Steinz (09:26.882) Thank Jorrit Steinz (09:36.089) Yes. Brent Peterson (09:37.196) Are you seeing now merchants going to and adding much more data to every single product? Jorrit Steinz (09:45.39) Absolutely, for some it's hard because they have hundreds of thousands of SKUs. For some brands it's easier and you see that for LLMs to give a good recommendation questions and answers are essential. Not only to directly feed in the LLMs but also if you sell on your own DTC shop because LLMs will index your site. So the better you can answer the questions for your product the more likely you are to float up into the conversations. And of course the reviews are a big part of that as well. Brent Peterson (10:19.286) Yeah, are you seeing the ability, I think one shortcut would be to use your existing data to create more questions and answers on your data and use that and then maybe have a human. We'll always have a human go back and review it later because it does get a little crazy, but is that one of the methods you're seeing merchants do? Jorrit Steinz (10:40.633) Yes, absolutely. And you can look at returns for instance. What are they saying? Look at reviews. Reviews are something of a very valuable part because those are the typical things I mentioned that are not in the product description yet. You can extract data from the images. You see you can add a lot more context, brand related context that is applicable to all the different items. Whether a brand is fully sustainable. put it in the product description as well. That it's fed into your DTC shop, marketplaces. So that's feeding a lot. And like you mentioned, there's a lot of AI that is helping product descriptions. Titles, descriptions, be careful with attributes because AI, although the number of hallucinations is smaller, always check for hallucinations. For instance, battery life on your watch. If it says three weeks and it's actually three days, be careful with that. easy way to circumvent that is have two or three specifically three LLMs to the same attribute and then see if there is an overlap and if it's fully overlapped is fine because the chance of all three hallucinating is zero. Brent Peterson (11:55.17) Yeah, well, it's not quite zero, but it could be zero. The other method I've heard many of the new AI or AI SAS platforms doing is having gateways in which you'd have a check and a check and a check. So maybe you'd go through run iteration, and then it would go back and check to make sure that the data matches what you're saying and making sure that it goes through there. The question answer is that do you From the LLM discovery standpoint, and this is probably not really in your wheelhouse, but where it lands on the page, does it matter for elements at bottom of the page or the top of the page? Jorrit Steinz (12:36.291) In the LLMs the ranking we don't know yet. Because it definitely depends also on the user context. Because if I'm using it, I'm using it already to discover information of what I have to buy. And it also takes a lot of things into account it knows about me. So there's already some sort of personalization. I heard an interview with DSC from Amazon saying well there's no personalization. Well it is, because Chetchipi knows a hell of a lot about people. So there's personalization already. I don't know yet how we can influence with adspand and seller, sales velocity and all the stuff we know from marketplaces. So that's to be discovered. The only thing you need to know is are your products there with enough rich content as possible. Brent Peterson (13:18.104) Yeah, and I... Brent Peterson (13:24.67) So we kind of started off with agentic and I think the ultimate vision of agentic is that the client will then, the consumer will deploy an agent that would scrape the internet for where they want to buy and then come back and say, hey, I found these three things. Would you like to buy this? Or even at some point, and I think in the B2B world, at some point you're going to deploy an agent who's going to go and then shop at three different places and then buy something for. Jorrit Steinz (13:48.453) Go do it! Brent Peterson (13:52.076) for the, to procure it, right? So I think that's where we're going. Have you seen anybody doing that yet? Jorrit Steinz (13:58.554) No, not at this scale. So you've seen Pilots, I know Google. So Gemini, instead of having to check out there where they send it in through an API, they're actually shopping for you on the shops. They're not to keep on doing it. So that's one option. But I think there will be more and more of these agents, especially for recurring, repeating purchases, agents will be able to buy it for you. my thesis would be that the agent will not have to do it on the shop front ends itself because they don't need it so it will be probably through a gender commerce protocol or an ACP so different protocols to to get that going without needing the front end that has had a lot of time with UX and UI they don't need it they just need the data and the transaction Brent Peterson (14:50.668) Yeah, do you think, so you did mention you don't really need the frontend, but I think that a lot of this is going towards an API feed to an LLM rather than scraping it, right? That makes it much more efficient for the internet even. Yeah, and I think that now it's up to the platforms themselves to give endpoints specifically for LLMs to find them. Jorrit Steinz (15:01.241) Yes, absolutely. And that's prepared already. Brent Peterson (15:16.678) Do you see maybe search going that way too and Google looking at API endpoints as a place to discover items rather than simply scraping the front end of the internet? Jorrit Steinz (15:27.235) Yeah, the fun thing is over the years, you've had APIs, but for product content, the most used is data feeds, whether it's an XML, JSON, CSV, that gets updated every 15 minutes, an hour, depending on how often is needed. And over the years also Google Merchant Center has drawn in all these product feeds. Bing has done so. All the comparison sites, it's very common practice. And you see that happening with the LLMs as well. So at Chetchupt you can look up the specifics of the data feeds they're expecting. So it's also a product feed, which is pretty old technology, but it works. One advantage of those product feeds is they don't need that structure in terms of mapping it to their taxonomy, their attribute fields, like any marketplace, because it's AI. They'll figure it out. Brent Peterson (16:20.8) Yeah, that's a really good point. I gave a talk at a conference a couple of years ago just around the idea that most merchants are still just putting, say, 10 bullet points on their product with the price and that's what you buy, right? I think consumers are, we're okay with that, but that's not okay at all anymore. Maybe we just kind of finalize our conversation the day around. the need to always enrich your data and always be asking, does my data have enough? And maybe a scale in which you'd look and say, what is the minimum amount of data that I need, right? On a product page. Jorrit Steinz (16:58.469) Absolutely. And I think attributes play a very important role. Battery life, diameter, processing speed, all those kinds of things. And are typically mapped towards marketplace where you use those filters on the left. But still these attributes matter for LLMs as well. So you need to provide some more context. But I would say it's the same as trying to answer as many questions as a person would ask in a physical store to a salesperson. at least an advisor not just someone working at the store and really in depth knowing the product. Where would you use it? How does it compare to others? Why is it better? And I think all these things need to be in the product content page so there will be way richer content than it was before. So that's it's gonna be an interesting challenge to do that at scale. That's where AI tools can help out but then you need to validate that it's correct. Brent Peterson (17:59.097) Yeah, and that should be sort of work like the parameter inside of an LLM where it's a connection point in which you can get data. And then there's a couple of different types of attributes. There's mapped attributes like that would map to just a single drop down item or you'd have a text based attribute. I know if you looked at a PIM, the hardest thing to try to manage is text based attributes. But I also think now LLMs are gonna make text-based attributes easier to manage in the future, Jorrit Steinz (18:30.787) Yes, and we do it as well. So in General Engine, we've got a lot of text based attributes coming in. And they can be different languages with slashes and different materials, for instance, and the marketplace have their own requirements and mapping between the two. use AI for that. It makes it much easier. And we just get a recommendation. All you have to do is validate and click on OK. So that saves a lot of annoying work to do. And going forward, we want for some customers to enable them to just say, just do it on autopilot. I verified the first, I don't know, 100 or 10. I'm okay with it. I trust it. Brent Peterson (19:06.894) Do you think that maybe the more important things in your tech stack is going to be your PIM, your product information management system, rather than your e-commerce front end? Jorrit Steinz (19:24.335) Yeah, of course I'm going to say that the marketplace integration solution is the most important one, but that's definitely that needs the data coming out of the PIM system. So you need both. So channel engine gets a product syndication, but without having the product information, there's not much to syndicate. And only having it your PIM, not getting it out there is also a challenge. Brent Peterson (19:42.082) Got it. Brent Peterson (19:47.35) Yeah, yeah, exactly. So Jarrett, we do have a few minutes left. As we close out the podcast, I give everybody a chance to do a shameless plug about anything they'd like. What would you like to plug today? Jorrit Steinz (19:58.245) Yeah, so if you're serious about e-commerce and the changing landscape, whether you're big company that needs to adjust quickly and doesn't want to rely on purely ERP and BIM system, you need a flexible solution. That also is enterprise-grade, but allows you to pop on new marketplace, test it out, be on the agenda commerce as quick as possible without disrupting all your operational flows, then give our team a call. And also the last... blog is got a lot of insightful information, great guides, ebooks, research reports that are interesting for anyone to download from our website. Brent Peterson (20:40.514) That's perfect. I'll make sure we get it's channel engine.com right? We make sure we get all those all those into the show notes. George Stein. Either one right? Alright, perfect. George Stanks is the founder and CEO of Channel Legend. Thank you so much for being here today. Jorrit Steinz (20:44.549) Correct. Jorrit Steinz (20:48.313) or channelengine.ai, you'll end up at the same place. Jorrit Steinz (20:59.088) Thanks a Brent.