Prompt, done, sell: AI hacks for your online store
AI writes texts, improves images, answers emails, and has long since become part of our everyday work life. But how do you use it concretely in your eBay shop without getting lost in 40,000 tools?
Answers come from Creative Technologist Max Mundhenke, who recently launched his own podcast „Kollegin KI“. In it, he talks with people who use AI in their professional everyday lives.
In this episode, we take a very practical look at how AI can support you with your daily to-dos. Max explains why you should always start with your own use cases, how to use AI as a sparring partner, and why “Human in the Loop” remains so important in e-commerce.
If you want to find out how AI can save you time, automate routines, and boost your creativity, this episode is perfect!
PS: With this episode, we're heading into the winter break and will be back on January 28, 2026 with exciting new episodes!
**Important Links**
- Kollegin KI podcast by Max Mundheke (https://open.spotify.com/show/6PtU35BUBRCtyHbAN3zDHB)
- eBay for Business on Facebook (https://www.facebook.com/ebayforbusiness.de/)
- eBay for Business on Instagram (https://www.instagram.com/ebayforbusiness_deutschland/)
- More info on the eBay podcast „Alles top. Gerne wieder!“ and our podcast hosts (https://www.ebay.de/verkaeuferportal/news/podcast)
** Chapter markers **
- 00:00:00 to 00:03:47: Intro
- 00:03:47 to 00:07:56: How Max became a Creative Technologist
- 00:07:56 to 00:15:39: How can AI simplify my workday and which tools are important?
- 00:15:39 to 00:20:22: Why are processes and SEO currently more important than GEO?
- 00:20:22 to 00:26:38: Custom GPTs, AI agents & automation
- 00:26:38 to 00:31:19: Internal resources & external help: embedding AI in the company
- 00:31:19 to 00:35:31: Outlook for the future & concrete first steps after this episode
Lisa: AI can write texts, improve images, or answer emails. That means it is no longer a promise of the future, but has long been part of everyday work life. The question is not whether, but how we use it to become faster, more creative, and more efficient. In this episode, it's all about the smart use of AI. Which tools save time? Which "doings" can you automate in your eBay shop even without major AI expertise? And how do you recognize the right game-changing tools? You'll find out now.
Max: AI is being used much more in communication. That sometimes works better, sometimes worse, to be fair. There will certainly be much more efficient supply chains, because AI can "predict" a whole lot and can handle large data sets much better than humans. And I believe that if we use AI correctly, there will also be far fewer careless mistakes. Because AI doesn't make careless mistakes. That's a human problem.
Lisa: A warm welcome to "Alles top. Gerne wieder!" the eBay podcast about retail and e-commerce. I'm Lisa Haag and today we're talking about a topic that probably comes up in every second meeting right now: artificial intelligence. But instead of just talking about buzzwords, we're taking a very practical look at how AI can support you with your "to dos" and why looking at your own use cases is the best way to find the right tools. To do that, I've invited a guest who works on exactly that. Max Mundhenke is a Creative Technologist, consultant, and host of the OMR podcast "Kollegin KI". He originally comes from the creative industry, worked as a copywriter, author, and strategy consultant, and today he'll show you how to move your business forward with AI. Hi Max.
Max: Hi Lisa.
Lisa: Are you happy to be here?
Max: Yes, very much. I'm not exactly sure whether I can keep the promise I can give everyone here about how to move their business forward. But I can certainly give a few pointers.
Lisa: We always have to open the scissors very wide and then see how we deliver in the end. Max, you and I know each other pretty well. I think that's worth saying up front. We also know each other outside the podcast. That means if we interrupt each other a little, then that's a sign of our friendship. And not rudeness.
Max: Interrupt me anytime.
Lisa: But we can give our sellers a little introduction to you. Maybe something that brings you a little closer. When you wake up in the morning, with your first coffee and your phone. Which app do you open first?
Max: Oh God, that's embarrassing. It's LinkedIn. Actually. Right now it's really LinkedIn, but I think that just shows how bad the German social media world is.
Lisa: Or what phase of your life you're just in as an old white man, opening LinkedIn.
Max: Or the LinkedIn and Handelsblatt app? No.
Lisa: But I understand LinkedIn. Do you also have this, that you're doing a lot on LinkedIn right now and getting both positive reactions, but people around you are also a bit skeptical about what kind of business bullshit bingo is being played there.
Max: Well, I'm not a fan of business bullshit bingo in the sense of producing it. I like looking at it, I think one of my first bots was also the Business Cringe Generator, where you could simply enter two or three terms and then you'd get the perfect LinkedIn post. These three things I learned about business development because my house burned down today, or something like that, and I built my chatbot around that, which kind of messed the whole thing up a bit. So it's relatively interchangeable, I have to say. I'm not a big LinkedIn fan, yet I still hang around there a lot.
Lisa: Okay, maybe we'll clarify why that is in another episode. But there's one thing you still have to explain to me. Creative Technologist. Did you make that up yourself or is it really a job title?
Max: I live in Berlin. I constantly make up self-descriptions for everything. No, I actually heard that somewhere during a talk. That's how I was introduced, and I adopted it because it quite well reflects what I do. I try to find creative applications with existing technology. So I test things out, try a lot of stuff, and creativity is definitely part of that. So yes, that's why Creative Technologist it is.
Lisa: Yes, I learned something new again. Thank you very much. So you're a Creative Technologist, AI consultant, author, copywriter – we've heard that, and now also a podcaster, all centered around the fascination with AI. Where did it all start for you?
Max: Well, the origin was actually in online marketing, even more in social media. When I started studying, I wrote a whole lot of stuff on the internet. On Twitter back then, and it was received pretty well. As a working-class kid, I reported from my experiences at university, and somehow that hit the spirit of the times, and that led to my first writing jobs and eventually to a book of my own that was published. And through that whole media bubble, I somehow ended up in online marketing, and after a few stops I moved into consulting. I was a communications consultant for three and a half years at a large international PR agency, and inevitably the topic of AI kept coming up more and more. I always use the example that was going around at the time: "What actually happens in the field of crisis communication? What happens when my boss calls me and says he wants me to transfer sum X? But my boss is sitting right next to me and doesn't even have a phone in his hand." How do we deal with deep fakes in a corporate context? How can we address these risks? And that's how I eventually came across AI in the field of crisis communication as well. And that's exactly when I was eventually laid off because restructuring was on the horizon. That's how it goes sometimes.
Lisa: You know it.
Max: You know how it is, and as a result I simply had an extreme amount of time and, let's say, not an immediate financial urge to go straight into a new job. So first I spent a few weeks wandering through the Cambodian jungle. Alone. I thought about what I would do next, enjoyed the time off, and came to the conclusion that there must somehow also be positive use cases for AI. Especially with all that doomsday mood. So many people were talking about the end of the world and saying: "No, there must somehow be positive things too." And when I got back from my trip, I said: "I want to nerd out on this now." And then I saw that it was relatively easy. If you just try a bit, you can learn quickly. So I started a project to build 100 bots in 100 days. And that's how I got into the whole AI game. Until I was able to host the OMR podcast "Kollegin KI" until recently. So now I've basically been a podcaster for two months or so.
Lisa: So yes, you had an affinity for AI. You were already involved with it through your work. But how quickly did you get to the point where you could build bots on your own? Because that's not your background as a marketer and in crisis communication.
Max: I'm not even a marketer. I'm actually a trained sociologist, let's say; I studied media sociology and media studies. But it's a field where, let's say, you can make pretty good progress autodidactically. And that may also be the first tip I can give all listeners. I learned a lot about AI from podcasts, books, newsletters, but also from AI itself. And that's what I find so fascinating about this technology. It's a technology that can describe itself. If we have questions about AI, we can basically just ask a chatbot. We can simply ask ChatGPT how a large language module works and then ChatGPT will give an answer. If we don't understand it, we can ask again and say: "Please explain it to me as if I were ten years old or explain it to me in sailor language or something." Or whatever else you want. And ChatGPT will do that. And you can learn a great deal from that. And that's basically how I taught myself autodidactically until I eventually built the first custom GPTs, which were kind of my entry point.
Lisa: Yes, that's a great tip, because what you're actually saying is: turn an AI into a co-worker, a sparring partner you can exchange ideas with, but also learn from. And I think many sellers ask themselves: "What can AI actually do for me concretely?" I keep hearing, okay, I can be more efficient with it, meaning save time. I can streamline processes with it, I can possibly reduce labor through the use of AI. You said, just start doing it and learn together with AI. But do you have a few concrete examples? If you put yourself in the position of an online seller, how everyday life can be simplified by AI?
Max: Basically, there isn't one single use case that applies to everyone. That's important to say up front. Of course, there are certain processes that are similar. In e-commerce, for example, those are product descriptions, and product descriptions can be created quite well by AI. Above all, they can be checked. So everyone out there with manually written product descriptions can simply put those descriptions into an AI tool of their choice exactly as they are and ask the AI what it thinks about them. Whether it can give feedback, maybe has criticisms, and then you can see whether you might have overlooked something. In fact, one of the main use cases at all is having existing human content double-checked by AI. I do that a lot. And that opens up entirely new use cases, because many people think I enter a prompt and then get a piece of content out and try to use AI that way. But it can also work the other way around, and that can help quite a lot.
Lisa: You just said, one of the many AI tools out there. I think the most common one everyone has tried by now is ChatGPT. If I look at how the landscape of AI tools is developing right now, I get a huge feeling of overwhelm. Do you maybe have one or two tools for our sellers that they should take a look at, where you’d say okay, that’s worth it, that’s where things keep going beyond ChatGPT.
Max: It's always quite difficult to recommend tools. That has to do with the fact that we simply have so many. The website "Is There an AI for That" regularly lists tools. Yes, it really exists. Feel free to check it out; I think it currently lists over 40,000 different AI tools. When I started spending every day full-time on this almost two years ago, it was still possible to try two tools a week that were new. There were also two relevant newsletters, a few podcasts, etc. That still worked. Nowadays it's impossible. You can't try every tool, it just doesn't work.
Lisa: AI burnout.
Max: AI burnout. Totally. Yeah, who doesn't know it? That's why the approach should really be to first find the use case and then choose a tool. So, three basic tools that you maybe should know. Especially when it comes to text processing, of course ChatGPT for everything related to creativity. But I also mean Copilot—many companies use Copilot. Unfortunately, it's not as strong as others. I would rather recommend, especially if you want a text-based chatbot, Langdock. In terms of functionality, it's basically exactly like ChatGPT. Just a Berlin-based company, German servers, GDPR-compliant.
Lisa: Also important.
Max: Yes, exactly, that's really kind of the "tool of the hour" right now. And otherwise, if you want to code, Claude is a pretty good choice. We have Perplexity as a search engine. Those are the common tools you should get familiar with. But as I said, they won't solve every use case either.
Lisa: And an important topic in e-commerce is always product images. And there it's kind of like: "With image generation, we won't need our own photo shoots anymore soon. We don't need studios, we don't need models." How do you feel about image generation? Do you have tools you can recommend to online sellers or also. You talked about use cases. In which use cases can I work on product image optimization, and where should I maybe leave it alone?
Max: So I would advise against first changing existing people with AI. So having model photos based on real people and then changing them so much that they might end up wearing your own clothes or advertising products. You can very quickly end up in legal areas that are simply difficult. What you can of course do is work with interiors or with objects in general. And there are definitely a few tools that are common for that. For image generation, if it should stay in the realistic range, I always like to recommend Flux. Flux.1 is a German image generation tool that is actually pretty cool. For product photos, you may need to dive a bit deeper into the technology. There are so-called low-rank adaptation models that you can also create yourself quite well. The so-called LoRA.
Lisa: What are low-rank adaptation models?
Max: That is... basically you "finetune" an image AI with it and say that the images or the setting should stay the same across different image productions with AI. That means: if you, for example, prompt a person and say: "I'd like a dark-haired, middle-aged man." Then... you would get ten different people in ten attempts. But if you attach a LoRA with a face, maybe. Then the person will look similar, and that way you can use it for products too. It requires a bit of technical understanding, but you can get really nerdy about it. Also with AI, of course. Otherwise there are good YouTube tutorials for it. If you can spare the time, anyone can definitely teach themselves.
Lisa: Of course, that's also important if, for example, you have your own "look and feel" for product photos in your eBay shop that carries through, that there is, as you called it?
Max: A low-rank adaptation model. LoRA.
Lisa: Good, then I'll make myself some notes to Google after this podcast. I think that's great.
Max: Very good. Yes, exactly. If you have, say, a bench and you take that bench, remove the background, take I think 20, 25 pictures from different perspectives, and train an AI on it. That usually works locally, it just takes a bit of time, but you don't need the big GPUs that cost hundreds of thousands of euros. You can actually do that for home use too. There are many good examples of this. And as I said, just Google it or look on YouTube. There are great tutorials there.
Lisa: And I think image generation or improving image quality is super important, because we always have to remind ourselves that product photos are simply our... our showcase for the item we want to sell. And if we think about important things like: "An image must be cut out on a white background in order to appear in Google search at all or something like that." Things like that can of course be implemented super easily with AI, or, for example, it's also a tool we've integrated at eBay, where you take a photo of your product and it gets cut out on a white background and added to your listing. Those are, I think, the minimum requirements we now have for AI tools. But yes, the "limit of the possible" is still not really tangible or visible, or perhaps not at all.
Max: So I’m, I’m curious to see what comes next. I can imagine that there will actually be individual product photos for the respective users who come to the listings. So that you really see, okay, your cookies tell me you’re looking for winter furniture, so I’ll show you the rattan chair tipping over in winter or something. Yes, so I can imagine that there will be strong individualization in e-commerce and that AI can of course support that pretty well.
Lisa: One topic we simply can't avoid internally, and one we get a lot of questions about, is visibility. We've already done so many episodes about optimization on the eBay marketplace or how to get visibility outside of eBay for your listings. And we always talk about SEO with Google and marketplace SEO, how you get visibility on eBay. And then GEO comes along, meaning optimization for large language models. You're already a bit in your... a bit disillusioned, but maybe let's take a quick look at GEO. What's your opinion on it and how much do sellers need to deal with the topic of GEO?
Max: So the idea behind it is that in the future, purchases will also happen via AI, that we will have agents that have our credit card data. And when we say we want to book a hotel, plan a business trip, buy certain things, that an agent just does that for us.
Lisa: Yes.
Max: To be honest, I currently just think that's a fairy tale. I think the agents I know aren't that good either and aren't error-free enough that you'd hand them your own credit card data and let them run with it. Be that as it may. GEO, I think, is a trend, a buzzword and a trendy term. Technically, it's simply not possible for us to get our products, our listings, somehow into large language models so that they spit them out. So that. That can't be controlled. We do actually have the option to do SEO, and I would still recommend that to everyone so that you can be found in search engines. Of course, with ChatGPT and others, we have something like web search, but even that doesn't always access the current HTML pages. So it's a trendy term right now, but in my opinion there's not much more behind it than just the new trend in the SEO field. And I think most companies are well advised to continue SEO, and to perhaps try integrating AI into their webshops as well. Those are always good things, like a chatbot, so you can chat with a virtual sales assistant and say: "I have this and that problem. What solutions can you offer me?" That's actually a pretty good thing, especially for technical products, and it can provide a positive user experience. But trying to passively get into some large language model is technically not really feasible.
Lisa: But I think that's a good help for prioritization for sellers. Where do I even start? Is my next step to try to understand GEO and increase the visibility of my products? Or do I look at how I can become more efficient in my doings or in my use cases, as you just mentioned?
Max: Yes, optimizing processes—that's, I think, what everyone should be doing right now. And like everyone else, we're of course looking for use cases, and that requires a certain amount of creativity. I have a kind of guide that I also like to use in workshops, where I say: "Okay, forget everything you know about AI for a moment, and make notes about the three areas that you basically work through every day or that come up for you." The first area is repetitive things. What keeps coming up? What needs to be done every day? And then just write it down without keeping AI in mind. The second are creative tasks. So everything involving product images, for example. Or what kind of creative assets are actually required? And the third are tasks with many variables. That's where we quickly get into areas like performance marketing or perhaps something like warehouse management in e-commerce. So where are there many variables, where is there a lot of data that I might also get access to and can then feed into AI accordingly? And once you've made that list with these three points, you can take that list and I always think it's great to just throw it into ChatGPT and say: "Hey, which of these projects is suitable for me? I have no prior experience, I don't have an AI team or whatever." Basically, you just enter what you do, and ChatGPT will help you. Or the AI of your choice. So as I said, it's the most valuable tip, and every workshop uses AI to understand AI. Integrate AI into your workflows and start small! That's, I think, the most important tip. I've heard so many things like: "We're going to fully automate customer support tomorrow." That always goes wrong, it always goes wrong. In principle, if your goal from the start is to replace people, then you will never get people to use this technology. And those are exactly the kinds of difficulties you can avoid.
Lisa: So not replacing people, but empowering them with the tools that exist. You said, I look at what's repetitive? I look at where I need to be creative? And the last one was.
Max: Where there are quite a lot of variables, exactly...
Lisa: Exactly. For which of those is prompting the right approach? Or just chatting with an AI? Or where do I take things a bit further into automation or even outsourcing? You mentioned custom GPTs earlier. Anyone who uses ChatGPT knows that I can now create projects in ChatGPT and that it's nice because then the AI knows what I'm talking about and can set rules. But could you maybe briefly explain the difference between a prompt, a project, a custom GPT, or even an AI agent?
Max: So prompting is first and foremost the be-all and end-all; we all have to learn that, and I like to compare it to Google. When Google came along in the 90s, we also had to learn how to use Google, because it's simply human-machine interaction, not human interaction. We need to know what we give the machine so that we get the output we expect in the end. And with Google, we now know that we write in keywords, that we don't need punctuation, and so on. That if we want exactly one keyword to appear in the search results, we put it in quotation marks. If, for example, we want results from eBay too, we can also enter eBay as a URL in the search, etc., and we have to learn prompting in exactly the same way. And I really recommend that everyone learn prompting at the beginning. Because prompting is the skill needed to operate AI. And after that, you can turn to assistants, to AI assistants. These are then, for example, custom GPTs, which are quite good for getting started. Of course, there are completely different ways to build your own chatbots as well. Those are basically the assistants I'm talking about. That means you give these assistants a meta-prompt, something that is placed earlier in the hierarchy. As an example, if I build a chatbot or a custom GPT. And I say in the meta-prompt: "No matter what input comes in, answer in three random emojis."
Lisa: So those are basically rules that I tell my AI: this is what I expect from you every time, no matter when I talk to you in this GPT.
Max: And if you then use that GPT, which I have given the three random emojis in the meta-prompt, and you write "Hello", it won't classically write "Hello, how can I help you?" or "Hello, how are you today?", but will give three random emojis. And those are exactly the prompt hierarchies you need to understand in order to build assistants. And that perhaps already makes clear where the creativity actually is. You have to think about how to transform input from users so that the AI delivers the corresponding output. And that is an extremely relevant thing. I like to compare it to a library. Imagine a large language model as a huge library with millions of bookshelves and a little robot driving around in it. A little AI robot, a bot, going from shelf to shelf. And you are at the entrance of this library and you hand this robot a note with your request on it. For example, you want to know, I don't know, where can I sell things online well? And the bot then drives around through all the shelves and.
Lisa: It can simply read unbelievably fast.
Max: Exactly. It reads very, very fast and comes out at the end and says: "Yes, take a look at this, eBay is a great thing, you can sell all kinds of things well there." Yes, and that's basically how you have to imagine large language models. That also leads to problems, because this library is of course not always up to date, and it doesn't contain only true books, but also a lot of internet nonsense that has somehow been crawled together. That means you have to be careful with the output. And the thing we do with an assistant is we give this bot an additional bookshelf. We place that in the library for it. And that may contain eBay articles or guidelines on how to list products on eBay best. And the bot then naturally goes through all its shelves and looks at that, but it definitely also goes to our shelf, and that way you can improve the output significantly. And that's basically, "in a nutshell," explained in a non-technical way, how a chatbot works.
Lisa: I understood it for the first time.
Max: Yes, gladly.
Lisa: Really good.
Max: Yes.
Lisa: But, I mean, that was really very, very well illustrated. Thank you very much. I think I'd like to bring this back to e-commerce again. Because especially looking at repetitive tasks—that's what somehow costs money, what annoys every online seller. What recurring tasks do you think can now be looked at with AI at the click of a button?
Max: At the click of a button is a bit exaggerated.
Lisa: Two, okay.
Max: Yes, two. So I'm not a fan of automation, so-called agents that actually complete things completely autonomously. There are plenty of examples of those causing problems. Large language models always have a certain error rate. You'll never have an agent that works 100 percent accurately. And in the end, it comes down to looking at where the human error rate is higher than the AI's.
Lisa: Okay.
Max: And then you can say: "Good, maybe I'll tackle that, I'll automate it." But then there are also things like "Human in the Loop," where, for example, you still need a person to look over the output. And that's especially recommended for personal data or for things that are simply business-critical. If you have a customer support agent, for example, my favorite example. And then they're allowed to hand out vouchers or something, then of course you have to look at it as a human.
Lisa: How much budget is that, actually?
Max: Yes, exactly. Is that even possible? And I would always keep such things in mind. But yes, repetitive tasks are of course extremely suitable. And you always have to think about whether it's really a time saver. Whether in the end you save the person who enters an existing text or keywords into an assistant, gets a product text out, and then manually reads over it once or copies it manually into a shop page. That's a matter of seconds. Does that need to be automated? And do you then accept the AI's error rate too? That's the question you should ask yourself. And in many cases, it becomes clear that it's actually nonsense.
Lisa: Yes, I think "Human in the Loop" is, especially when we relate it back to eBay listings, whether you use the AI-generated product description on eBay, for example. The responsibility for the listing ultimately lies with you as the person behind it. That means you also have to check again: Is it, as you said, internet nonsense and is it hallucination? Or maybe factually just not correct? Because the buying experience behind it makes you, as the selling person, accountable for that.
Max: Yes.
Lisa: Max, we talked a bit in the pre-interview about the next stage. What comes after I’ve prompted something or built a custom GPT, how can AI then also support me with more complex things where I might otherwise need to build expertise or look for some external tool or consulting? Things like optimizing my warehouse management or a prediction model for when I actually need to reorder items. In my head, these are all things I can now solve with AI.
Max: Yes, yes and no. So in theory, you can of course solve all that with AI, but these are such niche things that you should really rely on specialized providers. In that case, I would actually get more expertise or build it internally. But I also notice in my consulting work that there's simply often no time for that. AI is not something you do on the side. It reminds me a bit of the early days of social media. Yes, maybe you remember when interns did social media because they're young, so they can do it.
Lisa: Yes, they're still so close to the target group.
Max: Yes, exactly.
Lisa: That's still the case for us today.
Max: Yes. Yes, okay. But I mean that specialization has taken place. I mean, if I look around here in the podcast studio, five people are sitting here watching us. It's not like, oh come on, just some online content stuff, you do that on the side, but it's a profession, and AI is a profession too. That means I'd really dedicate time to it. I'd really train people thoroughly. The time is worth investing, because the use cases usually come from the employees themselves within the company. It doesn't help anyone to impose AI from above and say: "You have to use AI now." Ideally with the added KPI that somehow 30% of employees should be laid off. Try getting employees to learn a tool that is ultimately aimed at eliminating them. That doesn't work.
Lisa: But what I think is a super interesting point is a bit: what do I bring in-house as knowledge, and where is it also wasted effort because I simply don't have the time to keep up with it. And I bring in an agency or consulting or external help into the company.
Max: Well, the "make or buy" decision isn't new. And with AI, it's of course quite relevant too. I would always recommend at least initially bringing in experts and involving them, simply to build knowledge management internally. That can then eventually run on its own. And good consultants also take that as a goal, that they ensure companies can eventually continue on their own. And I would definitely recommend that. For the start, that means simply initiating this transformation process, which is actually happening in all companies right now, professionally, and then continuing on your own afterward—that, I think, is the right path, because then you can much better decide: What tools can I build in-house with my specialized AI team that I've set up, and which tools do I maybe prefer to buy externally? And even in that case, the specialists from the AI team also know: What is actually a tool that really helps us? Where might the dangers be? And accurately assessing that error rate—that's the kind of thing.
Lisa: Yes, or also saving the time of not having to test things so much.
Max: Exactly. Yes, yes.
Lisa: If you'd like to dive deeper into AI applications and exchange ideas with other sellers about AI in everyday life, then be sure to check out the eBay Community or our eBay for Business Discord channel.
Lisa: Max. I think some people also sometimes have understandable reservations when it comes to the topic of AI and AI. We've already said: "Hey, why should I teach an employee a tool that eliminates that person?" How do you perceive skepticism, maybe in your podcast, in the industry, or among people? And how do you respond to these doubts?
Max: I think companies can actively do that by empowering their employees instead of replacing them. Employees here are not cost centers to be replaced by AI; rather, employees can develop entirely new business cases, achieve very different standards in efficiency and new ideas, if they are enabled to work with AI. And I think we should view AI as an extension of our human abilities and not as an enemy.
Lisa: So more additional workforce and not instead of.
Max: 100%. That's why my podcast is called "Kollegin KI," because AI appears as a colleague and not as a replacing boss or manager.
Lisa: If you look into the crystal ball toward the future, say one or two years ahead. What do you think AI will change about retail or about us?
Max: I'm regularly asked about the future, and I keep telling the same story about the last time I answered that, in 2018 for the magazine of TalkWalker, the social listening tool, when they asked: "What social media trends do I actually see for 2019?" And I said: "Wow, definitely employer branding, campaigns, labor shortages, etc., huge." And what was 2019? Corona, and the trends were video calls. Yes, and then I said I was done making trend predictions like that.
Lisa: Yes, but at least you didn't predict a global pandemic and then stop thinking altogether.
Max: That would have been wild if I had done that. But no, I always say: "Stay flexible, try to be as flexible as possible." And that's basically it. All the studies out there on AI confirm that. The companies that benefit are those that position themselves flexibly enough to respond to rapid changes. That means no 20-year contracts with any vendors. New tools are constantly hitting the market that are worth testing somehow. That means looking at how I build my architecture when I set up AI systems internally so that I can perhaps change them again relatively quickly? We need to learn to handle it more flexibly. And that's, I think, the best advice I can give. What happens in e-commerce? Yes, I think the trends are quite clear. AI will be used much more in communication. That sometimes works better, sometimes worse, to be fair. There will certainly be much more efficient supply chains, because AI can "predict" a whole lot and can handle large data sets much better than humans. And I believe that if we use AI correctly, there will also be far fewer careless mistakes. Because AI doesn't make careless mistakes. That's a human problem. And that's why I say this use case of feeding human content into AI and having it looked over is actually a very exciting thing.
Lisa: And maybe to wrap up, if I finish listening to this podcast and then head back to my laptop. What's the first tip you'd give sellers?
Max: What you can do after this episode. Especially if you have various questions. That will certainly be the case. For example: "How do I train a LoRA now? What is a low-rank adaptation model or how do I use a chatbot? How do I build a custom GPT? What is a large language model and why is it nonsense for me to spend money to somehow get my products into it?" These are all questions you can ask AI. So if, after this episode, you open the chatbot of your choice. That can be ChatGPT, Perplexity, Claude, whatever you use, then you can ask these questions there too and at least get a starting point for an answer. Of course, they're never 100% correct, as I've said. So AI output is not always error-free, just like human output isn't. You can get just as bad advice from people as from AI. But it's definitely a first point of contact and can show you a lot. And I think I'm living proof of that. After all, I learned a whole lot about AI through AI, and now it's kind of my job.
Lisa: Now it's kind of your job, and you do it excellently. And I think we've given the listeners a few homework assignments. And if you take one thing away from this episode, then hopefully that AI doesn't necessarily have to be complicated and can really move you forward in everyday life. Because even small, well-chosen use cases can already be enough to save noticeable time, automate routines, and boost your own creativity. A very, very big thank-you to Max for the exciting insights and the practical tips. We'll put all the links and further information in the show notes as always. And if you liked the episode, then subscribe to "Alles top. Gerne wieder!" so you don't miss any future episodes. And we'll hear from you next time.
Max: And if you want to learn more about AI, including AI in e-commerce, then feel free to listen to the OMR podcast "Kollegin KI," hosted by me with exciting guests, if I may do a little advertising at the end here, dear Lisa.
Lisa: Of course.
Max: Yes, then feel free to listen! Everywhere you get podcasts. New every Tuesday.
Max: AI is being used much more in communication. That sometimes works better, sometimes worse, to be fair. There will certainly be much more efficient supply chains, because AI can "predict" a whole lot and can handle large data sets much better than humans. And I believe that if we use AI correctly, there will also be far fewer careless mistakes. Because AI doesn't make careless mistakes. That's a human problem.
Lisa: A warm welcome to "Alles top. Gerne wieder!" the eBay podcast about retail and e-commerce. I'm Lisa Haag and today we're talking about a topic that probably comes up in every second meeting right now: artificial intelligence. But instead of just talking about buzzwords, we're taking a very practical look at how AI can support you with your "to dos" and why looking at your own use cases is the best way to find the right tools. To do that, I've invited a guest who works on exactly that. Max Mundhenke is a Creative Technologist, consultant, and host of the OMR podcast "Kollegin KI". He originally comes from the creative industry, worked as a copywriter, author, and strategy consultant, and today he'll show you how to move your business forward with AI. Hi Max.
Max: Hi Lisa.
Lisa: Are you happy to be here?
Max: Yes, very much. I'm not exactly sure whether I can keep the promise I can give everyone here about how to move their business forward. But I can certainly give a few pointers.
Lisa: We always have to open the scissors very wide and then see how we deliver in the end. Max, you and I know each other pretty well. I think that's worth saying up front. We also know each other outside the podcast. That means if we interrupt each other a little, then that's a sign of our friendship. And not rudeness.
Max: Interrupt me anytime.
Lisa: But we can give our sellers a little introduction to you. Maybe something that brings you a little closer. When you wake up in the morning, with your first coffee and your phone. Which app do you open first?
Max: Oh God, that's embarrassing. It's LinkedIn. Actually. Right now it's really LinkedIn, but I think that just shows how bad the German social media world is.
Lisa: Or what phase of your life you're just in as an old white man, opening LinkedIn.
Max: Or the LinkedIn and Handelsblatt app? No.
Lisa: But I understand LinkedIn. Do you also have this, that you're doing a lot on LinkedIn right now and getting both positive reactions, but people around you are also a bit skeptical about what kind of business bullshit bingo is being played there.
Max: Well, I'm not a fan of business bullshit bingo in the sense of producing it. I like looking at it, I think one of my first bots was also the Business Cringe Generator, where you could simply enter two or three terms and then you'd get the perfect LinkedIn post. These three things I learned about business development because my house burned down today, or something like that, and I built my chatbot around that, which kind of messed the whole thing up a bit. So it's relatively interchangeable, I have to say. I'm not a big LinkedIn fan, yet I still hang around there a lot.
Lisa: Okay, maybe we'll clarify why that is in another episode. But there's one thing you still have to explain to me. Creative Technologist. Did you make that up yourself or is it really a job title?
Max: I live in Berlin. I constantly make up self-descriptions for everything. No, I actually heard that somewhere during a talk. That's how I was introduced, and I adopted it because it quite well reflects what I do. I try to find creative applications with existing technology. So I test things out, try a lot of stuff, and creativity is definitely part of that. So yes, that's why Creative Technologist it is.
Lisa: Yes, I learned something new again. Thank you very much. So you're a Creative Technologist, AI consultant, author, copywriter – we've heard that, and now also a podcaster, all centered around the fascination with AI. Where did it all start for you?
Max: Well, the origin was actually in online marketing, even more in social media. When I started studying, I wrote a whole lot of stuff on the internet. On Twitter back then, and it was received pretty well. As a working-class kid, I reported from my experiences at university, and somehow that hit the spirit of the times, and that led to my first writing jobs and eventually to a book of my own that was published. And through that whole media bubble, I somehow ended up in online marketing, and after a few stops I moved into consulting. I was a communications consultant for three and a half years at a large international PR agency, and inevitably the topic of AI kept coming up more and more. I always use the example that was going around at the time: "What actually happens in the field of crisis communication? What happens when my boss calls me and says he wants me to transfer sum X? But my boss is sitting right next to me and doesn't even have a phone in his hand." How do we deal with deep fakes in a corporate context? How can we address these risks? And that's how I eventually came across AI in the field of crisis communication as well. And that's exactly when I was eventually laid off because restructuring was on the horizon. That's how it goes sometimes.
Lisa: You know it.
Max: You know how it is, and as a result I simply had an extreme amount of time and, let's say, not an immediate financial urge to go straight into a new job. So first I spent a few weeks wandering through the Cambodian jungle. Alone. I thought about what I would do next, enjoyed the time off, and came to the conclusion that there must somehow also be positive use cases for AI. Especially with all that doomsday mood. So many people were talking about the end of the world and saying: "No, there must somehow be positive things too." And when I got back from my trip, I said: "I want to nerd out on this now." And then I saw that it was relatively easy. If you just try a bit, you can learn quickly. So I started a project to build 100 bots in 100 days. And that's how I got into the whole AI game. Until I was able to host the OMR podcast "Kollegin KI" until recently. So now I've basically been a podcaster for two months or so.
Lisa: So yes, you had an affinity for AI. You were already involved with it through your work. But how quickly did you get to the point where you could build bots on your own? Because that's not your background as a marketer and in crisis communication.
Max: I'm not even a marketer. I'm actually a trained sociologist, let's say; I studied media sociology and media studies. But it's a field where, let's say, you can make pretty good progress autodidactically. And that may also be the first tip I can give all listeners. I learned a lot about AI from podcasts, books, newsletters, but also from AI itself. And that's what I find so fascinating about this technology. It's a technology that can describe itself. If we have questions about AI, we can basically just ask a chatbot. We can simply ask ChatGPT how a large language module works and then ChatGPT will give an answer. If we don't understand it, we can ask again and say: "Please explain it to me as if I were ten years old or explain it to me in sailor language or something." Or whatever else you want. And ChatGPT will do that. And you can learn a great deal from that. And that's basically how I taught myself autodidactically until I eventually built the first custom GPTs, which were kind of my entry point.
Lisa: Yes, that's a great tip, because what you're actually saying is: turn an AI into a co-worker, a sparring partner you can exchange ideas with, but also learn from. And I think many sellers ask themselves: "What can AI actually do for me concretely?" I keep hearing, okay, I can be more efficient with it, meaning save time. I can streamline processes with it, I can possibly reduce labor through the use of AI. You said, just start doing it and learn together with AI. But do you have a few concrete examples? If you put yourself in the position of an online seller, how everyday life can be simplified by AI?
Max: Basically, there isn't one single use case that applies to everyone. That's important to say up front. Of course, there are certain processes that are similar. In e-commerce, for example, those are product descriptions, and product descriptions can be created quite well by AI. Above all, they can be checked. So everyone out there with manually written product descriptions can simply put those descriptions into an AI tool of their choice exactly as they are and ask the AI what it thinks about them. Whether it can give feedback, maybe has criticisms, and then you can see whether you might have overlooked something. In fact, one of the main use cases at all is having existing human content double-checked by AI. I do that a lot. And that opens up entirely new use cases, because many people think I enter a prompt and then get a piece of content out and try to use AI that way. But it can also work the other way around, and that can help quite a lot.
Lisa: You just said, one of the many AI tools out there. I think the most common one everyone has tried by now is ChatGPT. If I look at how the landscape of AI tools is developing right now, I get a huge feeling of overwhelm. Do you maybe have one or two tools for our sellers that they should take a look at, where you’d say okay, that’s worth it, that’s where things keep going beyond ChatGPT.
Max: It's always quite difficult to recommend tools. That has to do with the fact that we simply have so many. The website "Is There an AI for That" regularly lists tools. Yes, it really exists. Feel free to check it out; I think it currently lists over 40,000 different AI tools. When I started spending every day full-time on this almost two years ago, it was still possible to try two tools a week that were new. There were also two relevant newsletters, a few podcasts, etc. That still worked. Nowadays it's impossible. You can't try every tool, it just doesn't work.
Lisa: AI burnout.
Max: AI burnout. Totally. Yeah, who doesn't know it? That's why the approach should really be to first find the use case and then choose a tool. So, three basic tools that you maybe should know. Especially when it comes to text processing, of course ChatGPT for everything related to creativity. But I also mean Copilot—many companies use Copilot. Unfortunately, it's not as strong as others. I would rather recommend, especially if you want a text-based chatbot, Langdock. In terms of functionality, it's basically exactly like ChatGPT. Just a Berlin-based company, German servers, GDPR-compliant.
Lisa: Also important.
Max: Yes, exactly, that's really kind of the "tool of the hour" right now. And otherwise, if you want to code, Claude is a pretty good choice. We have Perplexity as a search engine. Those are the common tools you should get familiar with. But as I said, they won't solve every use case either.
Lisa: And an important topic in e-commerce is always product images. And there it's kind of like: "With image generation, we won't need our own photo shoots anymore soon. We don't need studios, we don't need models." How do you feel about image generation? Do you have tools you can recommend to online sellers or also. You talked about use cases. In which use cases can I work on product image optimization, and where should I maybe leave it alone?
Max: So I would advise against first changing existing people with AI. So having model photos based on real people and then changing them so much that they might end up wearing your own clothes or advertising products. You can very quickly end up in legal areas that are simply difficult. What you can of course do is work with interiors or with objects in general. And there are definitely a few tools that are common for that. For image generation, if it should stay in the realistic range, I always like to recommend Flux. Flux.1 is a German image generation tool that is actually pretty cool. For product photos, you may need to dive a bit deeper into the technology. There are so-called low-rank adaptation models that you can also create yourself quite well. The so-called LoRA.
Lisa: What are low-rank adaptation models?
Max: That is... basically you "finetune" an image AI with it and say that the images or the setting should stay the same across different image productions with AI. That means: if you, for example, prompt a person and say: "I'd like a dark-haired, middle-aged man." Then... you would get ten different people in ten attempts. But if you attach a LoRA with a face, maybe. Then the person will look similar, and that way you can use it for products too. It requires a bit of technical understanding, but you can get really nerdy about it. Also with AI, of course. Otherwise there are good YouTube tutorials for it. If you can spare the time, anyone can definitely teach themselves.
Lisa: Of course, that's also important if, for example, you have your own "look and feel" for product photos in your eBay shop that carries through, that there is, as you called it?
Max: A low-rank adaptation model. LoRA.
Lisa: Good, then I'll make myself some notes to Google after this podcast. I think that's great.
Max: Very good. Yes, exactly. If you have, say, a bench and you take that bench, remove the background, take I think 20, 25 pictures from different perspectives, and train an AI on it. That usually works locally, it just takes a bit of time, but you don't need the big GPUs that cost hundreds of thousands of euros. You can actually do that for home use too. There are many good examples of this. And as I said, just Google it or look on YouTube. There are great tutorials there.
Lisa: And I think image generation or improving image quality is super important, because we always have to remind ourselves that product photos are simply our... our showcase for the item we want to sell. And if we think about important things like: "An image must be cut out on a white background in order to appear in Google search at all or something like that." Things like that can of course be implemented super easily with AI, or, for example, it's also a tool we've integrated at eBay, where you take a photo of your product and it gets cut out on a white background and added to your listing. Those are, I think, the minimum requirements we now have for AI tools. But yes, the "limit of the possible" is still not really tangible or visible, or perhaps not at all.
Max: So I’m, I’m curious to see what comes next. I can imagine that there will actually be individual product photos for the respective users who come to the listings. So that you really see, okay, your cookies tell me you’re looking for winter furniture, so I’ll show you the rattan chair tipping over in winter or something. Yes, so I can imagine that there will be strong individualization in e-commerce and that AI can of course support that pretty well.
Lisa: One topic we simply can't avoid internally, and one we get a lot of questions about, is visibility. We've already done so many episodes about optimization on the eBay marketplace or how to get visibility outside of eBay for your listings. And we always talk about SEO with Google and marketplace SEO, how you get visibility on eBay. And then GEO comes along, meaning optimization for large language models. You're already a bit in your... a bit disillusioned, but maybe let's take a quick look at GEO. What's your opinion on it and how much do sellers need to deal with the topic of GEO?
Max: So the idea behind it is that in the future, purchases will also happen via AI, that we will have agents that have our credit card data. And when we say we want to book a hotel, plan a business trip, buy certain things, that an agent just does that for us.
Lisa: Yes.
Max: To be honest, I currently just think that's a fairy tale. I think the agents I know aren't that good either and aren't error-free enough that you'd hand them your own credit card data and let them run with it. Be that as it may. GEO, I think, is a trend, a buzzword and a trendy term. Technically, it's simply not possible for us to get our products, our listings, somehow into large language models so that they spit them out. So that. That can't be controlled. We do actually have the option to do SEO, and I would still recommend that to everyone so that you can be found in search engines. Of course, with ChatGPT and others, we have something like web search, but even that doesn't always access the current HTML pages. So it's a trendy term right now, but in my opinion there's not much more behind it than just the new trend in the SEO field. And I think most companies are well advised to continue SEO, and to perhaps try integrating AI into their webshops as well. Those are always good things, like a chatbot, so you can chat with a virtual sales assistant and say: "I have this and that problem. What solutions can you offer me?" That's actually a pretty good thing, especially for technical products, and it can provide a positive user experience. But trying to passively get into some large language model is technically not really feasible.
Lisa: But I think that's a good help for prioritization for sellers. Where do I even start? Is my next step to try to understand GEO and increase the visibility of my products? Or do I look at how I can become more efficient in my doings or in my use cases, as you just mentioned?
Max: Yes, optimizing processes—that's, I think, what everyone should be doing right now. And like everyone else, we're of course looking for use cases, and that requires a certain amount of creativity. I have a kind of guide that I also like to use in workshops, where I say: "Okay, forget everything you know about AI for a moment, and make notes about the three areas that you basically work through every day or that come up for you." The first area is repetitive things. What keeps coming up? What needs to be done every day? And then just write it down without keeping AI in mind. The second are creative tasks. So everything involving product images, for example. Or what kind of creative assets are actually required? And the third are tasks with many variables. That's where we quickly get into areas like performance marketing or perhaps something like warehouse management in e-commerce. So where are there many variables, where is there a lot of data that I might also get access to and can then feed into AI accordingly? And once you've made that list with these three points, you can take that list and I always think it's great to just throw it into ChatGPT and say: "Hey, which of these projects is suitable for me? I have no prior experience, I don't have an AI team or whatever." Basically, you just enter what you do, and ChatGPT will help you. Or the AI of your choice. So as I said, it's the most valuable tip, and every workshop uses AI to understand AI. Integrate AI into your workflows and start small! That's, I think, the most important tip. I've heard so many things like: "We're going to fully automate customer support tomorrow." That always goes wrong, it always goes wrong. In principle, if your goal from the start is to replace people, then you will never get people to use this technology. And those are exactly the kinds of difficulties you can avoid.
Lisa: So not replacing people, but empowering them with the tools that exist. You said, I look at what's repetitive? I look at where I need to be creative? And the last one was.
Max: Where there are quite a lot of variables, exactly...
Lisa: Exactly. For which of those is prompting the right approach? Or just chatting with an AI? Or where do I take things a bit further into automation or even outsourcing? You mentioned custom GPTs earlier. Anyone who uses ChatGPT knows that I can now create projects in ChatGPT and that it's nice because then the AI knows what I'm talking about and can set rules. But could you maybe briefly explain the difference between a prompt, a project, a custom GPT, or even an AI agent?
Max: So prompting is first and foremost the be-all and end-all; we all have to learn that, and I like to compare it to Google. When Google came along in the 90s, we also had to learn how to use Google, because it's simply human-machine interaction, not human interaction. We need to know what we give the machine so that we get the output we expect in the end. And with Google, we now know that we write in keywords, that we don't need punctuation, and so on. That if we want exactly one keyword to appear in the search results, we put it in quotation marks. If, for example, we want results from eBay too, we can also enter eBay as a URL in the search, etc., and we have to learn prompting in exactly the same way. And I really recommend that everyone learn prompting at the beginning. Because prompting is the skill needed to operate AI. And after that, you can turn to assistants, to AI assistants. These are then, for example, custom GPTs, which are quite good for getting started. Of course, there are completely different ways to build your own chatbots as well. Those are basically the assistants I'm talking about. That means you give these assistants a meta-prompt, something that is placed earlier in the hierarchy. As an example, if I build a chatbot or a custom GPT. And I say in the meta-prompt: "No matter what input comes in, answer in three random emojis."
Lisa: So those are basically rules that I tell my AI: this is what I expect from you every time, no matter when I talk to you in this GPT.
Max: And if you then use that GPT, which I have given the three random emojis in the meta-prompt, and you write "Hello", it won't classically write "Hello, how can I help you?" or "Hello, how are you today?", but will give three random emojis. And those are exactly the prompt hierarchies you need to understand in order to build assistants. And that perhaps already makes clear where the creativity actually is. You have to think about how to transform input from users so that the AI delivers the corresponding output. And that is an extremely relevant thing. I like to compare it to a library. Imagine a large language model as a huge library with millions of bookshelves and a little robot driving around in it. A little AI robot, a bot, going from shelf to shelf. And you are at the entrance of this library and you hand this robot a note with your request on it. For example, you want to know, I don't know, where can I sell things online well? And the bot then drives around through all the shelves and.
Lisa: It can simply read unbelievably fast.
Max: Exactly. It reads very, very fast and comes out at the end and says: "Yes, take a look at this, eBay is a great thing, you can sell all kinds of things well there." Yes, and that's basically how you have to imagine large language models. That also leads to problems, because this library is of course not always up to date, and it doesn't contain only true books, but also a lot of internet nonsense that has somehow been crawled together. That means you have to be careful with the output. And the thing we do with an assistant is we give this bot an additional bookshelf. We place that in the library for it. And that may contain eBay articles or guidelines on how to list products on eBay best. And the bot then naturally goes through all its shelves and looks at that, but it definitely also goes to our shelf, and that way you can improve the output significantly. And that's basically, "in a nutshell," explained in a non-technical way, how a chatbot works.
Lisa: I understood it for the first time.
Max: Yes, gladly.
Lisa: Really good.
Max: Yes.
Lisa: But, I mean, that was really very, very well illustrated. Thank you very much. I think I'd like to bring this back to e-commerce again. Because especially looking at repetitive tasks—that's what somehow costs money, what annoys every online seller. What recurring tasks do you think can now be looked at with AI at the click of a button?
Max: At the click of a button is a bit exaggerated.
Lisa: Two, okay.
Max: Yes, two. So I'm not a fan of automation, so-called agents that actually complete things completely autonomously. There are plenty of examples of those causing problems. Large language models always have a certain error rate. You'll never have an agent that works 100 percent accurately. And in the end, it comes down to looking at where the human error rate is higher than the AI's.
Lisa: Okay.
Max: And then you can say: "Good, maybe I'll tackle that, I'll automate it." But then there are also things like "Human in the Loop," where, for example, you still need a person to look over the output. And that's especially recommended for personal data or for things that are simply business-critical. If you have a customer support agent, for example, my favorite example. And then they're allowed to hand out vouchers or something, then of course you have to look at it as a human.
Lisa: How much budget is that, actually?
Max: Yes, exactly. Is that even possible? And I would always keep such things in mind. But yes, repetitive tasks are of course extremely suitable. And you always have to think about whether it's really a time saver. Whether in the end you save the person who enters an existing text or keywords into an assistant, gets a product text out, and then manually reads over it once or copies it manually into a shop page. That's a matter of seconds. Does that need to be automated? And do you then accept the AI's error rate too? That's the question you should ask yourself. And in many cases, it becomes clear that it's actually nonsense.
Lisa: Yes, I think "Human in the Loop" is, especially when we relate it back to eBay listings, whether you use the AI-generated product description on eBay, for example. The responsibility for the listing ultimately lies with you as the person behind it. That means you also have to check again: Is it, as you said, internet nonsense and is it hallucination? Or maybe factually just not correct? Because the buying experience behind it makes you, as the selling person, accountable for that.
Max: Yes.
Lisa: Max, we talked a bit in the pre-interview about the next stage. What comes after I’ve prompted something or built a custom GPT, how can AI then also support me with more complex things where I might otherwise need to build expertise or look for some external tool or consulting? Things like optimizing my warehouse management or a prediction model for when I actually need to reorder items. In my head, these are all things I can now solve with AI.
Max: Yes, yes and no. So in theory, you can of course solve all that with AI, but these are such niche things that you should really rely on specialized providers. In that case, I would actually get more expertise or build it internally. But I also notice in my consulting work that there's simply often no time for that. AI is not something you do on the side. It reminds me a bit of the early days of social media. Yes, maybe you remember when interns did social media because they're young, so they can do it.
Lisa: Yes, they're still so close to the target group.
Max: Yes, exactly.
Lisa: That's still the case for us today.
Max: Yes. Yes, okay. But I mean that specialization has taken place. I mean, if I look around here in the podcast studio, five people are sitting here watching us. It's not like, oh come on, just some online content stuff, you do that on the side, but it's a profession, and AI is a profession too. That means I'd really dedicate time to it. I'd really train people thoroughly. The time is worth investing, because the use cases usually come from the employees themselves within the company. It doesn't help anyone to impose AI from above and say: "You have to use AI now." Ideally with the added KPI that somehow 30% of employees should be laid off. Try getting employees to learn a tool that is ultimately aimed at eliminating them. That doesn't work.
Lisa: But what I think is a super interesting point is a bit: what do I bring in-house as knowledge, and where is it also wasted effort because I simply don't have the time to keep up with it. And I bring in an agency or consulting or external help into the company.
Max: Well, the "make or buy" decision isn't new. And with AI, it's of course quite relevant too. I would always recommend at least initially bringing in experts and involving them, simply to build knowledge management internally. That can then eventually run on its own. And good consultants also take that as a goal, that they ensure companies can eventually continue on their own. And I would definitely recommend that. For the start, that means simply initiating this transformation process, which is actually happening in all companies right now, professionally, and then continuing on your own afterward—that, I think, is the right path, because then you can much better decide: What tools can I build in-house with my specialized AI team that I've set up, and which tools do I maybe prefer to buy externally? And even in that case, the specialists from the AI team also know: What is actually a tool that really helps us? Where might the dangers be? And accurately assessing that error rate—that's the kind of thing.
Lisa: Yes, or also saving the time of not having to test things so much.
Max: Exactly. Yes, yes.
Lisa: If you'd like to dive deeper into AI applications and exchange ideas with other sellers about AI in everyday life, then be sure to check out the eBay Community or our eBay for Business Discord channel.
Lisa: Max. I think some people also sometimes have understandable reservations when it comes to the topic of AI and AI. We've already said: "Hey, why should I teach an employee a tool that eliminates that person?" How do you perceive skepticism, maybe in your podcast, in the industry, or among people? And how do you respond to these doubts?
Max: I think companies can actively do that by empowering their employees instead of replacing them. Employees here are not cost centers to be replaced by AI; rather, employees can develop entirely new business cases, achieve very different standards in efficiency and new ideas, if they are enabled to work with AI. And I think we should view AI as an extension of our human abilities and not as an enemy.
Lisa: So more additional workforce and not instead of.
Max: 100%. That's why my podcast is called "Kollegin KI," because AI appears as a colleague and not as a replacing boss or manager.
Lisa: If you look into the crystal ball toward the future, say one or two years ahead. What do you think AI will change about retail or about us?
Max: I'm regularly asked about the future, and I keep telling the same story about the last time I answered that, in 2018 for the magazine of TalkWalker, the social listening tool, when they asked: "What social media trends do I actually see for 2019?" And I said: "Wow, definitely employer branding, campaigns, labor shortages, etc., huge." And what was 2019? Corona, and the trends were video calls. Yes, and then I said I was done making trend predictions like that.
Lisa: Yes, but at least you didn't predict a global pandemic and then stop thinking altogether.
Max: That would have been wild if I had done that. But no, I always say: "Stay flexible, try to be as flexible as possible." And that's basically it. All the studies out there on AI confirm that. The companies that benefit are those that position themselves flexibly enough to respond to rapid changes. That means no 20-year contracts with any vendors. New tools are constantly hitting the market that are worth testing somehow. That means looking at how I build my architecture when I set up AI systems internally so that I can perhaps change them again relatively quickly? We need to learn to handle it more flexibly. And that's, I think, the best advice I can give. What happens in e-commerce? Yes, I think the trends are quite clear. AI will be used much more in communication. That sometimes works better, sometimes worse, to be fair. There will certainly be much more efficient supply chains, because AI can "predict" a whole lot and can handle large data sets much better than humans. And I believe that if we use AI correctly, there will also be far fewer careless mistakes. Because AI doesn't make careless mistakes. That's a human problem. And that's why I say this use case of feeding human content into AI and having it looked over is actually a very exciting thing.
Lisa: And maybe to wrap up, if I finish listening to this podcast and then head back to my laptop. What's the first tip you'd give sellers?
Max: What you can do after this episode. Especially if you have various questions. That will certainly be the case. For example: "How do I train a LoRA now? What is a low-rank adaptation model or how do I use a chatbot? How do I build a custom GPT? What is a large language model and why is it nonsense for me to spend money to somehow get my products into it?" These are all questions you can ask AI. So if, after this episode, you open the chatbot of your choice. That can be ChatGPT, Perplexity, Claude, whatever you use, then you can ask these questions there too and at least get a starting point for an answer. Of course, they're never 100% correct, as I've said. So AI output is not always error-free, just like human output isn't. You can get just as bad advice from people as from AI. But it's definitely a first point of contact and can show you a lot. And I think I'm living proof of that. After all, I learned a whole lot about AI through AI, and now it's kind of my job.
Lisa: Now it's kind of your job, and you do it excellently. And I think we've given the listeners a few homework assignments. And if you take one thing away from this episode, then hopefully that AI doesn't necessarily have to be complicated and can really move you forward in everyday life. Because even small, well-chosen use cases can already be enough to save noticeable time, automate routines, and boost your own creativity. A very, very big thank-you to Max for the exciting insights and the practical tips. We'll put all the links and further information in the show notes as always. And if you liked the episode, then subscribe to "Alles top. Gerne wieder!" so you don't miss any future episodes. And we'll hear from you next time.
Max: And if you want to learn more about AI, including AI in e-commerce, then feel free to listen to the OMR podcast "Kollegin KI," hosted by me with exciting guests, if I may do a little advertising at the end here, dear Lisa.
Lisa: Of course.
Max: Yes, then feel free to listen! Everywhere you get podcasts. New every Tuesday.
By
1 comment
warenweiser
·8 months agoYes, AI doesn’t make careless mistakes - but sometimes it makes major errors because of incorrect data. Just because many data sets exist doesn’t mean they’re correct too. As long as a human still looks over it at the end, you can use AI, as long as the human is competent in their field in the end.