Hi everyone. My name is Mathieu Lavoie Sabourin and welcome to the Masterclass R360 Commerce. I'm a product manager at Coveo and I'm here with my great friend Vincent Bernard. My name is Vincent Bernard. I'm director of Applied AI Solution and we're very happy to show you latest and greatest in terms of Agentic and Commerce. So we're going talk today a lot about B2B, B2B manufacturing, right? B2B buyers are not like your casual buyers. They're not just browsing. They're looking for complex queues, They have, you know, complex leads, and they're in a rush. Right? They're not looking to spend some time and shop for surfboards. They're looking for, like, an important piece of hardware for something that might be broken that need to fix and actually get out there right now. The not the funny part, but the difference part the different part between what we've done before and this, because you may have seen our classical I call it classical, but the brand new B2C Agentic solution. It was our introduction. It was also the easiest path for us to go live. Now we started to investigate how can we make Agentic solution works with B2B. And like you said, B2B has tremendous complexity. Usually that complexity forces them to fall back on legacy technologies or things that are not as modern. Do you think with what we're coming with we can bridge that gap? Absolutely, but it required us to make the little rectangle box, the intent box, the search box a lot smarter. To be able to bridge and understand different intents, different complexities of queries, different obviously skew, complex skew queries, but also sometimes troubleshooting queries. Right? You might be looking for a new piece, a new robot arm or a new tire or something like that, but you'll also actually be trying to repair something that's broken. As such we need to have the intelligence to be able to help those purchase buyers or B2B buyers navigate through all that complexity, navigate through large catalogs and get to the right outcome. Yeah, the catalog is larger. It's usually also it has views so if I am a distributor tier one, I'm gonna have a different view on your business than if I'm a a lower tier. Also, for some of the B2B websites, the one we'll show you today is a new vertical for our favorite Barca company. So Barca is a fictional company. They now are in the robotics industry. So you'll realize that these products are more expensive and they also are more sophisticated. So they're to require education. They're to require support. They're to require maintenance. They will also live in an environment. So they will require fitment, for instance Absolutely. Which is the our favorite word these days as this is where we get a lot of challenges and a lot of, cool solutions. You could also get a lot of variance from different pieces depending on who you are, depending on the type of, company you work for. Yep. So And and the more you dig in B2B, the more surprises you get in that box. You get, like, different markets. You get different prices. You get different there's a lot to unpack. And a lot needs to be assembled together, but then she make, a complex robotic arm work. Yep. So we'll jump in the demo, but, bear with us. We'll show you the the generative experience, the conversational experience. But then also after, we'll make sure for you to understand what's under the hood and what do you need to have in your Coveo deployment to benefit from all these different new features. Fantastic. Let's begin. Yes. So let's get get it started off here on Barca Robotics, a brand new vertical for this company. It's really impressive what these guys are becoming. Definitely. We're opening up here the search box, and you'll see that you're going to have a lot. And the reason why is to kind of showcase to the user quickly where they can go. This is a unified website. So in a unified website, you're to have forums, you're going to have reviews, you're going to have some manuals, large PDFs Documentation. Documentation, and products. Here I am interested into learning more about my robot, which is the AC one one one space zero zero six. You can see it here in my recent searches. This is But no one remembers the actual SKU numbers, right? No one does. And even if you do, if you work let's say in DIY, let's say you're at Canac, most of the employees will search by SKU, by code actually, not even SKU, internal code. And those are numbers. And and at one point you may want to go in a family of product for instance. So we'll dig in that a little bit later. So if you search for something like a c eleven, you're going to find the entire list of products compatible or that belongs to this family. At least that's how they made their queue structure. So you see here that we managed to find a c one hundred eleven zero nine, d zero one zero seven, etcetera. So very good, like, search overall. You can see a list price. You can see in how many warehouses it is available. You can see it's in stock. And this is all because I'm anonymous. But then if I connect myself, it's just going to morph and be adapted to what I want. Just going to narrow the scope. And you see here, we're actually showing this traditional search experience. Right? And search is not completely gonna go away, especially for the, like, those vague queries. But what happens very often then is the user will start to dig. The user will look into something a little more complex, a little more advanced, and this is really where the agent agentic layer can come in. We can help the users navigate both a more standard open search or a more guided conversational experience. Yeah. Let's like, this is fine. I got my robot already, so I'll just ask quickly a question here. Compare the strongest options or this one even. Which result is the safest replacement choice for a c eleven when downtime matter most? This is where the complexity comes in. Yeah. Exactly. So here you'll see and this is still a prototype we're building. A lot of it is already available. Everything that is the commerce search that you've seen maybe in the previous master class. Here, we decided really to start merging, I'd say, the knowledge experience with the commerce experience, which is the big news we're announcing today, basically. We started to merge all these products we have together. And when you do these kinds of queries, you'll see that the first off, the agentic framework will take its a little bit more time, a little bit more thinking because it has way more depth to go in. But at this point, it starts to showcase. And what do we see here? How do you do you call or how do you explain what's happening here to me? Yeah. Well here we have, you know, a couple of pieces that can actually answer this question, right? So we have just a complete question, but the complex question doesn't necessarily require just looking for one piece. We want look at a few different options, right? So the agent is there to help guide that shopping experience, comparisons, could provide a list here or in this case a carousel, for example. Carousel is optimized right now for for for actually not trying to go to to to have not too many height on screen. Otherwise, it's just it kills all your real estate, so we're just using a vertical approach here. But then this is what we call a layout, a layout for displaying products. But then we also have at the bottom some interesting follow ups. You see, they want you to continue without retyping, obviously, in a one last Of Simplify the user experience. At this point, we'll find an installation guide, for instance, and that's just to start, like, really merging it. You see here that it kept my whole my whole context. I started the session initially with AC eleven. Now I'm hyper focusing on one of them. You see also kind of a feedback loop here. So what's happening? What's the kind of action this guy can take? So the agent here is doing a few things. Right? It's obviously gonna search for products, but furthermore, it needs to understand the intent. Because as you mentioned, we're working here with layout. So the agent is not just finding products. It's also determining what's the best, most optimal way to display what the user's looking for dynamically. So we're like we're getting into essentially a narrow or a paradigm where the agent is not just doing a search for a human. The agent is actually deciding and helping the human and displaying the results in a way that actually is more convenient to guide the user down the purchasing and journey. The installation guide I got here is pretty cool. I got a little tile for product, then I got an installation summary. You get the required tools, which is always fun. Pre installation checklist, the warning and safeties, and then the best part, in my opinion, is really a step by step procedure. Yeah. And look, the the warnings if you have some. Here, if you're properly aligned, it's going to cause vibration and bearing wear, which is terrible. You don't want that. So the step by step is very interesting. I think the only way you can do something like this reliably is through good content. Right? Absolutely. The content obviously is the key under, and perhaps maybe that would be a good option to dive into that. And we'll get back to the layouts further on because we have a lot of different layouts to show. But some of you might be wondering, how do you make sure, like, all the content I have, because here it's like mixing different types of content. You have products, you have technical documents, you have guides. How do you bring all these things together for that agent and then display it into layouts? So we'll come back to layouts, we'll come back to that fantastic technology, but let's dig maybe into what happens under the hood. Where does this all come from and how does Coveo make this available for the agent to search within? The one of the motto we have internally is keep your data where it is. We are not in the in the business of like cloud storage. So no matter if your data is in Salesforce commerce or in Salesforce, like your your KB's article and support pages or in SAP or in name it, we don't That's your database. Yeah. We'll take it in. We'll take it in. So Coveo is known as the Switzerland of the data. We will not take part of any parties. We just like to play with everyone. We're agnostic. We're very We're agnostic. At this point, here for this specific implementation, you'll see the kind of data we gathered. Most of them are through push sources. The reason is that Barca Robotics has a very custom ERP in house so all these different parts are referenced somewhere in the system. These customers decided to create some little functions to push it to Coveo. So it goes from their system to our street. You see this company managed to index some blogs, client accounts, documents, KBs, product, redirect rules, name it. Some of them are through the catalog sources, which is always interesting, but then the rest of it, like support cases, work orders, name it, there is a lot. Do we need all of that to create something good and Coveo conversational? Well, The more you have, obviously, the better the experience can be because we have the technology to be able to bridge and bring all these things together. But, of course, we have customers that start with less. We have customers that have way more. We have customers that have dozens and dozens of sources. The beauty of our index is that not only can inject everything, but it can also actually create the mapping between documents, availabilities, and products in order to deliver a fantastic experience when things actually make sense and we can give the right documentation for the right product. I told you earlier about the snowblower that I bought last winter, and the snowblower actually came with three documents. So three large documents that actually explain how the snowblower work. And one of the documents was actually referring to three different SKUs. So those links are very important in the context of an Agentic search to allow the user to find the right SKU, find the right documentation for that SKU and any other resources that might be helpful in that context. Yeah. You know, if you start selling multiple products in multiple markets and multiple languages with versioning, it just creates kind of a very hard to manage thing, this is where we shine. We'll show you a little bit in detail some of them as I think it's interesting for you guys to understand the level at which these agentic solution will be will go. These agents are as good as the data, so that's why we're spending an extra time on this. If we go on the content browser, for instance, and we open something called an availability, what we call availability can be a store, a warehouse, or even a distribution, list or a customer list in this case. Let's go in the Midwest. If you open these guys up, you'll realize here that you're gonna find the available items for this specific item. So you see all these queues that we were referencing to earlier, they're all there And this is it can be live if you want, meaning that we have an atomic system that can really index small pieces of information. We call them partial updates. Meaning that you can send a single SKU here saying, I don't have it anymore, And then we'll remove it, or you can say, I got this one back at this distribution center. So if you have that kind of enterprise bus, let's say you're using Kafka or a messaging system like that, every time there is a change in your inventory, you can route it to us, and then we can adapt in real time. And since this feeds the search and your agent is based on search, it trickles back, which is one of the beauties of this What happens when a manufacturer or distributor potentially has different prices for the same product, or different attributes that must be deflated to different audiences depending on who they are. Is is that something we handle as well, Vince? We do, but it was a it was such a headache initially when we got that problem. Most of the manufacturers using search right now will have the brute approach, a brute force approach of just flattening it down. So you're to have an item in the search, which is your snowblower, with your price as a customer. So if you have ten customers, you're to have ten items for a single item. But then this list goes from a hundred thousand customers to even more, so it creates kind of an explosion in terms of combination. And we decided to resolve it by nesting these prices directly in the product. So let's just find a product over here, Push product, this guy. So if we take this access force, you'll see here that recently we unlocked what we call a price dictionary. And you see it's a dictionary, and the reason why is that if you open it up, you're going to find like the name. This can be the client identifier, or in this case, we're using A region. A region. Or maybe not, Pacific Robotics. This looks like it's a name of a customer. Usually, you use a number, but you can use it doesn't matter. And then for each one of these customers, you'll see a price in there. So this is how we resolve it. When you're querying your front end, you're logged in as one of them. So you're only going to receive in the EC price dictionary field, you're only going to receive your value. You won't see the others. As in your authentication token, we enforce this kind of context of the dictionary itself. In essence, we're really built from end to end, you know, starting with the root in the platform all the way to controls and all the way to delivering experiences to serve those complex B2B use cases. One of the elements I feel is perhaps that we haven't touched on so much is we have, you know, a fantastic index, we've shown some layouts already. Now how do we allow merchandisers, we have people who are managing those ecommerce or customers, you know, that we that we love and appreciate, to managing those experiences and making the relevance better, controlling who gets to see what, controlling product's gonna be displayed for which region or for different type of customers, understanding the context, and then even getting feedback, understanding what's actually working, how much are we selling, what's my average order value, what's my conversion rate. How do we help merchandisers achieve that? That's a good question. And the more you go in the agentic world, the the less control we get. That's my feeling. Hence, the I I think why people are afraid to jump in, in B2B. These agentic solution promises a lot, You know? Oh, yeah. You can fix everything with your layouts, but but, like, is it true? That was the need for human in the loop nonetheless. Is it true? And and and the reason is yes. The goal here is to really take the best of machine learning but also combine it with the traditional, I'd say, approach. So this is the Merchandising Hub. This is where you go to operate your search. This is basically specialized into optimizing ranking rules, boosting promotions, exclusions, recommendations and a lot of different features. This one is pretty cool as it's data driven so the entire interface will give you insightful things regarding revenue per visit, average order value, conversion rate, or even like the spread of these different metrics based on where they are coming from. You see here we launched the listing pages, so we started to get revenue out of them. Real insights to help you understand how your business is going with Coveo. The, interesting part in my opinion here is the copilot. So we're here to talk about Agentic. Tell me more. What's that guy at the top here? Well, you know, the Merchandising Hub offers, like, a a plethora of very powerful tools to help merchandisers manage their B2B store. Right? But, of course, we're helping manage an agent. We thought, well, why don't we also actually provide an agent that can help the merchandiser manage an agent? So really getting into, an agent to agent world where we're gonna provide additional tools with AI agents that understand the APIs, that understand all the configurations that exist in here. And if you're new to this, for example, you're not familiar with the Merchant Hub, which we've already built to be very simple and easy to use for merchandisers, but if you wanna go even faster, we're providing a co pilot directly within the Merchant Hub that can allow you to gain insights but also even apply rules that will then further improve how your agent and your search performs moving forward. One of the good example here we can do is give me insights regarding my storefront overall performance. Those are suggested. You can type in whatever you want. You can help you can get help for creating a rule, deleting one, or managing overall your experience. Or even getting deeper insights into specific queries and specific categories for example. What's usually hard with data sets like this is that you're going to have a per query kind of strategy here but then what's the patterns? So this is where we're to really be able to do it. See the interesting, we got a search drop off at the end. That's I mean, the Barca business problem at this point. Strong listing monetization despite low volume, so that's interesting. Put your eggs in the basket where it pays off. So here you see that you don't have a lot of volume, but the revenue stream is interesting, so good place to to merchandise. And then the recommendation engine is delivering premium transaction. Good job on that. Interesting. Yeah. So we're helping recommend actually big orders in our recommendations. This is very sweet. Additional to this insight you can get and also to the help you can get by the Copilot, you can also manually go there and fix things yourself. So for instance, we got that comment in one of our surveys that some queries like robots were a little bit too wide, and that's normal. We're selling robots. The word robot is It's part of everything. Is peppered all over Course, for example, right? Yep. So we have courses on how to actually manage and build robots, but if someone is just looking for a robot, it's quite clear that they're probably just looking to write robot parts. And you see it here as an outlier, like most of these queries are SKU specific, people know what they're looking for, either partial SKU search or full SKU search. But then here what you realize is that this guy has a bad recall. Everything comes back. So we'll try to go and fix it. So it's going to be fixed both for my agent, my website, and for my customers that are complaining. Here, if you look at it, you're going to see your metrics performance and then you can see the preview. And what they're complaining about is this. So you scroll, you get your robots, but then you get like training on robots or courses, which are not ideal I think. So let's create a rule. In this rule we'll use the exclusion here. So let's exclude a type of product which is service and training. And at this point, if you run that and you review your rule, you'll see by scrolling down that we don't have them anymore. We only have like arms and then we also have like some robots that are controllers for instance. So that's how you really control search, I'd say. And that closes that chapter of like how to implement through Lee getting content in Coveo through the push API or through the connect the catalog connector, the availabilities, price per users. How to manage, how to optimize, how to get insight, how to make sure that your search experience is optimized, also the relevance that the agent, the conversational agent, also uses to deliver product discovery experiences. Yeah. Now let's see. I mean, it looks good on paper or on screen, but now let's have a look at the future or basically what's live now, what's not. Let's call it the road map section. Yeah. So we really do believe that the future searches intend driven experiences. Right? So we saw at the beginning of the presentation today a traditional search. We saw a conversational experience. And a lot of the merchandisers and B2B manufacturing companies that we work with are wondering, how do I make those things work together? Should I show a should I have an agent only and have an experience that's agent only driven? Should I have an experience that's search only driven? Should I have two separate experiences? And have a chatbot. And have a chatbot, for example. Or can I perhaps just join them both, right? And this is really what we believe. We believe that the future really is gonna be intent driven. It starts with one single intent box where AI and ML can help optimize that journey of the experience depending on the user's intent. So The first thing we see here, which is call available, is the single intent search. So I'm looking for something specifically. What what happens? Well, here obviously it's a very clear intent, right? Single intent. We're actually gonna show instead of a big search page that could have many, many things. If the intent is clear, let's let the LLM and let's let the agent find us what we need and give us a very clear single intent type of experience here where we'll have the right products that we're looking for. The since you are scoped, I'd say, in a specific segment of the catalog, at this point, you'll see filters for instance, which will help you manipulate this a little bit better. It also follow-up questions, which is kind of typical in the in the different layouts we have. If we continue here, you see a multi intent search. What does it mean? Well here essentially the agent was able to detect that we're actually looking for two types of products, Two slightly different things. So let's actually show both. Instead of showing like a big melting pot of search results, right, with different things and tons of filters and lots of noise, let's help the user focus a little bit. There's two intents here. User's probably looking to buy two things. Let's actually just help the user show both with the multi intent layout here. So you get carousels, one for each, so it's a good separation of concern. And then, again, since this agent will split the query in two queries, you can have merchandisers to make sure that if you get a query with Drive system, promote the NextBuck robotic one because that's the new gen and that's the good value for the The profitable one. Yep. So it's very insightful, and then you can merchandise at the component level, even at the subcomponent level if you want it. That's really the beauty of agents is that if the query contains multiple intent, the agent doesn't need to send one big ugly query that's gonna be kind of a melting pot. The agent can split that in multiple queries and of course then generate a fantastic dynamic conversational experience. And since those are weighted sometimes by popularity or by my conversion rate, You can see here that harness of cable here, which I'm sure is very practical, but there are the chances this guy maybe this guy is powering all the robots, so the conversion rate is through the roof. So if you do one big query, this one's gonna be first. Even if it's a side project or kind of a you don't wanna promote the the dress pack cable first, you know? So it really allows them to get the best of both worlds, you know, AI ML powered behavioral enhancements, lexical semantic search with an AI agent on top to understand the intent better and guide the user journey. Now we get one of my favorite part, think, which is the melting pot of everything. And the reason I say that is the question is very broad. Like, tell me everything. Everything. Nothing short. Right? He asked for it. At this point, tell me everything. What kind of layout is this? Yeah. So this is a product education layout. Right? We're mixing again, we're starting to bring in some of that content that we mentioned earlier. You know, those big documents, technical documentation, guides, troubleshooting. Guy. Yeah. Right? Exactly. Right? That we saw the other master class. So we're starting to bring this together with the product to help not only just provide the right product to the user, help guide and help educate the user on how to use that actual product, especially when we get into complex B2B manufacturing. It's a key part of the user purchasing journey. And this is how also you drive confidence to your buyers. You know what you're talking about and you know that in confidence this is not just a purchase, it's a relation of business you want to build with. You build a robot, you're gonna have maintenance, you're gonna have like updates, you need to be in control and also be Able to install it to maintain and to be a trusted adviser for your customers. So you can see this guy goes in various parts where you're gonna have guidance, best practices, safety. You can have reviews in there. And those can be in your index. At Barca Robotics, they are highly organized people. They send all the data to to Coveo. This is the perfect situation. Life is not as perfect as this. So, basically, your your your training could be in another system and for a reason You don't wanna index it, you can just have a layout that contains the the the URL link for for the system to go poke that, bring it back, and display it afterward. Indeed. And the beauty here once again is that everything that you're seeing here is dynamically displayed at query time, at runtime, on an AI agent based on what's available in the index. Based, of course, on some preconfigurations, we do want to give those layouts that we're just looking at to an AI agent. But the AI agent will know which layout it needs to show based on content that's available for a given query. Another one we see here, a big seller I'd say is the product comparison. So tell me, how does it work? Yeah, this one is really about helping improve those conversion rates and average order value. So buyer is looking at two different things, The buyer is not really sure what he needs to buy. Of course, like I mentioned earlier, we have ML models that understand and have learned from behavior. I've learned what's most popular for a given query, so we can use this to recommend in this example what's the best product that will serve the user's needs the best, right? Provide comparisons between different options and actually help guide that buying journey so that the user can come to a conclusion and move forward with the right product. Based on the context, based on your entitlement, based on your location, based on all the factors we discussed earlier, it all trickles down. Personalized for that user. Yep. It's personalized, but it's not AI personalization like in the past. It's personalized, meaning it ticks all the boxes of everything you have in common or everything that you're entitled to see, which is the true personalization in my opinion. Lastly, you see at the bottom, we're happy to provide a little summary because why not? We got all the data we need to do so. If we continue here, we get the recommended configuration, a bundle. Yeah, this is very interesting for those large purchase orders, especially B2B manufacturing where one product on its own is not going to build your robot in this example, right? So you need to buy a few different products, and again the AI agent understanding the relations between the products like we were sharing earlier, for example, different sources in the index, availabilities, products, technical documents. The beauty of our index is we can all link these things together and as such we're able to provide those guided bundle experiences here where we'll find the right products, we'll bring them together in one big bundle, and help the user make that bigger, more complex purchase order. It's it's always fun to be able to come that's kind of put together these things. The solution architect in me would find very interesting to get with all these layouts, the digital requirement to do so because it would be very clear for a customer to understand if I wanna do this, what do I need to provide? These will comes in the documentation as we build and develop these. Last one here is the product search. So you recognize probably it's we're not very far from the product search we have here. What's the difference? This one is a product search page. It's an actual search page. This one is an agent that responded with this. Exactly. This was done dynamically. Right? So we've understood that, you know, some intents are very clear, like, and this is really the first item on our road map here. What what we've just seen so far is already available. Now we're getting into what we wanna do next. And the reason why we wanna make this available next is that powerful intent box that we mentioned can not only help guide a conversational experience, but we've seen that vague intent can sometimes benefit from a search. The user doesn't really know what he wants. The user's searching for just vaguely products in general. Right? And as such, let's perhaps let the user in those situations refine, use the standard search experience with facets to obviously get to the right. And once the intent becomes clear, we can navigate them back to a conversational experience like we were seeing before. And some of the customers in B2B, for instance, I was working with a customer in the appliance for restaurants and that kind of thing. At this point, you're gonna have parts and pieces, knobs, hoses, whatever, but you're also gonna have, like, open ranges. And I might just wanna see them all, just to browse Perhaps. Scroll. Perhaps. The infinite scroll section where I'm just trying to find the one I like, not necessarily Discuss about a problem. So the good old search interface is still valid. It's still a valid behavior I wanna have. And although B2B buyers might have a more specific intent, sometimes it'd be you're evaluating a provider and you wanna see a good idea of the range of products that they offer. So search still for Cheetah Skip. It's part of the mix. Lastly here, the buying guide, which I think is kind of a updated version of the education one. So you see here the query, I'm setting up first collaborative robot, so what do I need to do? This kind of goes hand in hand with the education but really aimed at let's help the user buy based on what they're looking to achieve, what they're looking to build. We're to find here you see at the bottom, if you need help sizing your gear, an application engineer can help you request a free sale design configuration. The interesting part here is the call to action. You could call for a sizing guide, for instance, which is another part of your website Or you could have a layout for that. So when you start understanding that these blocks are a reaction to an intent that goes either by searching data in Coveo or calling an external API, you start understanding the building blocks and how can you build something really The power of bringing us together an agent. Now, is the part that really starts to excite me. Right? The other master class with my colleague We saw it. It. We're looking at troubleshooting. Right? And these these two types of intents can actually come together in the B2B buying experience where, you know, you might have a piece that's broken, you actually need to fix the piece, you need to buy a new one, and potentially you need to troubleshoot first. So this is the types of experience you wanna bring. And as we were showing earlier, having all of this in the same index inside Coveo allows for us to juggle between these two types of intent. The purchase intent, the help support intent, and bringing those together. So quite excited about this one to help not only, like we've been showing, increase buying, right, increase conversions, but also actually reduce cost to serve, reduce inquiries. There's a few more layouts actually that that speak to that. I was naive in the day, but I was saying if you buy a very sophisticated laptop for instance You search on the website laptop, my intent is to buy, and I had these kinds of cases and issues that because the the the website back in the day wanted to show all all the time to catch everything. It was frustrating for me to get a brand new laptop and then I see problems on the Troubleshooting, broken, how to repair. Broken, how to repair. Was like totally afraid to get that laptop. Here it's very interesting because you're gonna show only when it's needed that part of information. When the intent is clear that we have a problem, we need to troubleshoot, we need to fix something. And sometimes the query, the answer is you need to purchase a part. Indeed. It can be a solution, you know. I think it covers what we had to show today. If you have any questions regarding how to deploy this, what's the new feature, how to get more or if you just want to talk to Matthieu, please talk to us. Reply to your CSM, stalk us on LinkedIn or even send an email. We'll be more than happy to help and get back to you with your answer. Matthieu, it's been a pleasure. Likewise. Thanks, Gavyn McLeod. Thank you.