Hello, everyone. We're just gonna give a few moments for people to join the webinar today. Okay. Hi, everyone. Thank you for joining our webinar today, a GenAI in e commerce. My name is Courtney Christianson, and I work on the marketing team here at Coveo. I'm really excited to be part of today's session, and I'll be taking you through portion of this webinar. I'm thrilled to introduce my colleague, Sergio, who is our senior director of e commerce at marketing at Coveo. He'll be taking you through the top use cases and misconceptions of gen AI. I've a couple housekeeping items I want to first cover. You'll be in listen only mode But we do wanna hear from you. Please use the chat to ask any questions. I'll be moderating the Q and A section at the end of the presentation. We'll be sending out a recording of this webinar about twenty four hours after this webinar today. So please keep an eye out for that email. For those of you joining, welcome to our webinar, Gen AI and e commerce Coveo to Sergio. Thank you, Coveo. Hopefully everyone can hear me well. You know, I'm obviously joined here as well by Simon. Simon is our VP of product for e commerce here at Coveo as well. You know, together, we have a world of experience in the world of e commerce, myself, over ten years now. And a very, very keen interest in AI and specifically gen AI and Simon Simon likewise. So kinda are gonna kinda set the scene and then we're gonna go into into sort of a question and panel format where I'm gonna ask Simon some questions around what he thinks about GenAI and e commerce in this application. Because the main thing we're trying to do here today is discuss yes, the exciting use cases, and there's a lot of hype around GenAI. We are very excited about it as well. But at the moment, it's kind of all the rage and it's everywhere and it's being applied to everything. And I suppose the question that we wanna put out there today is does that Does that make sense? Does that make sense in every situation? You know, at Coveo, we've been, you know, creating applications with AI for about fifteen years. Always adopting the latest and greatest and what's available in market to bring it out and scale it to our customers. And we're really excited about GNII. You know, we we've done our own consumer research, We found out that consumers. It's about fifty four percent of them expect a chat GPT like experience that was their words, to be part of their shopping experience. And interestingly that appetite increases depending how young someone is. So it decreases with with age. But likewise, you know, we're we're a very mature and considered SAS business, and we wanna make sure that everyone is aware of the misconceptions around NAI and where people are using it as an app, you know, to as part of their application where it doesn't really make sense. You would use different technology for that. So Simon and I are gonna dive into that, in in in detail. And the kind of to back up that thought. We're not trying to we're not trying to be party poopers here. We're trying to just be very mature in terms of our application of this new technology. You know, Ghana have said that the investments in GenAI are likely to decrease by twenty twenty five due to the high costs and that those costs are passing the value that GenI can can generate. And I think that is really an important part of what we what we wanna do around staying focused and making sure that JNIIs applied to the right parts of the e commerce journey and and the solution that can be can be provided. So, you know, hopefully, you guys are gonna all go away today learning a bit more about Genio than you did and and specifically its applications to e commerce and where it makes sense. And the really important thing as Courtney already said is please fire in question as we go through. We'll do with the q and a at the end. You know, some of them are very open to to chat about that with all of you. And know, hopefully we can have a nice a nice discussion about that towards towards the end of the session. So I'm gonna jump straight into the questions if that's okay with you, Simon. Sure thing. Nice. So first question is, how do you envision generative AI impacting the future of e commerce search And what are the key benefits it can bring to online retailers and their customers? Sure. So there are two, you know, it's been it's been several months since since the craze started, and now, you know, things have matured a little bit. We saw a lot of, you know, different investigations, some a lot of hype, a bit of And the things that we started to see, first of all, I would see the tactical element that I've been applied, that could help overall, you know, the e commerce experience, but also including product discovery and search, our application where it comes to solving some sort of the data crisis to at least accelerating the data crisis. So we saw, you know, commerce engine starting to offer, products product description helper. So really, again, here, not completely generating the product's description, but helping category manager and merchandisers to create these descriptions. And, obviously, you know, good data in, and good results out so that that will definitely help search in the long run. That being said, it's not necessarily, you know, an application would call it an application not necessarily a feature, but an application that is specific to a product discovery platform. It's more something that you'll find you know, in the commerce engine. And if it was for me as well, I would like to have it more into Tim providers as well, because that's really where the source of truth is. But Tim They're not there yet. So we're seeing, you know, a bit of this. Also, one of the thing that product discovery never necessarily tackle is traffic generation. So mostly, you know, product discovery engine will help a lot once your shopper are actually on-site. But, before that, you know, getting them in in your content is, is always a challenge, show creating rich content generating rich content, is something that definitely, would be easier to do with gen AI. So creating new content so that you have, you know, better SEO ranking and such And, obviously, a product discovery engine that can leverage that rich content, would, you know, would, would have a really good synergy, with that new strategy. So those are two, I would say two things that are applied that we're seeing that are functional today. With Jenny, I even with chat GPT almost as is. Now when it comes to product discovery itself, we see it really with shopper Education. So I will touch that a little bit later, but, you know, I have maybe an example here of query. So I'll just show a few slides. So here's an example of, you know, the type of question that we see and and pretty much show so Cavell is more in commerce we see it in our different line of business. But you can see here on the commerce side, the type of question that we'll answer is not, you know, do you have you know, the red Nike shoes, that Michael Jordan was wearing or something like that, because this is quite easy to answer with semantic search and and cognitive search, what would the type of question that we'll see are pretty much discovery questions. So things that happen before the product discovery experience happened. So in this case here, you know, I'm a student entering law school. I am in a tight budget. Where's the right stuff for me. So these kind of question, typically, you know, sometimes we think, okay, what if I ask this question on my current e commerce solution today? Do I receive good results The reality is they do not happen on your site today. It's not just because your search engine is not able to answer them or anything like that, but it's really because it's not part of the flow of the current shopper experience in twenty twenty three. And might not be in the future, although, you know, we aim to change that. So what's happening today is when you ask these kind of questions, you end up, you know, on Google, and then you'll end up on, you know, a site like this one, like, I'm searching for what are the best, baby monitors, for example. And then The first thing that you see at the bottom is for each element of that list, you'll have a link to Amazon. And then, you know, if if you're not Amazon or Walmart or Target, you've just lost the sale. That that's pretty much what happened. So what we're seeing here is as you start generating that rich content, using GenAI again, or being held by GenAI as you start building the expertise of your brand. So for example, you're DIY store, you start building, you know, content around how to build a patio, how to fix, whatever. You might be able to to keep your user into your solution. And then avoid the bounce, avoid the re the redirect to Amazon, avoid that initial search, in Google. So mostly, you know, if you're able to answer these questions with rich content, with summarization of rich content, you'll be able to kind of educate your shopper to stay on your site and do you believe that you have that expertise. So if you already have rich content today. If you are, you know, a brand that already has created lots of blogs or articles or anything like that, you're already in a good position where Jenny could do, a summarization of this content and then showcase, you know, a, such a certain result showcase at least an explanation of what your your shopper need to buy. And then from there, you can bring them into the more typical product covering experience that includes, search, obviously, listings, recommendation, and and pretty much every other tools that you have in place. So I would like to say that, you know, GenAI in search, in in in the product discovery experience will be a game changer today. I think it's more a progressive element where GenAI, if you if you do the right thing, if you get already your content out there, and you start to build that that brand expertise, JanAI could be a game changer to make the intro on, you know, what what we so far, what we have seen in the last eight months or so, around, you know, the the the changes that could be powered by Genai. And and and I I agree with you, Simon. It says, like, you know, we can't expect customers to change their shopper pattern behavior. Like, if they're used to going through a certain thing, it may change over time, but right now, you know, everyone needs to meet the customer where they are. Just for for argument's sake, like, if someone has a very loyal customer base, very large catalog multicastrew retailer, two kind of questions here just to spin off the back of that. Like, do do they need to consider those kind of questions that you just showed, you know, for their own search experience. And two, are you aware of any information that we can provide to people. I'm not sure if it's really out there yet. Of cup consumers changing their search behavior based on, like, a bit more of a conversational chat GPS kind of search versus a, you know, we know that the the the number of one one word searches for e commerce is, like, takes up ninety percent of of total queries. And then two word searches make up the rest of the nine percent, and then there's a no point north of longer tail stuff. So What's your thoughts on that? You know, should you do it and will customers potentially change their their views or not? If you have, obviously, you know, the understanding long query and returning the right content, should, you know, it's it's a thing you should invest in. That being said, this is semantic search. I just want to be extremely careful that, you know, did we're used to that kind of gen gen AI chat GPU experience, but it really about semantic search and natural language query. It is obviously one element of semantic search. And to be honest, the, so far, what we're seeing is the investment into understanding shorter query, but understanding the intent behind these shorter query has actually driven a lot more value than understanding long queries. Again, because these long queries today do not happen. And and the reality is, even with the the arrival of Chad GPT, and even with you know, the surveys that say that, you know, you're looking for a more chat GPU experience, the number of keywords and queries, what we have seen so far, has actually reduced. So so it it went the other way around. So people expect a chat GPU experience, but we have to be careful. It doesn't necessarily mean that the expect to send long query, what they expect is to have a more human interaction with the entire experience, not just with the search, but with the overall navigation with the the people they talk with, with support, with their self-service after sales, so we have to be extremely careful that we don't just take chat GPT and then transpose it, and now it has to become the search because this is not the change of behavior we're seeing. And I don't think we'll see it, you know, early. Now that being said, if you have rich content today, if you have answers to these questions on, you know, what is the best on this or what should I start with as a beginner in that sport, If you have this content or you know where to find it and you can index it with a search engine, in this case, for example, like the Cavell search engine, then definitely look into these long queries, could be interesting. And in Cavell, we have started an early access for, self-service in a knowledge discovery, with NAI. Yeah. And I think that's in it. Like, that's what we need to remember. Right? Like, because we've got chat GPT esque, technologies, you know, generative AI is not a catchall to replace a search. You know, it's about accurate retrieval is one of the most important parts of it. And we all know that that can that can not be, one of the traits of generative AI can hallucinate. There are those issues where it can spit back the wrong thing. And that dip again, it depends on what kind of business you're in, but especially in something like b to b or where there's very specific queries where you need to respond with very specific products. That's an important part of that that person's buying journey. You know, gen AI might not be the the right option for that. So a lot of people here will probably be thinking about how can implement NAI within their business. Right? It's it's gonna be at some point, but, you know, within commerce as well. So for any of the the leaders here, you know, thinking about how they could implement it, what are some of the best practices they need to think about, you know, when it comes to top use case selection, vendor selection, you know, how should they approach it? Because I'd imagine there's a whole lot of people out there now with a lot of options, and it's probably even more confusing than ever before for our poor e commerce buyers. They were always in market having to look for stuff. So any help for them in terms of guidance. The the the the funny thing that happened, well, funny. I mean, stat and funny was that, people kind of forgot that they, they have entitlements. They have secured documents, especially in B2B, lesson B2C. But in B2C, you have stock. You have local inventory. You have, you know, if you are multi region, multi brand, you brand or region specific products and such. So so people kind of forgot they they looked at chat JPT and said, look, I was able to to, to do my entire trip to Italy, which at GPT, so I could probably do a full on shopping journey, and then you realize, well, know, I'm asking Chad GPT about this and I realized that these shoes don't exist or they're not available in the US or something like that. And, you know, people started to realize, and then it got even to a point where even for internal use cases and all that with with, security, for b to b, for entitlement, it ended up being completely disqualified because, it would not respect your catalog in any way, and will they lose the name which is obviously, extremely dangerous. So one of the thing that we have worked on, quite a lot on the on the Gavell front, for knowledge discovery first, really, for rich content, but we we do the same also for product discovery as grounding, or what we Well, grounding is a is a term that is quite known. So, you know, if we look here again, I have a, you know, a slide here. So Gartner, Mike Londis at Gartner, said, recently that he doesn't see, chat GPT or JNAI or LLM, yes, but let's say JNII Technology augmented search, he sees it the other way around. He sees search augmenting GenI use cases, and I I totally agree with him on this. And this is an example here as a a simple kind of architecture of how we do, grounding it at Cavell. So mostly we use the entitlement, the security that are present in the index, that have been retrieved by our connectors. And we use this to ground the vector database that is then used, to to generate the answers. So mostly We will respect your security. We'll respect your entitlement. We'll respect your local store and such, and make sure that you know, the content that is generated will no will not go beyond, the different bounds. So if you, you know, as a buyer, if you are being promised by, you know, a a new vendor that says, you know, we've created a layer of application that looks really good on top of of a a Jenny vendor such as OpenAI, JetGP, and such, you know, as them Okay. But what about the grounding? What what about my content? Are you able to respect my entitlement? To respect my inventory? I think that's a very valid question. Because most of the time, the answer is no. Unfortunately, it's something that is quite hard to do, and and requires also a basis a base index requires, a a container that already has that information in ends, which is not necessarily the case for. For all vendor out there. So I would say, you know, that's that's the first thing I would say. Don't don't forget. It's just, you know, with with that entire height that was created around NAI, I think people tend to forget that they have business requirement and limitations that need to be taken, into account. Yeah. For sure. You know, you don't want it suggesting a product that isn't available on on the site or there's nothing to do with the content that's already that's already, there. So you complete, completely agree. Any other use cases or anything you, like, you know, misconception wise, you know, we should flag to everyone so that they're aware when they're going through a process of looking for someone for for gen AI? The cost as well. And the cost and latency, so we have, we have ran a few tests on very large catalog, and just not just large catalog, but also on large volume. And here I'm talking about, you know, volume that goes around a billion queries a month or something like that. And a lot even of the providers, the the big players out there, such as Open AI, Microsoft, the service that they offer, for example, Microsoft from Azure, we actually got to the limit of of what they can provide. So even, you know, cost aside, they were not able to scale to the demand that, for example, we had for certain customer that were in the retail space. So thinking that, you know, every single question, every single query that will be asked by your shopper will be answering a generative way, is a very inefficient way to do things. And if you have any sort of you know, echo concern, you know, environmental concerns about it. It is requiring quite a lot of energy to to power this, but you know, that's a different topic altogether. But even cost wise as well, a a query sent to a JNI, system is usually much more expensive. Then a query that used normal semantics. Again, because it needs to do a lot of processing, lots of generation around it. So these are concerns also that are pretty valid. So don't necessarily think that you'll be able to take something that works in the small demo environment, answers some rich question and then all of a sudden, put it in front of all of your shoppers. You're gonna have to be smart. Again, what I kind of said in your first question, you know, answering questions that are not necessarily typical in a product discovery that happens somewhat before usually happen with Google. This is really where you should focus. So longer queries queries that happen in a certain situation, you know, you have to be smart, when you implement it. And, again, you know, there will be promises of of, of the klondike, you know, you just put this thing on your site and start answering all questions, I have to be very careful with this. Yeah. And, you know, I think that's why Gartner say what they say around the cost. Right? The cost the cost was to pass the value when things are looked into in more detail in a year's time. Possible all of this, but what's the ROI, you know, what incremental, you know, KPI have we generated whatever you're interested in? From those experiences, and that's where it becomes really important. And I think that's why where you consider and where you side to use JNAI needs to be really, really thought through. And I know that's not the most exciting message in the world because it isn't. It's you know, it's fun to be part of the hype and go after everything with everyone else. Right? But at the end of the day, and as a vendor as well, we know that things will be evaluated in a in a year time, two years time, we wanna make sure that the the values there. So kind of close off in a little bit of this, and then we'll kind of go into into some of the the the kind of more what can we do to generate more more revenue and that kind of thing? Is there a good example that you have of a tactic or functionality that often gets classed as gen AI when stop it with a completely different technology? Absolutely. And there's been a lot of obviously, for SEO reason, the Jenny IChat GPT has been, pretty much sprinkled on every blogs that have been released recently. And it it is extremely dangerous. One that I've seen quite often is, detagging or product tagging. So could you tag products that have, you know, missing missing categories or, you know, bad. You you have a marketplace you receive from a lot of providers. You're trying to you know, augment the data using tagging, and then you think, you know, GPU could do that. It could. But it's absolutely inefficient. First, it will be expensive. Second, it will hallucinate like crazy. Again, I said at the start that creating rich content, creating description is something that ChadGP does well supervised. You know, when you have a a a bunch of people, you have, your your category manager, you have your merchandisers, you have your content creators that would look at it, review it, make sure it makes sense, make sure the products that they talk about or products that are really in your catalog. There's a lot of things to consider there. If you use it for things that are very specific, in this case, for example, categories in zero shot learning, which is learning that is not supervised, it's gonna go all over the place. You'll have hallucination all over the place. So you have to be extremely careful with that. And to be honest, deep tagging is something now that is quite cheap, that a lot of smaller provider are offering. You have Amazon Amazon recognition. Does it Azure, as an ML service for it. And it will simply infer with the rest of your catalog. So it will look at, you know, the these products are somewhat similar with their image. For example, what category they're in, and it's gonna start proposing tagging categories and such at a very competitive price. So I've heard a lot of tagging being put under the umbrella of GPT, be very careful on this. It is an extremely inefficient method. I've also heard about one to one personalization. Right? Because before that GPT craze, if we go if we look back in twenty twenty two, it was all about one to one personalization. And to be fair, it has shown a lot more value today, like, real proven AB tested value Dan, anything that has come out of Gen AI when it comes to product discovery. So one to one personalization is still something great. At the moment, most NII model do not keep a state. They were not in the same bill for that. So you need to combine them with any sorts of vector search. So user session vector compared to vector, a vector map. So, you know, the semantic search using intent detection, and vector based search are really the core of one to one personalization, not generative AI technologies. So you have to be careful with that as well. You'll have rich answers, but now not necessarily personalized, and they will not personalized in session. So so that's another thing that I I hear a lot, and you have to be careful with that. I can even see there with the deep tagging stuff how you could quite easily assume and, like, misconclude that that was generative AI, right? Cause like, it's generating a tag. You know, it feels like it that's going it's all down that road of everything that GenAI says it is. But yeah, like you say, again, the cost is a really important part that we need to not get carried away with the excitement of it all. It's significantly more expensive. It's not like it's a little bit more expensive is, you know, in the multitudes of ten times more expensive, that kind of thing. So, you know, it's not a it's not a joke that it's, it is a bit more. So I can see how people can can can make that that mistake. When it comes to kind of holding this original state, can you can you just explain that a little bit more? Like, So, you know, why I can't GenAI do that? You know, why why is that not an not an available feature? Because it kind of feels like to me when I when I've used chat GPT, there's a chat there and there's some sort of state because we know we we've had a conversation. Is it when you then apply it outside of that kind of scenario? The the before and after. So so what what the GPU has started to do is to keep a state within the conversation. So you can ask, you know, oh, I'm looking for this kind product, and then do you add it in black or something like that. So that conversation is kept, but knowing your taste, where you come from, where you're going, this is really where it lost it. So being able to do also real time personalization, keep in mind also that that generated content is a little bit slower. It will not necessarily do it, because it doesn't know the rest of the catalog doesn't necessarily understand it. The way that we do actually have a slide for it. I think it's gonna be easier if I if I showcase it. So, you know, when you use product vectors, what happened is pretty much you take your catalog and you create a vector map of it, what you see here on the left side. So mostly you kind of redistribute the different the different product into a physical or or, well, a vector position. And then from there, you have a user the moment they start interacting with your catalog, you just start to position them you know, into that vector space. You can inform your generative, model of where you are now in space, you're gonna be able to tell them, you know, they are interested into that kind of category or brand of product. And then it could leverage this. But it needs to happen before. You need to have the technology that will track it before. And then once, you know, that conversation is done, and then you move on to just normal discovery where you start looking at products and have recommendations and such, what has happened in the chat also need to be reflected into that vector space. And it's not by default into the way that you build edge and of AI model, that that's that's not, how it was that's not what it was built for, and it's usually not constrained within the catalog as well. So it's it's idea that there is a finite system that is a catalog is not necessarily there, when it is built so we're getting a little bit, you know, deeper into the science here, but this is mostly what's happening. It's just it has no necessarily understanding of what is your catalog, what is it start in its end, which is what a vector map will do. So then, you know, I think the thought on everyone's mind will be what about people using chat bots, on their e commerce stores powered by gen AI to help with the sale. Now what's your thoughts on that? Cause it's been a quite a bit of movement around that, I would say. You think. The there's always that kind of blade runner dilemma of when you, you know, when you see something futuristic, you hope it's always a role Right? That we we always we always want to see robot. That's what we wanna see the future. And and when we got the chat GPT experience, you know, the the generative AI technology, the LLM on the need have been going on for years before that. It's, you know, it didn't come out of nowhere. But the fact that Open AI put it into a, you know, human human like interaction format where you could chat really made it kind of alive, obviously, and and that's where we started to see the robot. That being said, just taking it and putting it into a chatbot, is a very limiting, way to to take, you know, GenAI. And again, in the chatbot technology, if there is no search behind it, if there's no semantic, semantic search powering that chatbot technology in the first place, you're gonna have that same grounding problem that I mentioned before. So, you know, that butt can start to to hallucinate. There's also the idea that UX wise yes, everyone, again, want that chat GPK experience. They want to have a more natural flow, some sort of a a trusted interaction with your solution. It does not necessarily mean that they want to check. And and people have had a hard time to differentiate it to it's not because the system understands me, understands where I am, where I'm going, what I mean, that I necessarily want to have a conversation. And and we need to really separate these two. Because you'll end up, you know, at the moment, if you go on Amazon dot com, unless it changed, you know, in the last twenty four hours, there's still no Jenny, chatbot available. Even though they have the biggest catalog in the world, and probably, you know, benefit from having some sort of, direction. And the reason for this is simply because they know that their user right now, their flow is not to start a conversation. So so maybe Amazon will do something. I mean, they have all of the people, all of the money, all of the technology to make it happen It's not a question of time, is really a question of priority and they don't feel like at the moment, they can take this technology and put it into a chat format, and it will bring in value. So so I would be very careful. I think, you know, the the hype around the chatbot, chatbot providers saw this as, you know, their new way to kind of position their product, and and it's great if it works for them. But so far, you know, for pretty much all of the prospects, customers, partners that I've talked to, that I've interacted with, I've seen very little value out of are not meant to chat, but using, using JNAI, at the moment. I think if you're if you're gonna try it, know, do what I hope everyone would do anyway, which is a, b, test it and make sure you run significance and see what value it's generating for for the business because it won't be cheap. And if it's cheap now, it might be, you know, could if if it's based on usage, you know, that could go up quite considerably. But I think your point on Amazon is is key as well. Right? They know their search work pretty well. They know how people search on there. They just wanna retrieve product information. They don't need to start a conversation for that. I can see some situations where you might wanna start a conversation, like, don't know what gift to buy my partner for, you know, I don't know, a fifth year wedding anniversary, something like that. Then there may be something around there, but Again, I think those questions usually start somewhere else and that's part of that flow. So we we need to be we need to be cognizant of that. So just kind of moving forward into into the, like, areas where, you know, everyone here is is hoping there's something GenAI can do to step up their KPIs, be it conversion, average order value, reduce abandoned carts, you know, whatever it is. What what is it that you think is available within the tech if utilized in the right way that could start helping boost everyone's KPIs? Yep. So if we look at the, you know, the original use case that I saw, where it's, you know, to answer complex questions, and making sure that these complex questions are tackled on your solution instead of being addressed on on, on Google, it could increase, obviously, your conversion because you have more traffic that stays. You have less bounce rate etcetera, etcetera. Again, here, it's a, it's a more complex use case because it's not just putting GenAI. It's also the creation of rich content, gaining trust from your user, even a bit of marketing, you know, did you know that we have expertise and and all of that? So it's more of a process than a technology, to to increase or to reduce that bounce rate, increase traffic, increase conversion. On the AOV front, if you start explaining a bit, you know, why you should choose a product versus the other, why this product is better quality, and it's more expensive. You could increase your AOV, but again, maybe that more expensive, more, product also has, you know, smaller margin. So overall on your bottom line, that might not change much. So you have to be very careful with that. Also, you know, with JEDI, you you're being more transparent eventually to your users because you explain why this product is being shown, why it's recommended, So you might tell them, you know, at the end of the day, you might not need that very expensive product, and you should probably take this one instead. What I see though is, after the sales the sale is done, in self-service reducing possibility return rate. Because with the pandemic, we saw, you know, an an another pandemic, another epidemic was the return, just being, just increasing drastically and and, a lot of retailers trying find ways to reduce the returns. If you have, especially for more complex products, for clothing little bit harder. I guess AR could help you more than than GenAI. But if you have, for example, electronics, laptops, all these things, explaining a little bit, you know, or or trying to help you use our self serve, could reduce, you know, your support cost, your return cost, even before, you know, if someone is looking, I know myself, for example, I bought, for example, a TV, and I'm looking for a sound system to go with it. If I buy the wrong thing, I'm gonna have to return it. So trying to explain a bit what are the complimentary product, what is compatible with what could it help with, with the reduction of return as well. So that's where I see it. So again, it's not as, we saw, you know, that one to one personalization overall AD tested has a pretty significant increase in conversion AOV and overall our PV, that's what we see more and more, especially if you if you look at the long tail, you can avoid the popularity loop and then part even to showcase product that have higher margins. So, with these technology, we were pretty close to the money. We were pretty close to, you know, that RPD, that margin. With with GENI, it's still not there yet. It's still things that, for example, like, where we see, real money is is more on the efficiency side. So increasing the efficiency of your merchandisers, again, through product inspection. But on on really the product discovery GDA will increase conversion rate increase and all. It's still not there. There there's still a lot, of of UX to be done even around the application of GenAI. But we see it a lot though on the self-service front, especially here at Kavail, where started, you know, that beta, and we see some great results coming out of it. So so that I can tell you. I I can guarantee that it can help for your self-service, reduce return, and such. Yeah. And I think that's still, you know, still very valid for people on on the webinar because a lot of the self-service journey starts on the, you know, dot com or equivalent website. Right. So, if you can help with a bit of case deflection and making sure that your your call centers aren't overwhelmed if you have them or your your chat support isn't overwhelmed. There's a huge amount of cost saving to be to be had there. It directly impacts some people's KPIs, but it depends on how the business has been, structured. Sometimes it does impact commerce teams, and sometimes it doesn't just on being cost conscious, you know, and especially when it comes to implementing new technologies, because again, need to consider that It's not only the the technology license, it's the process of buying it, it's the process of implementing it, and then ongoing costs of the total cost of ownership. You know, what are the key areas that Genai sort of demonstrates good cost effectiveness, and some sort of like ROI for for commerce businesses versus being this thing that we keep saying is extremely expensive and needs to be careful. If if you choose a query, again, the the number of query that receive that says, for example, what is the best product for me? How do I how do I build this? How do I do this? It's still a fairly small number of overall queries, overall request that you receive, you know, in a product discovery journey. So if you just focus on those, the ROI could be great, because again, you're gaining trust even even if it does not necessarily translate to conversion today, it could improve the trust, improve the brand Pretty much your your brand, the brand expertise or or the perceived, brand expertise out there. And he could overall, increase, Coveo time increase the the amount of traffic and such. So that's where I would say, you know, the best ROI lies. It's really, you know, if you focus just on these questions and you make sure that you have rich content that can support it, that can be summarized, properly. I would say in this case, if you use it in that situation, the the sheer volume that you'll receive is usually quite acceptable. And on the other end, the the expertise that you'll showcase can be very beneficial, even for the conversion AOV overall down the road. Yeah. That makes sense. So, just to kind of round off a little bit here and and finalize yeah, we we could go we said that we were gonna go into some more of the potential risks and challenges when adopting the technology, but I think we might covered that, you know, during during this. I think it made that clear. Maybe in terms of a bit more of a summary, you know, first, let's start with the positives. Let's just summarize again, like, what's the top things? You know, you've even said product descriptions. There's elements of, you know, routing the right queries the gen AI might make that makes the most sense, but not all queries. But can we just go through, like, the top things that you think are positive? And then anyone considering the technology at the moment and probably in market, what things should they consider and be just be really aware of, make sure you test, make sure you validate, it'd be good to summarize there and then we can go into the q and a. It depends. If you're fighting today versus if you are a retailer that wants to be the everything store and you're fighting Amazon today, I would say focus on your self-service portal. So instead of investing a lot in product discovery, focus on knowledge discovery because you want to reduce your support costs. Now if you're ninety percent of, you know, the rest who are not necessarily directly competing with Amazon at least retail to retail. You just want to be a brand that is focused on a certain type of product, and you want to showcase that expertise, then I would say, the the the the the main use case is to generate content, generate expertise, and summarizing that expertise, so that you gain brand trust And you make sure that even if you advertise on on Amazon, even if you have products on Amazon, people will start, you know, they will buy your first product And then it will say, well, you know, that product was good quality. Why don't I go on the main site, on the mother's site? And then they get there and they realize, oh, the experience is just much more than what I have on Amazon. There's a lot of expertise here. I'm I'm being told, you know, what are the related product that could fit with what I've purchased before. Self-service is great. So I would say, you know, differentiate this way by using GenAI. So mostly focusing on creating rich content and summarizing rich content by doing knowledge discovery. And join us in the beta if you want to test it, which we have ongoing at the moment. And then for for, like, these pitfalls and just to be careful, like, if you're in market right now and you, you know, you're looking for stuff, you know, I I'm gonna put out there that make sure you test it. It's gonna be my number one thing. If it's something that you know, AB test it and be really, really and scrutinize it and make sure that the cost is gonna be worth it. Costs should go down in the long run, but there is a lot of compute power when it comes to GenAI. So it's gonna it's gonna take a while, I think, for that to to drop, but it will inevitably does. But that would be my my key one. Simon, I don't know for you. Is it just is it just make sure that you pull the mask off and check that it's actually Jenai behind and not may well, I mean, you know, promises there, like, oh, we're gonna fix your, we're gonna fix your data issue, with NII red flag, you know, there are lots of other better way to fix that. There are a lot of provider out there that does keep tagging and much cheaper much more efficient. So I would say, you know, go go that route. But then the rest of it is, you know, it will not replace your search, not today at least, you know, in the future, maybe. But it will not replace your search. So if you buy If you want to use, you know, a JNI provider, make sure it has a search engine underneath a vector database that can ground it, that can personalize it that's that's the most important thing. I would say, you know, if if you come up with a new provider that come in and say, I'm I'm putting just a layer on top of JetGPT. You'll see it's great. Most likely will not, be, so just be careful with that. And and really, again, creation of of rich content. So if you use, you know, a generic provider that has, an index underneath, make sure they can index rich content. So we see a lot, you know, of, of of e commerce search provider, that only index product data. So in this case, how can they ground, your risk content, they will not be able to. So just make sure you you have that as well. And that would be and and do not, do not think that the after sales self-service, you know, does not does not gain anything from GenAI because it does. We've seen a lot of of great things coming out of it. So also, you know, look at a provider can do that part of the business. It's quite important. Can we just check something, like, just a validate? Because I'd I'd imagine a lot of people here have content that they wanna index. What do you mean by index content properly? Because I think I think that might be something where people saying you can do it and what what does that mean versus actually being able to do it? Like, I'm gonna ask on behalf of people, like, do you mean that if for instance there's a downloadable PDF It's not just about, like, indexing the name and file type. It's actually the content within the PDF. You know, do you mean to that level or What do you mean to that from somewhere? Absolutely. So the rich the rich card, then that's a really good example of that. Being able to index also, you know, in seeing the website, like, just web scraping the website. Yes, it will work, but you'll end up with just a blob of content, making sure that the index automatically will, properly put these elements into fields. So you'll have your description. They'll understand, you know, the method tag on the page, even just for web. An indexing content that is not available even on the web. For example, you know, you have a a legacy database that is on prem, do you have a provider that is able to index that, are you able to index content that comes from the PIM, for example? Are you able to index partial content as well? So you know, you have some marketing content in your CMS, but most of the rest of it is in the database somewhere on the file system. So these are also connectivity security support in these connectors, being able to properly extract, metadata at at indexing time. Also being able to do advanced processing. We have a thing at Kavell called the indexing pipeline extension where you're able to correct data as you index it or even fetch additional data, for example, that's is in your CSS, on your page or something like this. So all of these elements shows mature rich indexing product. And there's not a lot of of provider out there that does it properly. Yeah. And I think that's a good way to round off because if you if we, you know, If the view is here that generally it's about content and and knowledge for customers, and that is part a huge part of the buying process it can make a bigger difference in some industries than others. But that's an important part of it. You need to make sure that what it generating its answers from is your content, and that means your content everywhere, not just like what's easily accessible. You know, on a on a on a web page. It's everywhere because then you've got your own mini version of chat g p t, then that's actually utilizing everything you have and it's trained on that and it's grounded on that so it can't go outside of it. And I think that's that's that's a really important part of this being successful if you're kind of if your business does have that kind of content focus. Great. Well, thank you very much, Simon. That was super interesting. On the webinar, if you have any questions, feel free to start putting them in. Courtney, I think you're gonna go through any that have come through so far, and we'll tackle those first. Yes. Absolutely. So reminder, please use the chat if anyone has any questions. I will read them aloud to our presenters. But in the meantime, I'd like to kick it off. Is it necessary to make any specific hires to explore the AI route and would it be recommended? Yes. Is it recommended yes? Does it need to be higher, in in AI? Absolutely not. I think, you know, scientists and researchers are expensive. So and and could be better used somewhere else. To be, you know, if if I go back to that initial idea that, you know, you want to showcase your expertise, you want to bring that experience that is generative, not just putting, you know, generative in the chat bot or something like that. It requires UX. It requires web development. It's crazy still to this day how sometimes we try to aid the test to AI technologies, generative or not. And and our our customers or prospects or our partners are unable to do so. So make sure, you know, if you don't have these resources, you you partner, with someone that can do it. So web development is is more critical than ever. Because you're gonna want to AB test it all of it. And and you're gonna want to AB test not only, an engine versus the other, but also an experience versus the other. You know, we mentioned that chat. But before, could you show, you know, a a situation where you have, for example, the search on the left side, but then you have the chat on the right side that looks what you put in the search or something like this. It's all things that is it possible with AI, but how it's gonna look like, desktop, and then mobile, is really where the challenge is. And and it's not your AI provider that will help you with that. So yes, hiring, making sure that your web dev team, and your mobile mobile development team are staffed properly so that they're able to adopt these technologies and properly test them, and don't spend too much on on AI resources. I've I've seen a lot of companies building their AI team. I don't think people realize that Amazon has thousands of AI resources, you will never compete with them, on your own. You know, you you that's why we're here. That's why, you know, they are third party vendor that focuses only on AI. Thank you so much. As anyone disagrees with anything we've said, I think is interesting to put that in the chat as well because, you know, you it could be a valid point we've missed somewhere and, you know, we happily discuss it at the same time. I do have one other question right now. What is the ideal profile of a company that is positioned to leverage AI? While minimizing well, minimizing associated risks. Yep. So I would say company that starts today, I mean, I'm bit of a broken record here, but I want I want you people to understand that, you know, shopper educations at the core of it all. So if you already have content, if you already have users that trust your expertise, for example, you're you you are in a niche somewhere, or you are known as the leader in that stage, for example, you're you're a leading outdoor, retailer or something like that. You are already starting with somewhat of an event And you you're mind minimizing risk because you already have that content that sits there that is just waiting to be consumed, so you can test you know, NII on it, see how it's how it's returned. And even if at first, you hallucinate a little bit on that content, it's not the end of the world. You know, you will not end up, you know, showing a product that don't exist or a price that is wrong or something like that. You might just say something that is might not be completely through it. So so I would say, you know, this is the the best way to test to to bring value and to minimize risk. Is if you already have that rich content out there that you can use to educate your shopper and to bring them or or, you know, try to fetch them away from from Google. Thank you. If there are any last questions from our audience, please use the chat but right now I'm not seeing anything else. So I just want to say thank you to both Sergio and Simon for the presentation today, kinda going through these misconceptions and GenAI for e commerce. Is there any other final remarks from our speakers that we wanna say? No. I think we'll thank you for for listening. I appreciate it. Up there. It's a wild world, but, you know, we're here to help you. Wonderful. Thank you both. And I hope everyone a good day. We'll send out the recording to everyone who's watched it. Thank you.