And now, of course, we can't do this podcast without talking about AI. By 2025, it's certain that if you don't have AI in your processes, you're dead. The thing is, the goal of Chat GPT or LLM is to please as many people as possible. There are lots of tools out there today that we've been integrating gradually before Chat GPT arrived. The first thing you need to do as a retailer is to hybridize your solution with a tool that will optimize your product pages. You see, in the future, you can have fun with AI and potentially need to do less and less of that. Without having to go through ultra-complex rules of if the product is this, then show that product. It's done automatically by visual intelligence. AI that interacts with someone and at the end asks them if they want to be kept informed of new product launches, an action to add the person to a Klavio list, for example. Hello, you're listening to Lodely Talks, the podcast that talks about eCommerce, quite simply. Once a month, I welcome an inspiring figure from the French eCommerce ecosystem for informal, friendly discussions on topics that fascinate them. The aim is to decipher eCommerce trends and share practical tips to make your eCommerce shop a success. I'm Joseph Aubry, co-founder of Loyoly, the loyalty and referral platform that allows you to engage your customers through more than 50 mechanisms. Sharing, user content, customer reviews, and much more to increase your LTV and reduce your CAC. If you like Loyoly Talks, follow and don't hesitate to give us 5 stars on Apple Podcasts or Spotify to support us. Enjoy the show. Of course, we can't do this podcast without talking about AI. While preparing for this interview, we identified several types of AI, so I'd like to go over them with you to see how they relate to the use cases you presented. Which is the most relevant in which context? I think this is a subject that listeners don't necessarily have a lot of visibility on. So, we have predictive AI, visual semantic AI, and generative AI. Could you quickly explain the differences between these types of AI and what they are used for in the customer journey? Sure, yes. I think you've hit the nail on the head. Sometimes AI is a bit of a mystery. We don't really know what it is, what it's used for, or how it works. So that's what we've tried to clarify at Nosto, to show how we use AI. We actually use the different types of AI you mentioned, which will serve the customer, or rather the end user, the final customer, but also the eCommerce, and the operational teams. First, you mentioned predictive AI. Predictive AI allows us to analyze a vast amount of data, draw conclusions, and identify trends in order to predict future purchasing behavior. In very concrete terms, this will enable an eCommerce brand, for example, to see which visitors have high potential and to create a segment of high-potential visitors. So basically those who are HPV. Exactly, HPV, quite a few. That's not bad. Basically, those who are close to making a purchase and, once again, triggering the appropriate mechanisms. It could also be based on weather forecasts for the next week or two, for example. If it's going to be very cold, I'll suggest a certain type of product because that's what sells best overall during these weather episodes. Interesting. So that's one thing about predictive AI, and then we have semantic AI. This is really useful for on-site search. It allows you to understand the user's real intention when they type in a query, even if it's complex. Today, users tend to use searches on eCommerce sites in the same way they use Google search. So sometimes it can be extremely complex, but that's where semantic intelligence comes in. It allows us to understand the user's intention and therefore provide relevant results, even if the query is complex. And that's really important because a study we conducted showed that 70% of users go straight to the search bar when they arrive on a website, and 80% of them leave the site if the search results are not relevant or effective. So that shows the importance of intelligence like this to provide ultra-relevance. Then we have visual AI. I'll try to keep this brief, but visual AI is more of a lever that will benefit teams because it will enable them to mimic human vision. It will be able to recognize a style, a pattern, an emotion, a shape, or a flower within an image. A flower, for example. And in very concrete terms, for an eCommerce merchant, I think it will be particularly interesting in the fashion sector, where it will enable visual merchandising, i.e., showing products that are visually similar without having to go through ultra-complex rules of if the product is such and such, then show such and such a product. This is done automatically by visual intelligence, which will match products that are similar or visually different according to the merchandiser's wishes. Okay. And finally, generative AI, which is what everyone is thinking about, I imagine. Exactly. It's the one that everyone has been talking about a lot recently. It's a type of AI that we've also integrated into our system. The idea is to streamline the work processes of eCommerce. So, for example, we use it to generate automatic lists of synonyms for searches. Okay. So I think any eCommerce retailer who has worked a bit on improving the relevance of their search results knows that it's extremely time-consuming to analyze all the pages, all the queries that return zero results, and then decide which synonyms to use, and so on. What AI does isdoes it for them, providing a list of appropriate synonyms, which they can then accept or reject. They retain control and can accept or reject a suggestion. And it's really interesting because, in fact, the first three types of AI have been around for 10 years, if I understand correctly, and it's generative AI that's a little newer in this field. And from what you're saying, generative AI is used more on the merchant side to save time than on the consumer side. At least, that's the impression I get, because generative AI today isn't powerful enough, or at least we're not confident enough to let itgenerate results for users directly. It's more interesting for us as a company to control it a little and save time internally. I think there will always be control mechanisms; I think it's essential. That's the developments we've made. I think that we're also going to move towards a lot more things that will be, let's say, customer-facing or visitor-facing, particularly with the development of technologies such as conversational AI, like ChatGPT, where search is being redefined a little. So I think we're clearly heading in that direction. And actually, yeah, that was a question I'd had in the back of my mind for quite some time. In the era of Chat GPT, where, as I saw in a LinkedIn post this morning, you'll soon be able to integrate your Shopify checkout with Chat GPT. In fact, you can buy from Chat GPT. My first question about this is, have you guys already managed to structure your product pages and recommendation blocks in such a way that they are optimized for generative search engines? I don't know if that's how you say it, but we understand each other. It's up to you how you want to look at it. What's certain is that these are topics that interest us. We have a team, we have a lab that keeps us up to date on everything related to artificial intelligence, so they're very interested in all these topics, including KAI agency, really all of these topics. So, of course, these are things we're thinking about. Today, it's not something we offer directly, but of course it's something that's at the forefront of our teams' minds. Yeah, and you mention AI and teams, so I have a couple of questions about that. On the one hand, internally at Nostaud, and more generally in companies in general, what are some of the barriers to adopting AI? I don't know, sometimes some of your clients are afraid of being replaced by a super AI merchandising agent that will do everything for them. Do you see behaviors like that today among your clients or even within your own company? I think that AI, especially with Chat GPT, was a bit complicated to use at first, but I think that's changing because we're realizing that it can't really replace the human brain, at least not yet. However, it does provide a significant boost to productivity. So there you have it. I think that at Nostto, at least, it's very uncomplicated. On the contrary, we are very encouraged to use everything. It might be a little strange for a tool that has been rooted in AI since the beginning. I mean, for us, AI is not a new topic. Nostto was born with AI integrated into it. So at Nostto, we are very encouraged to use all forms of AI that can improve our productivity. What do you use, for example? I use Chat GPT quite a lot. Yeah. Mainly because I write a lot of content in English, so it also allows me to do an initial check on my writing before giving it to the content teams who are working on FRED, for example. It's not Chat GPT that writes for me, but it does allow me to have sentence structures that are often better than what I could have done myself in English. So for me, that's the main benefit, and then I use it for lots of other things, but it allows me to summarize long documents, for example, and things like that, and therefore move forward much more quickly. Yeah. You still have to proofread quite a bit, though. Yes, yes, yes. But still, there's a lot of work that goes into it, which is not insignificant. That's for Nostto. But you were talking to me about the client side, and I have a lot to say about that because I've worked on the client side, so I've also had clients that I've supported in their personalization strategy. And the first obstacle I often heard was a slight concern about not knowing how to control or master what the AI will actually deliver in terms of results, and also not understanding how it works. Typically, for example, a retailer might want to personalize the products on their category page, so they want to show personalized products to Florian, for example, and push products that are red or floral for me, but they still have very important profitability goals, so they can't just say, “Alia, go ahead and show personalized products at the top of the category page.” It's too strategic to say, “I'm going to leave it completely open.” What they can do is use control mechanisms to add a rule to their merchandising to say, “I also want you to take into account the products on which I have a better margin, for example, in this weighting that will allow you to sort the category pages.” What will work better afterwards, and what I always tell them, is to test this rule with my margins added, or this rule where I basically let the AI control everything, see what happens, and see what performs best in the end. And in some cases, it may be linked to a single factor, in some cases it's related to controls, filters, merchandising rules, and so on. For me, what's important and what I was trying to tell them is that good technology today must integrate AI, but ultimately the strategy is still driven by humans, so they will actually integrate all the control mechanisms that allow them to guide the specific strategy of an eCommerce merchant. Yeah, totally, yeah, that was interesting. I had another point. Yeah, go ahead, go ahead. I also think there's something we hear a lot, which is a bit of a fear of the black box. That is, we don't always know why a product has been recommended. I'm very into product discovery, but why was a product recommended? Why is a particular product in a particular position in the search results, and so on. And sometimes you need to be able to explain that because other people in the company will say, “Hey, that's funny, I typed in this search and I saw that product there, I don't understand.” So it's important that the person in charge of merchandising is able to say, “Well, it's for this reason.” So typically at Nostos, we've implemented what we call inside scoring. We show why a particular product has been placed where it is, what factors were taken into account in displaying that product in that position, whether it was the merchandising rules, AI, etc. So it's also important to be able to explain these elements to the user. And that way, you can clearly understand the whys and wherefores of each recommendation that is made and avoid the black box. Exactly, we are able to say what influenced the decision. It's the same thing, it's a scoring system. What influenced the weighting and ultimately the promotion of this product rather than another. And you can change the weighting if you want to. Okay, absolutely. We always review the merchandising rules or filters, and we can add lots of things. Okay, okay, great. Okay, I had Camille from Nostto on the podcast last week, and we talked quite a bit about AI, since that's kind of Nostto's positioning with personalization. So one of the types of AI that we're all a little familiar with now with Chat GPT is generative AI. What's your relationship with this type of AI for producing SEO content? Well, I've been against using it for a long time. It has to be said that the quality has improved a lot, but you never know, you can never rule it out. We've already done a lot of testing and we've seen that for us, at least in the way we were doing it, there were ranking issues. The thing to understand is that LLLLLLLLLLLLLLLLLLLLLLLLL mainly looked for training sources on existing content. Ultimately, if you produce a text, it's going to be a text based on existing content. And there's another thing, which is that the goal of Chat GPT or LLM is to appeal to as many people as possible, so it needs to use language and responses that are average. You see, it has to appeal to everyone. That's really the goal. That's the reward for Chat GPT. That's how it works. When we're happy, we give it a thumbs up, we train it, we give it a carrot. So the answer has to be as appealing as possible and average based on content that's already online, otherwise it wouldn't rank well. Today, what works is generating content, but adding things that others don't have or enriching it with your own databases, your own customer questions, and so on. So that works. That doesn't answer your question about whether we use it. And in fact, today, I have people in-house who are passionate about it and will always respond much better than any AI. A quick shout-out to Olivier here. And there you have it. Passionate people convey technical information, that's for sure, factual information, but also emotion. And it's true that for now, emotion on the AI side is, well, it's not exactly crazy. So we're not there yet, but of course we use it to help us write plans, structure content, create product pages, and lots of other things, but we use it a little less for high value-added content. Okay, that makes sense. It's interesting what you said about the reward aspect of the ultimately mathematical model of AI, especially with the latest updates we're seeing today, where you have screenshots on Twitter of people saying “I am God,” and so on. And Talia says, “Yes, of course, Gregory, it's you, you're the chosen one and everything you want to be.” So it's true that it can quickly go off the rails depending on how it's set up on their end, towards something less authentic in the end. Yes, that's right. And it's content that you can generally find. If you make your request in a fairly basic way and your neighbor makes their request in a fairly basic way, you're going to end up with content that is very, very similar. I don't think that's relevant. You have to add your own touch, and only a human can really do that, or at least a good writer or someone who's passionate about it. Yeah, definitely. And today you have traffic coming from Chat GPT. It's funny because yesterday we recorded with Laurent, whom you know well, Dupagnier, who we're also passing the project on to, by the way, and we were talking about this whole aspect, and I said that I'd installed a little something, so you can check if you're getting any traffic with Google Analytics. We use Google Analytics and, at the same time, we've linked it to another tool called Perriple, which checks the crawl. So it's the bot and Google Analytics that give us the traffic, the real visits. So that allows us to make a correlation, I'll say correlation, it's very frowned upon scientifically because in fact, there aren't that many links, which is very frowned upon scientifically because in fact there aren't that many links, but it allows us to see whether the LLM is working well for us in recording information and whether it is generating visitors on the Google Analytics side. Today, we're making a lot of noise about it, but in reality it's one in 100 for me. Okay, fine. But do we need to keep an eye on it? Yes, definitely. Yeah. We've got our hands on something. It took Google years to go from one percent to 40 percent for me. LLM will definitely send more traffic. However, is this the revolution that everyone is waiting for and talking about? I think we're a bit caught up in the hype right now. Okay. On the part I'm talking about, traffic generation, ALM optimization, and so on. In any case, we're making a lot of noise about nothing for now. Yeah, you think that the day people are going to buy in Chat GPT is maybe not right away. Yeah, that's right, I mean, the more the engine improves, the more relevant the response will be, and ultimately, will the person go further to get more information? No, if they get the answer right away, they'll close the tab and move on to something else. I'd be surprised if they clicked on the little result above, you know? Well, let's see how it evolves. On the other hand, when it comes to purchasing, yes, we need to keep a close eye on that because it was one of the advantages of having an eCommerce site, being able to control customer loyalty from start to finish, and so on. That's a good question for you too. I mean, if the purchase is made on the LLM or if the purchase is made on marketplaces and if the purchase is made in TikTok shop, ultimately the feed is harder to understand. So we'll have to develop it in any case. So I'm not in a rush for that to happen, but eCommerce is definitely evolving and we're going to move towards this type of purchase, I mean, delocalized, where you become a rapIM and you have to feed several sources that can sell directly, you know, your products. Yeah, totally. And still with this SEO in mind, do you have any idea what JI0JS0 is called? Yeah, there are lots of terms, SE0AIAISE0. And so, can you understand how they analyze your site? It's LLM. So there are some general rules. Basically, when LLM needs information, it goes and crawls a search engine. So Bing for OpenAI, for example, will take a certain number of results from the first and second pages. Generally, to come up with an answer, let's say that to get 100% of your answer, it will pick from all the sites it visits to reconstruct a complete answer. Okay. So it takes 20% from the first result, 5% from the second, and so on. So first of all, you have to work hard on your SEO to be on the first page, at least. So that's important. And then finally, since it feeds off content it finds on the web, if we say that the best jewelry store is Okara, the more often we say that it's my competitors, the more when you search for “what is the best online jewelry store,” Okara will come up. So there's a notion of popularity which, unlike SEO, where popularity is links, you see, it's the number of links that point to you. Here, popularity translates into the number of mentions of your brand. That means you also have to think about your content. Generally, when you're at home, you're not going to say “Okara is the best online jewelry store” at the top of your content, so you'll have to think of cool ways to do it. Instead of saying “at our store,” you'll say “at Okara,” “at your Okara jewelry store.” You may need to modify your content a little so that your brand appears a little more, shines through a little more in your content. You see, before you would have said “on our store” or “on our website in our categories,” and you would have stopped there. Now you'll have to steer it a little more, more often than your competitors. And then you have to be aware that the LLM doesn't read the content like Google does. Google needs content that is quite long, very comprehensive, and so on. What we've seen with LLM is that it likes content that is short and concise, with small paragraphs. There's less for it to digest. Yeah, that's right, and it's easier for it to condense sentences that are already condensed, you see. So the FAQ sections you might have on products, all that stuff, is a godsend, so that's what you need to focus on. I'm giving you the rules as they stand right now, but they change every week. However, if you've already done your SEO well and you're well established on Bing and so on, you're on the first page, you've got a good chance, and then you can modify those little quotes, build your reputation, be on trusted sites. It's not that different from SEO, really, there are just a few small nuances to consider. That's really interesting as a starting point. I think it'll be of great interest to our listeners. Now, you mentioned artificial intelligence and automation earlier. It's hard not to talk about it on the podcast. You explained that there were AI solutions that would improve the overall customer experience and personalization to improve conversion. So that's what I think you were proposing with your startup, if I understood correctly. Yeah, we were actually doing a lot of data in terms of development, and we were developing at the time—we won't go into that—but based on visitors' journeys, what is their actual intention? So we have a lot to do on the AI side. You mentioned talking to your customers in a unique way. If you can also make a visitor's journey unique, you can monetize the traffic you have, which we talked about, and you buy, you buy. More and more expensive, so there you go. On AI, I think we got a big boost at the beginning with the latest advances. First, there's the AI that we use at PrestaShop. We need it internally, and we use it at all levels, but also in our development processes. We don't necessarily generate code with AI, but our tools are equipped for it. We are increasingly seeing that our support tools, our development tools, our collaborative document sharing tools, etc. are becoming more and more hybridized with AI. So the challenge for everyone, including PrestaShop itself, is to use AI effectively, and therefore the challenge of training will remain, because people are at the heart of the system and at the heart of the process. We saw the first use cases when GPT version 3 was released. The first use case came a week later on PlayStation, through a community that I would like to commend for being very responsive. That's the beauty of our model. Three modules allow you to optimize your product pages, and these are the first use cases in SEO for acquiring free traffic. So, the first thing you need to do as a retailer is to hybridize your solution with a tool that will allow you to optimize your product pages. You can do this in bulk with the right modules for your solution. Of course, there is also the support aspect for the initial interactions, which is becoming increasingly effective. For example, I think that in terms of internationalization issues, AI is increasingly becoming an accelerator, so of course you're going to have to adapt your eCommerce site to payment and logistics solutions, which we've already talked about, but on the product pages, how do you generate a product catalog that's optimized for the country? Well, there you go, I've got an accelerator. So it's an extremely powerful factor in terms of agility, but I'm also going to get it through tools that are becoming increasingly powerful. We were talking about the engagement of our customer base. Scenarios are now based on AI that learns from your database, but ultimately from merchants and their purchasing practices. We developed a start-up and developed our models, but now they are available in CRM solutions such as Klaviyo, which will have scenarios that adapt to the context of your merchants based on what they learn and what the tool learns about purchasing practices on your site. That's great, because succeeding on the web means having the one-to-one interaction that you have in the physical world, where you recognize the person. That's the holy grail, and we're at the beginning, but we're getting there. Yeah, there's still a lot to do with all that. It's pretty impressive, and it's true that we too are trying to get to grips with these issues to further personalize the overall customer experience, make it more consistent, and also increase its relevance. Yes. That's because in the flow of messages we receive, the various and varied requests we get as consumers, we feel a little bit like we're just a number. So I think this is a real step forward for retailers, thanks to this personalization shown to each customer, who is recognized individually at their true value, simply to improve their satisfaction. We've talked quite a bit about examples of automation in the last two or three questions. What kind of automation can brands implement? Is it when you always get the same question? Instead of asking your question, you click on the question and the answer is displayed. What is that? That could be an option. It's something that's mainly done on chat, where you go to the chat and you see the most common questions, and it works very well. I haven't given any exact figures, but in any case, from the accounts I look at from time to time on Gorgia or with people I talk to, I'd say easily four-twenty-five percent of requests, or actually the person clicks on that button and it doesn't create a customer service ticket, it does things like how to choose my product, what is your return policy. These are actually pre-sales questions that people often ask, and there isn't necessarily any added value for brands in answering them, except of course that it can convert, but it's in their best interest to allow customers to find the answer themselves directly in the chat while they're on the site. So it works very well. We call it “quick with think,” but basically, you click and you get your answer. And sometimes there are even intermediate steps. You can also create more or less complex things. I'm an existing customer, a new customer, and depending on that, you can go pretty far with it. There are few brands that go far enough in the complexity of these options, but you can go very far. You can even make API calls to check which subscription a person has and, depending on the subscription, which options they have. You can go really far. For me, it's something that's really underused because, at the end of the day, consumers don't necessarily want to contact a brand for anything. First, they will always prioritize sales options, but it has to be easy to find and access the information. So, to come back to your question about automation, there is the “sales” part of allowing people to find the answers to their questions themselves with what you were saying about chat, quick replies, etc. It could also be tracking and managing orders directly from the chat or the center, as we were saying earlier. Otherwise, in Gorgia as a tool, you'll have automations that aren't necessarily automatic responses. It could be automating things to make customer service easier. It could be macros, services. It could be macros with variables or even macros that perform actions. So, for example, you could have a macro to edit the delivery address, or you could say, “We've updated your delivery address,” and when you send it, you edit the delivery address directly in Shopify from Gorgias, for example. So macros are going to be a really big part of automation in the sense that you're going to save a lot of time with macros, and then there's automation via rules, so basically it's going to be a bit like triggers where you say, “When I'm asked this, I'll tag the ticket in this way,” a question in French for the French team, a question in English for the English team, or maybe I have a question from a VIP customer, so I'll send it to a specific team. It's a pre-sales question, so I'll send it to the team of advisors who really know the product well. So there are all these types of automation that will allow you to prioritize and manage customer service better and faster. Okay. It can also be automatic responses. I was talking earlier about business hours when you have a lot of customer service teams working Monday through Friday from 9 a.m. to 5 p.m. You can have an automatic response that says that outside of business hours, if someone asks a question about a return, they will be directed to the help center on the returns page and told that if they have any questions, they can contact us again. There are lots of other things like that, and now there's also conversational AI, although I don't really like to call it conversational because thethe goal isn't to have AI that goes back and forth with a customer 10 times, but rather to be able to respond concretely and in a timely manner to the right question at the right time without necessarily having to have a conversation as if it were a human. But the reality is that this is going to become the norm very quickly. We're launching AA Agent on July 1st. I don't know when the podcast will be released, maybe it will already be live. But we're releasing it soon and we already have 400 customers using our AIGENT. Some of them are already able to automate 20% of their customer service requests with something that is still in beta, so in fact, more and moreany brand will be able to automate a large part of their customer service thanks to AI. This is still in its infancy, but it's really something that's open to everyone, and there will be no more excuses for not being able to respond quickly, in a personalized and optimal way, regardless of the size of the brand. It's more limited to brands that have large teams, high volumes, and are forced to optimize this. Now, anyone, even a solo finder, will have access to tools that enable a great customer experience. And speaking of AI, what it will actually do is greatly reduce barriers to entry. Mechanically, as I was saying, the level of customer service will increase dramatically. Can you tell us a little about the key features of your AIG? Absolutely, but it's still something to be approached with caution in the sense that it's a really incredible opportunity, but for me, it has to be done well, and that's not always easy. Yeah, that's clear. For example, if you take chatbots, I've already interacted with lots of chatbots, and the first thing I do every time is speak to a human. I need to talk to a person because they usually respond with things that are completely irrelevant. It's hell. It's great to have AI that can respond to customer service requests, but for me, what's much more important is making sure that it's going to respond with something relevant and that there's always a human being available quickly behind the scenes. But the opportunity is to make sure that questions that don't add much value, like “where's my order,” can be handled by AI, freeing up time for humans to handle more human conversations that make more sense. And how does it work? Is there a Gendt? Is there something that is truly autonomous and connected to all the tools and will respond autonomously, or is it more of an aid for the support agent, who prefers to ask questions and then has the support agent validate, in quotation marks, since Lea has given them a pre-response or something like that? What we're developing right now is a way to respond concretely to customer requests. So it will be based on both the history of customer service conversations in Gorgia. Okay. On Shopify data, so the order and customer data that is available in Gorgia. The help center will be the main source of information for Hygel to work well. It's important to have a solid help center, not necessarily the customer-facing help center on the website. You could have, I don't know, twenty-five help articles on your website, but if you have a help center with two150 articles for your AI with all the internal processes and all the information, so it's really key to have articles that the AI can use as a basis for answering the most common questions. Basically, the goal is to be able to answer certain questions, or even transfer them to a human, but after gathering information first. Then it's a choice of whether I'm going to automate a large part of customer service. Do I just want AI to handle certain types of questions?-for example, I don't know, someone who makes a return request, does the AI first check whether the order was placed less than a certain amount of time ago or whether it is less than a certain amount of time ago according to our procedure. Is the product in good condition? Does the person need to take it to a pickup point or send it back to us directly? Whatever the different steps are, once all this information has been collected and transferred to a human, it depends. The beta version is for responding to requests, but very quickly, within the next week, it will also be able to perform actions, particularly on Shopify, such as editing an order, canceling an order, or refunding an order. I know we're going to start with loop return, so return requests. Very quickly, these will be actions via the API. So what I imagine is, for example, the AI interacting with someone and at the end asking them if they want to be kept informed of new product launches, which would trigger an action to add the person to a Klaviyo list, for example. So for me, the real potential is also in performing actions, always with a view to doing something that the brand wants and ensuring that no actions are taken that shouldn't be taken or anything like that. But in terms of possibilities, it will really allow us to automate a lot of things so that humans can spend their time on things that make a lot of sense. Another big advantage is that it allows you to manage things with the full context of the customer, so all the conversation history, all the information, no human error of forgetting something or forgetting the context of something 24/7. So in relation to that, we were talking about response time earlier, even if it's not the ideal response if you configure it inin a way that still responds to something relevant, and at least it's super fast and you're assured as a customer that it's been transferred to the team, etc., and they'll get back to you as soon as possible. You can even ask the AI to tag it as urgent depending on certain requests, etc. It's really made this work easier, and it's also multilingual. Tomorrow, you're a French eCommerce merchant and you want to launch in Germany and Spain. You don't necessarily need to recruit a team that speaks those languages right away. You have young people who can handle the initial requests as they come in, and you can test things out at that level too. Yeah, 100% sure, it's really going to be a game changer in organizations. It's going to be a game changer because there will be less need for humans to manage large peaks in demand. Typically, there are always lots of brands that tell me they're swamped after Black Friday, which can require up to 4 people. In fact, this will allow them to manage the influx of requests much better and, as we said earlier, it's accessible to anyone. So now, anyone can use these technologies that, just a few years ago, were only available to large multinationals with huge budgets that had developed their own LLL model in-house. So now that's the situation. We're for everyone. It's crazy, yeah. No, I really think it's revolutionizing the way we interact with our customers. And in a few months or years, when we contact a brand, we'll also expect an AI to respond, but we'll think, okay, if it's done well, it'll give me a relevant answer, even if it's not entirely accurate. At least I'll get a very quick response, and I know that the AI will understand the context, share the right information, and so on. On the other hand, I would also expect that, since there is AI responding to requests, if I need to talk to a human, I will be connected much more quickly than is currently possible. So there will be both this expectation that maybe an AI will answer me, but also that I can very quickly get in touch with a human if I need to. In terms of response time and resolution time, this could increase quite drastically. All this needs to be balanced carefully to ensure a good experience and that the AI doesn't respond with irrelevant or nonsensical answers. Ultimately, I'm convinced of the positive impact this can have, but I'm also aware of the risks involved in implementing this type of tool. Just like I've seen lots of people who want chatbots in place, but in the end it was an extremely frustrating experience, which wasn't the goal at all. So yes, but you have to do it well, in a thoughtful way, and test the different scenarios and how it works beforehand. Yeah, testing all that stuff is going to be pretty cool, I think. But it'll be interesting to see how you can use it in your processes today. So today, it's not quite there yet, but I imagine you were talking earlier about importing review databases into ChatGPT to ask it for angles of attack or the main issues that came up. Do you have any other examples from Straad-up internally? How do you use AI on a daily basis? There are lots of tools today that we're integrating, that we integrated gradually before Chat GPT arrived. There's one I was telling you about just before, which is improving the quality of a video or photo. Today, we have ultra-high-performance tools, and a year or two ago, it would have been incredible to have that at our disposal. So we also have... What are the names of the tools? We have Topaz, which is a very good tool, a piece of software that has been updated fairly recently. So you have the photo version, the video version, Topaz AI Photo, and Topaz AI Video. And, it's the same thing, they have several artificial intelligence models depending on the types of videos you have, and it's really quite interesting. Then you even have, so these are software programs with licenses, but you also have online versions. I use Créa Point A I a lot, but only in one case. Okay. And that's to add texture to photos. Ah, okay. But it's not just about improving the quality of a photo. It can recreate material based on a prompt. And it's really impressive. For example, let's say you're working with a client who no longer has the pack shots for their products. You can go get them from their website, run them through this type of artificial intelligence, and then you get content that you can use in motion design, which you can reuse for all your ads. And that alone is a technical barrier that has been removed. A few years ago, we would have ended up with poor-quality pack shots in the ads, but now it's completely open. We also have tools that allow us to translate voices in videos, as well as tools that lip-sync the content. So, instead of just a French or English version, we'll be able to offer several languages. And that's the same thing, it's a technical barrier that was removed a few years ago. Sometimes we had to do voice-overs, but that was a bit cumbersome with actors, or we had to create two pieces of content, and today it's much easier. It's incredible, really. Are there any other areas within Stripe where you use AI for process optimization or to find ideas? We also use AI, not to find ideas, but more to find things that we don't have from our potential customers. Okay. With generative AI. So, rather than looking for them, sorry. Stock footage is actually images or illustrations that you have in banks, whether they're photos or videos, and we're going to use Midjourney to find additional content. And that's exactly where all the work has been for the past two weeks. I've been working hard on this, and the main issue is how to integrate our clients' specific products into the AI results. Yeah, because that's the barrier. In other words, you either have content that you've created yourself, or you ask someone to add a tree in the background because that's what you want. We add sunshine because it's summer, or vice versa, i.e. we add pack shots to existing content. Exactly, or you might say, I had this case, summer is coming, they haven't had the means to go and do a shoot on the beach yet. First of all, because the weather isn't good, and because you need time and money to go there. You have to go there. That's it, when you're in Paris, it's not necessarily easy. And so, quite quickly, we're able to take an illustrative photo with artificial intelligence to integrate our product on the beach quite easily. And actually, I've been taking photos for quite a long time now, especially film photography. So, it's a bit the opposite of that. But it helps a lot with AI because the hardest part of Midjourney is creating realistic photo content, doing very, very distinctive things, like steampunk. You see worlds that are very codified, which is more accessible, but creating very realistic, photo-realistic content is a bit of a headache. So, by giving it the type of camera you use, your composition, the lens, the grain you want, the type of film you want, it gives it a lot of indicators to generate content that will be much more realistic and therefore usable in ads. Okay, that's cool. Are there any other tools we haven't mentioned that might be interesting to explore further? AI is crazy. Yeah. No, but I have another really crazy one that's really going to appeal to creatives who are starting to get really into After Effects. It's actually a plugin for After Effects, which is a really advanced piece of software in the Adobe suite for creating motion design and SFX animations. So, it's actually a plugin that bridges the gap between ChatGPT and After Effects, integrating ChatGPT, or rather OpenAI ChatGPT, into After Effects. It's called Clut GPT, and with this tool, you can create expressions and scripts directly from the software. Typically, there's one that imports subtitles into After Effects. To be able to create completely customized animations, the basic way to import them is as a file, a text layer. And I've written a script that allows us to import each text track onto unique layers, so we can easily rework them later. And I did that in five minutes with OpenAI in After Effects. And it's the same thing, before, you had to be a developer to be able to do this kind of thing, and a good developer specialized in After Effects. So, it's not exactly common knowledge. That's for sure. It's crazy, really interesting. It gives me lots of ideas. It's endless. I spend nights testing tools. So what's a little frustrating is that you have to test a lot of AI tools to find the right ones, the right way to use this or that tool. Yeah, that's often the case too, it's like you said earlier today, if you ask someone to do something and you're not happy with the result, all I can say is use this film, this camera, and so on, but it's still, that's the real thing, that's what will give you a result that's really good compared to what you're expecting. It's more about how you use it than the tool itself. Yes, exactly. The tool is just a tool, it's the same as any other tool. And that's precisely the problem: once you've mastered the tool, how do you get the rest of the team, the rest of Crea, to use it in the same way? And there, you see, on Midjourney, I've created a prompt formula that helps generate prompts with drop-down menus, you see, you choose your Lens, you choose the effect you want, and with the help of a few GPT chat prompts for the description of the universe, and so on. Behind the scenes, it automatically generates your Midjourney prompt, which is well optimized based on the various tests we've done beforehand. It's crazy. Yeah, in fact, your job is changing too. I'm going from creative to engineer. Exactly. But at the same time, it's actually really exciting. Because I think it's something we need to be interested in. It's going to become essential. I don't think it will replace anything. At least not today. There's no magic AI that will integrate into your product and give you something that can be used as is. Yeah, of course. I know I've talked about this a lot with Cabaïa, who are very advanced on these topics and very interested in all the opportunities it opens up. But the problem is having something that can be used by a brand. And that's very complicated without intervention. That's what we were saying again, you have to be able to use the tool well. And that's something you really have to learn. You have to spend hours on it. I don't know how many hours it takes to master a tool. I think it's 10,000 hours or something like that. There are figures like that. Indeed, mastering a tool is what it's all about, whether it's Inia or something similar. Yeah, exactly. The same idea. And then, disclaimer, if you have really large volumes, you can use predictive tools that will trigger you, for example, based on your customers' repeat purchase habits. Okay. That doesn't work if you have small volumes, but it works well if you have large volumes. Obviously. And with Advance Analytics, you can train the algorithm to explain, based on the products, what the changes are in terms of repeat purchase. Okay. If you sell treatments, don't use this trigger because the basic trigger doesn't take into account whether the customer bought something three months ago or a month ago. So it's stupid, the alerts won't make sense to the customer. Okay, okay, I understand. But you see, in the future, you can have fun with AI and potentially need to do this less and less, because what could be better than AI that's really smart and knows the normal repeat purchase time for people who bought a one-month treatment, you see? Yeah, that's clear, yeah, yeah, it's true that it's going to be a pretty interesting development. Is Polar Analytics, for example, a tool that allows you to do things like that? We use it less, I think Polar is more on the Acquise side, maybe. On the analytics side, you know, post, it's really comprehensive. Yeah. But that's more on the side of, yeah, once you've made the purchases, once you've done your CRM work, you're going to look at what's happening, how your cohorts are behaving, and all that. I'd say that in the future, either Klaviyo will make progress on AI and that will do the trick, or you'll have tools, you'll have to work a bit with Quantify or things like that. Yes, there's Quantify too. They offer this kind of approach, or maybe you had it at the time, and there are others. Okay. But we do it less with our clients because often these tools weren't developed enough, and I think the volumes are less suitable. If you have 15 refs in real life, that's all you have when you do your analysis, and I think you have a good basis. Okay, it'll be interesting to see how this whole area evolves, which is a bit predictive too with everything that's happening in AI. Actually, eCommerce is a really broad topic. Shopify, I'll try to be fairly concise. Shopify has developed several proprietary systems. What does it do? Yes, commerce. There are several. There's one. The first one that came out allowed you to create automated product descriptions. Okay. So basically, how does it work? You submit a mini description and they develop it for you, adding a few more keywords, etc. You can suggest them for SEO, by the way. Yes, I can't say exactly what the algorithm takes into account. Okay. But in any case, it adds qualified words. For example, on Shopify, you can ask them to make you some test sites. The best known is actually snowboard sales. Oh yes, yes, we know that one. And so, those who know, know. And I think the story behind it is because I'm afraid of Iron Cotton, but it's not like it was the first site on Shopify, or maybe it's because the guy who makes the snowboards launched Shopify afterwards. I don't have the anecdote. I don't have the anecdote. So you told me I could say I don't know, I don't know. And so, that allows you to avoid, well, when you write, when you do a test saying write me a description of a red snowboard, okay, made of polyurethane, and so on and so forth. You write a sentence that's like, I don't know, 20 words, and it gives you a description of 150 words. Okay. So maybe it goes from 20 to 150 words, it's not going to be mega high quality, but there you go. But in any case, it works pretty well, I've tested it, it works well. You can also generate AI on product images, especially for backgrounds. So, for example, you have your images in PNG format. Let's go back to the snowboard. You have a snowboard in PNG format with the background removed. You say, “I'd like to add a background with a mountain, a lake, etc., at sunset,” and it creates a background for you. Oh yeah, like that, yeah, yeah, I don't know because that's already active. So I haven't tested it to be honest, but in any case they released it recently. So there's that. The other thing that Shopify has released is something called Sitekick. I think it's in beta testing. Anyway, they've been talking about it for a long time, but it's not 100% operational yet and I don't even know if you can access it. I haven't checked recently, but anyway, SideKick is supposed to create automatic things for you in your Shopify back office. You say, “SideKick, what's my average conversion rate over the last 37 days?” And it will say, “Your average conversion rate is this.” That's cool. Can you give me the sales stats for Black Friday, which took place between May 31 and whatever. I have a question. Can you create a discount coupon for 20% off a certain product? You do that in a chat. And it says, “OK, I've created a coupon that's active now.” That's it. Awesome. So you have lots of stuff like that, it's really interesting. Even for us, as developers, it's quite interesting because Shopify does very regular updates and sometimes we discover things. Anyway, I was talking about it last night at the party where Harley was. I've been calling him Harry since the beginning, but his name is Harley. And I thought you said Harley, see if you can say it like that too. We'll do that. Let's say you said that. And then I was talking to Younes, who is quite well known and also very well integrated, very, very integrated in the Shopify ecosystem. And talking to his associates, he said it's really hard to keep up with all the Shopify developments. It's really a full-time job. And so, it allows you to be a little more aware of what's new at work, etc. So it's really interesting, and they're happy to develop things without talking about AI specific to Shopify. They've developed AI that allows you to compile data from your site with data from other Shopify sites to be more effective in retargeting marketing. Wow. And I don't know the name of the system, but in any case it has a name, no, I forgot, I can't remember it. But basically, it allows you to take stats from all eCommerce sites, especially those that are similar to your target audience, and allows you to be more effective at retargeting. Okay, that's cool. So I can't say for sure, but I don't think it's active in France yet, like a lot of things, but in any case, that's what's happening, and then on the more general side, obviously on the SEO side, especially right now, there are some crazy things you can do if you want to optimize your SEO positioning. Honestly, you can use tools like Reboom, which I discovered recently. You know it? No, not at all. It's pretty crazy. It's an application that allows you to... No, I'm talking nonsense. It's a SaaS solution. It's metric, it's mathematical. Yeah, sorry, it's a SaaS solution that lets you enter a topic. You tell it you don't have anything to say about that topic, and it comes up with 10 articles on those topics. So it scrapes the entire web and creates articles. You copy and paste them into your blog, and it's pretty crazy. I did some tests for us on CR0 and so on. It's not perfect, I didn't post them because, for image reasons, we thought, well, you can clearly see it's done by AI, so let's not be pigs. But really, if you don't care too much about your image, you can do it, you know. And maybe we'll test it out, honestly. I'd be pretty keen to test it, just to see what it's like. But I think that today, if you really want to make it big, especially if you're entering a competitive market, I would definitely do it straight away to test the tool. OK. Because that's what the blog is for. It's not where you get your most traffic, to be honest. But it does help you position yourself in terms of benchmarks. And that's pretty much what I have in mind. But there are lots of other things too. It's endless. You have the chatbot part, so Gorgias Crisp, which also allows you to have much more advanced assistants. Tomorrow, when you have questions about your loyalty system, you'll have a kind of tunia that will answer you. I don't think so tomorrow. You see, I've guessed what's coming next week. And much more. There you go, I'm sure of it. No, but that's the thing. So, it makes sense that everyone is implementing this kind of thing. So, it's going to be flooded in 2024. Maybe it won't be the year of AI, maybe it will be 2025, but it's going to take off very quickly in 2025.-twenty-five, so it's already being talked about a bit, you know, but there are very clear concrete cases where you have to implement AI on your website. It hasn't happened to me yet. That's why today I'm telling you about these cases of AI. But in fact, we don't systematically integrate AI into our processes today. Maybe we're not, maybe we're already behind, you know, I don't know, but in any case. In 2025, it's certain that if you don't have AI in your processes, you're dead. Thanks for listening to this episode of Loyoli Talks. I hope you enjoyed it and found lots of tips to try out for your brand. If so, become a follower so you don't miss the next one. Spread the word and leave us a 5-star rating on Apple Podcasts, it really helps us out. Finally, if you need to increase your LTV, feel free to contact me on LinkedIn or on our website point I o. See you soon.
.png)
.png)
.webp)
