Coveo Solutions Inc. (CVO) Earnings Call Transcript & Summary

Thank you for joining us for today's webinar for an update on Coveo's Relevance Generative Answering capabilities with the company's Founder, Chief Technology Officer and President, Laurent Simoneau. My name is Paul Moon, I'm the Head of IR for Coveo. Nick Goode, our Chief Business Officer, will also be available during the Q&A section and we'll be helping to moderate today's session. Before we begin, I'd like to review some quick housekeeping items. [Operator Instructions]. If your question isn't addressed during this session, we'll be sure to follow up with you after the conclusion. Please be aware that today's webinar is being recorded, and you'll be able to access the replay through the News and Events section of our Investor Relations website. Before we kick off, the information presented in the session should be considered alongside our annual information form dated May 30, 2023, accessible on SEDAR-plus at www.sedar-plus.ca as well as our latest financial statements and MD&A. Please keep in mind that the content provided here is not intended as legal, tax, regulatory, financial or accounting advice. During this presentation, we may also make forward-looking statements which are subject to applicable securities laws. Additionally, we may discuss non-IFRS measures and ratios designed to facilitate consistent performance comparisons or discussed key performing indicators essential to our industry. For additional details, please refer to the respective sections of our latest earnings press release dated August 8, 2023, and with that, I'd like to turn the webinar over to Laurent to begin. Laurent?

All right. Thank you so much, Paul, and good morning. Good morning, everybody. I understand it's a good early morning to some of you. So thank you for taking the time to be with us today. We tend today to stand more or less 30 minutes on content and give you the most recent update on our platform development with a strong focus on Coveo Relevance Generative Answering. I'm Laurent Simoneau, Founder, President and CTO of the company. I hope that this content is of interest to you. And we are also looking forward to a good, thorough Q&A after the content. So with that, I'd love to start with just a highlight on our platform. One of the key differentiators of Coveo in the market is we have one single platform that runs from all of our customers tiers, one single code base, there's 20 updates a day. And this is from that platform that we're looking at generative answering in the context of Q&A. So in the middle here, we have our key capabilities of the platform where we are known as an intelligent search platform for the enterprise. There's prescriptive recommendations, unified personalization, merchandising, mostly in the context of commerce. And now we've added generative answering here as one of the key capabilities of the platform. At the bottom, you will see our connectivity and content layer where we have the ability, and that has been to a long history of investments and refinements and innovation, we have a very strong connectivity layer to basically any system within the enterprise that has an API and that is worth indexing. And then at the top of this slide here across our various lines of businesses, we have native integration where either employees work and/or users consume information. So you will see different systems on the top and have a more detailed slide on this a little bit later. And I remind you that our core principles of Coveo, we consider ourselves as the intelligence behind the experience. We -- while we have out-of-the-box experience layers and UI layer, for customers, typically we also get integrated in the flow of work. So you will see a search box within enterprise systems, such as Salesforce, ServiceNow that are often powered by Coveo, and we have an API -- basically an API play multiple ways for our customers to integrate Coveo into their storefronts, into their different portals, their different experiences. We like to do this with high scale, so a lot of queries, a lot of content, a lot of transactions, a lot of diversity of content. We focus on data, on behavioral data, so we can understand the journey in the context of service, in the context of commerce, in the context of employee facing workflows, and this allows us -- as to build specialized AIs. And we do all of this with the utmost respect for privacy and security rules and regulation. That's basically the most -- that foundational and it's the most important thing that we do. And if I double click on the previous slide, that's the more detailed slide, I won't go through all of those boxes, obviously. But I just want to highlight a few things here. First of all, there's the indexing here, the search and indexing engine that is our own. And we've added the vector database directly within the index. I'll talk a little bit more about that later. But we believe that's a key differentiation versus those specialized systems that will do question answering. We have security cash directly in the index that has been there for 15 years is the ability for documents that have permissions to be indexed with their permissions and also to have roles and groups being synchronized and replicated directly within the index. We have the connectivity layer here on the left side and then we have the AI, machine learning subsystem here that supports different methods, different models, different approaches. You have our different integrations at the top and the ability to integrate through directly through headless at the API level, but we also have visual components frameworks. And then we have the ability to connect to a large language model. And most recently, we have announced additional -- so we are in a calendar year here. We have announced additional regions that our primary regions for Coveo for Canada at the start of '23 and then we've announced also active [indiscernible] for our U.S. customers, allowing us to support 5, 9 SLAs for those customers who would like to buy for 15% premium on subscription. This is fairly new, but this is quite important for a subsection of our customers that are working -- that are supporting workloads that are critical with Coveo. So a lot of maturity here, as I said, about 20 releases a day, no downtime, virtually no downtime for the past year. And we expect this to support the workloads of our most extreme customers. Think about Black Friday and e-commerce and think about some major product releases in some of our customers on the service side. So I'd like to focus now on Coveo Relevance Generative Answering. We have announced this earlier this year with a focus first on customer service. Now customer service is really knowledge that is both on the self-service side for customers, but also on the assisted service side for agents that are trying to solve these customers. So we power the entire journey of service with our service cloud integrations with Coveo Service Cloud integration, and we've added Relevance Generative Answering within that platform. We're adding it at every step of the way. There is -- we have also announced earlier this year, our data program Relevance Generative Answering. And we have our best and most advanced customers on that program. We have 25 also additional enterprise customers in the advisory group. And what we're seeing right now is a lot of excitement. I would say that we are quite impressed by the maturity of these customers and their expectations. We are seeing great early results and we look forward to share more later this quarter about those results. So what these customers are looking for, right? And this is my colleague, Louis shared that slide with some of you earlier this quarter, these are at the high level the 9 CIO Headings and why they care about this in the context of generative answering and Gen AI. Security, privacy, dealing with enterprise proprietary content in multiple content sources, currency and factuality of content is also a priority, the source of true verifiability. And then there's high cost of generative AI and coherence of the multiple channels, search and chat channels, for example. All of this needs to be taken into account for any Gen AI projects, especially generative answering. So this is how we are addressing this problem. This is Coveo on the left side. And some of you have seen this, but I think it's critically important to show that again. Coveo is the left side of this slide, right? That's classic Coveo. So you've got secure connectors that are feeding an index that with Relevance provide great results, classic results. A few key words, navigation, slice and dicing of the content. So this is a classic search. This is not going away. 80% -- 90% of the queries depending on the customers will still be powered by that in a search box. There's native UI integrations, there's mature administration, there's mature stack of analytics, right? And a lot of large -- most of the large enterprise customers had various forms of data already implemented. And then what we've seen in the past month is the emergence of disk. Question answering systems built around RAG principle that are going to use a large language model and vector database from specific source of information to provide answers. We think while appealing, we think it's a bad model. It's a bad model because it implies different search boxes, different search methodologies to provide different facts to the same question. We think it's bad and we think that the enterprise deserves better. So that's what we've built. We have consolidated the right side with the left side. So we already have the left side, right? That's historically what we've been building for more than a decade now in the cloud with a lot of maturity. So in the past months, we have substantially scaled our ability to do vector search directly within our index. So for those of you who are a little bit more into those details. So we're -- we've added the embeddings at the indexing level, so it allows us to do in one single operation, classic search, navigation, the AI and machine learning relevance and the semantic search, all in the same operation. Therefore, it allows us to leverage all of the connectivity, the security, the scalability of the platform and also to reduce costs. By doing this here, we also have the ability in the context of semantic search to also pass a large language model to create a response grounded with the best exert or passages from the semantic search that are returned with the highest relevance. So from a back-end side, the advantages are depth, breadth, and freshness of content, security. It all comes out of the box basically and the administration analytics. And then from a result and relevance, we have one search box that deals with the different methods of search and a query for all of the queries, the generative answers are granted from the most relevant paragraphs or passages extracted from semantic search. But again, those paragraphs may be highly relevant for the query, but come from a document that is not relevant, that is of low quality. On the flip side, we may have good decent answers from great content that are highly relevant, that are highly credible. So because we have access to both sides, we can balance this for customers and we can come up with the best combination of both. Personalization is also important. And using all of these key ingredients, we think that we have the best protection against hallucination, because basically, we're going to ask the model to generate an answer based on the most relevant, personalized and secure content that is resulting from the search. So with that, I'd like to pause here the slides and switch to a real demo. This is customer zero. This is Coveo. This is on our online documentation. So you can go and try it, you can test it for yourself. As long as you know what's in there, there's a little bit of content and it is very popular for our customers, but it's a little bit [indiscernible]. So there's a search box here, right, on our documentation. So this is a classic search, right? I'm looking for service cloud, and I will get results here from our system. That's Coveo Classic what becomes interesting here is, let's say, I will ask questions here. How this will determine relevance. So when the system will do is based on the different results at the bottom, get the best passages and that to the large language model that will return an answer here, right? So you see that the answer is generated and we have the [indiscernible]. It's coming from those results at the bottom. And we have the source of through, right? So about search relevance here, search results ranking here, I can open the results. All of this is based on APIs. So our designers and our customer designers are testing all sorts of designs and all sorts of the UI. So expect this to evolve quite rapidly over the next few months. Let's do another one. Here steps to install Coveo for Salesforce. Okay, that's interesting, but I don't -- I'm not sure I like -- I think the response is a little bit bulky. So let's try something different. Detailed steps to install Coveo for Salesforce. So I can start playing with the answer styling directly in the prompt, which is super helpful. I can get a little bit more detail than say can Coveo work into ServiceNow in 300 words and how to install it in a number of bullets. So let's try that. There you go, coming from multiple results here, right? And then what's pretty cool also is you sell all those facets to the last. So this is metadata that is coming from the content that has been in there. Obviously, customers will decide what they show here depending on what attributes and what the data is available in the result. But let's say I do this here. Let's say I do detailed steps on how to change resolve template. So what's interesting here is, I have multiple facets here, right? So if I click on e-commerce, the result template will be different. So the answer is grounded in the context and the context is powered by search -- classic search. So what we are starting to see, looking at the analytics of our customers and testing that is going on is that a lot of the behavior of the end user is tailored across slice and dicing, across let's refine and let's navigate into the content, therefore, you can conceptualize the answer with a few clicks. Very quickly, this will become available into our customer community. Therefore, results will also be grounded to what the logged in user has access to, which will give an additional layer of personalization. I'd like to comment on what's coming soon where actually it's been tested at our customers as we speak. There's a first step, is about in line citations. So citations that you saw at the bottom, we want to inject that directly within the text. The challenge here is that many large language models are not very precise, and we'll have a tendency to hallucinate when suggesting an in-line citation. So it will suggest that this is coming from citation #1, but it's not. In the enterprise, you can do that. We have ways to validate that and refine that and remove citations that are not credible, but it's a little bit of additional work, but we are in the process of releasing those. And also the answer styling. So remember my detailed steps and 300 words and so on. So we are testing various ways here from a UX perspective, to make this available for the end user so that he can decide to have a bigger answer, shorter answer and answer with more step by step search bullets. So this is being tested right now. And then later on, this is where it becomes quite strategic. It's the -- it's to have the ability to transform the search box into a conversation box also. We believe that -- and this is confirmed by our customers. People and users are way more comfortable with a search box than with a chat box when they have the choice. We are not in the process of transforming our search box into a chat box. However, if the end user wants to have a conversation, wants to have follow-ups, may the search box is the right anchor to do that, maybe the search box is right UI to do that for most users. So that's what we are also adding in building. And we expect that this has the potential of changing the customer engagement model down the [indiscernible]. We're not done yet, but I thought I would highlight a few things that we are working on that have a lot of potential. So there is the ask follow-up here and then suggestions of all queries. And obviously, those follow-up queries are personalized, are driven by the context and are specific to the user that is currently testing the system or using the system. We also have an opinion on e-commerce. So there are various potential use cases in e-commerce that we're testing. One of them is more around semantic search. So what are the best satelboard for beginner, right? But what becomes really interesting is this. So highlighting the key attributes that as an end user, I should care about related to my query. What are best satelboard for beginner. Well, Coveo through semantic search and some LLM, we have the ability to surface the best basically attributes or facets that should be considered for best satelboard for beginners. Now what we're going to do with this is obviously highlight that on the left side on the [indiscernible] but also there may be some conversational workflow that will come out of that. This is early, but I thought I could show you that as something that we're working for the future. Finally, security is critically important. We -- and I want to go through that slide. Each and every of our customers invest and use Coveo because of our ability to secure their content, secure their usage and make sure that the foundational of -- the foundations of Coveo is as secure as it can be. So from a generative answering standpoint, there are a lot of questions, right? So how do we deal with security in that context? First of all, make your content retrievable. All of our connectors have what we call a security provider in it. And this is what's indexed. This is what's synchronized in the index. We deal with structured and unstructured data with document-level security. Then the grounding of the prompt is dynamic. So I showed you by clicking on a [indiscernible] which would change the answer. Well, basically, the grounding is dynamic based on the results that are coming out of a secure search and personalized search, and that's another element of security. Every prompt and responses are auditable. We are -- we can track that, just like other queries, and there's the ability to go see what happened, who did what and when. There's zero retention by our large language model provider. Currently, it's open AI, GPT 3.5 running on our Azure plan. So there's zero retention here. And we are working on data matching to add an additional layer of security. Since the large language model is running directly on our Azure Cloud instead of on another cloud that customers may not be, I would say, mature with. This is less of an issue as of today. So with those key elements, we can have an advanced conversation about security with our most complex customers. So in summary, cost matters also. While this is evolving, while this is getting better, we -- our initial testing suggest generating a response with GPT 4 is about 1,000x more expensive than generating search results from a classic search, 1,000x. Now you could argue, and we also argue that a GPT 4 answer is worth probably 10 search in terms of richness and complexity and all of that. Okay. So it's a hot -- it's 100x more expensive. So how do we deal with that? So there are various ways to deal with that. One of the ways is to only use generate answering for a handful of the volume of the queries that are happening and classic search for the rest of the queries. So that's first step. And the second step is there are various ways to optimize this. So large language model providers will do a better job down the road. We're sure Moore's Law will help. But at the same time, there are all sorts of cash in mechanism that will put in place down the road on [indiscernible] that will help manage costs. Trust is not an option. It has to be foundational. We expect to see domain adaptation of the large language models down the road and more and more open source, we saw a lot more too that are doing a great job down there. But there will be all sorts of large language models out there evolving, making this required. So we are designed to be a large language model agnostic. There will be some work of the prompt level to reduce hallucination and deal with some, I would say, dysesthicity from each large language model, but this will be pretty light. The foundational system is large language model agnostic, so it means that those customers that have their own version of large language model internally that they are fine-tuning towards 3 times a year. If they want to use that in the system, it is designed to handle that with limited investment. So our belief is that in the end, any systems will require relevance across all content and all interactions in order to generate great answers. And that's what we have built, and that's why we're quite excited about the future. So with that, I would pause here and let's open up for Q&A.

It's great, Laurent. Thanks for that. So there's the Q&A function in the webinar at the bottom next to participants. So if you have a question, please go ahead and use that to ask it, and we'll get through as many of these as we can. So first, a question, Laurent, can you give us a sense of how much personalization goes into the platform for each customer? And are you finding that the platform is easily transferable across industry and customers? Or does it require heavy lifting to make that work?

So that's a great question. So personalization will vary from one use case to the other. So let me give you an example. In e-commerce, in B2C e-commerce, most of the activity is done through anonymous sessions. People are not logged in. So the personalization there will happen using product vectors, which means that looking at the behavior of data of all of those in different sessions, we know we create a natural path within the product. So with a few clicks, not on this user will get personalized results automatically, but that's the anonymous way of doing things. Then you have the logged in or authenticated way of doing things where we look at the user history, we look at various aspects like themes and topics that this user may be interested in that has been joined automatically, and then we're going to personalize based on that. So service and knowledge, people are mostly logged in. Commerce, B2C commerce, people are mostly anonymous, that's where it differs. And I think the question is -- the question also implied, do we need to do customization and so on? Well, the core capabilities are the same. So the customization will be in the -- do we want a log or do we want to track additional metadata for user and things of that nature that is specific to the business, but it's typically pretty light.

Yes. Maybe talk a little bit about the implementation process, Laurent, for an existing customer, how easy is it to get up and running with Relevance generative answering?

That's a great question. And while we need to be careful because we don't know yet what we don't know as of today with those that we went through the process with. It's substantially easier than expected. Actually, most of the work has already been done, meaning the connectivity work, the indexing, the relevance and so on. Therefore, adding this capability on top of the existing stack is pretty straightforward. Now we're going to see multiple situations that are corner cases because that's the business we're in and that's fine. But the status as of today is quite positive.

Perfect. Several questions on ROI here. So Laurent, in terms of the early indications around ROI and some of the metrics that we're monitoring in terms of folks that are using Relevance generative answering on the service side. What are some of the things that we're looking at? And I guess, how are we going to be able to show folks that there is an ROI here? And I guess maybe a follow-up there, what's the timing around that in terms of when we're going to feel confident that there's a clear ROI here?

Yes. Yes. We're in the early innings of this, but there is -- we -- our expectation is that this will have impact in case of flection. Obviously, since search already has a huge impact in case of flection, it's been very well documented in our case studies on our website. So this will augment the case of flection impact. And that's what we're starting to measure and are going to start to measure with our existing customers, but that's kind of the low-hanging fruit. And then we are going also to see things like I showed in commerce. It's currently very early. But down the road, this should have an impact on the revenue per visit and conversion down the road. So that's the -- that's how we are looking at this one.

Perfect. In terms of costs, so this is -- and I know you touched on this, what are we seeing so far in terms of costs relative to our expectations? Obviously, we think the way we approach this is unique and the way our platform is architected, it's in a way where we should be able to do this and run Relevance generative answering in a relatively low-cost way. Is that still what we're seeing so far with the early customers?

Yes. I think that our initial assumptions were the right one. You need to be careful about this, I'd like to say slightly Gen AI is like a very funny moon. And sometimes when you come back home and you look at the credit card bill, it may leave some bad memory. So there's -- there needs to be search to cover the 80% or 90% or 95% of queries that will not be good for Gen AI. So that's how we control the whole thing with all of the obvious optimization that we're putting in there. That's the job we are in. That's what we do for living. We optimize our scale, and we make sure that we have best-in-class COGS, platform COGS. So that's culturally what we're looking at.

Yes. Another question on the PII side, Laurent, is I think you mentioned that data masking is something that we're looking into in the future. How are customers dealing with data privacy today? And is PII management a hurdle for enterprises in terms of what we've seen so far?

Yes, that's a great question. So it is less of an issue, given the fact that technically Azure is part of our cloud. If this has to go to an external service, let's say, open AI or whatever, right, in their service, while great, this opens up a new set of questions. And it potentially requires data masking and additional measures to put in place because Azure is part of our cloud, those customers are quite fine with it.

That's great. Laurent, in terms of the beta program customers so far, are there any interesting use cases that those customers are wanting us to help them with that are not on the near-term road map? Or is it more of the same in terms of self-service and the existing use cases that we're helping them with in the beta?

Yes. I think what's pretty obvious is self-service side and the assisted service side for the agents or the help desk in the enterprise actually. So that's the obvious one, that's [indiscernible] low-hanging fruit. I think I alluded a little bit to what we're [indiscernible] with commerce in B2C and then I would also say that we have conversations around commerce where we expect to summarize content, right? How to build this patio or how to configure this kitchen, right? So there is a potential of using Gen AI to provide richer responses to those kind of queries. But this, I think, will be step #2.

Yes. And just to follow up on that, Laurent, because we've got a couple of questions on this. So we've talked about, obviously, service and starting there. What's the sequencing in terms of of how we plan to roll this out to -- when I say this, I mean Relevance generative answering, to each of the LLM will be, so we talked about commerce. But what's kind of the timing there? And what's the order of operations there when we talk about websites and workplace in addition to commerce?

Yes. So in our 40-ish customers right now, we're -- we have either actual projects or concrete conversations. Most of them are around service and workplace, around the big knowledge umbrella, if you want. So it's kind of the same workload. It's either for employees or for customers. I think that website is a natural extension of that and we will see some websites coming out of these projects or I expect that this will happen. And then commerce is a different animal because it's -- typically people in the B2C commerce store, they don't ask questions. They slice and dice content and navigate. They go into a category page. They select this, they select that. So we -- it remains to be seen to the extent that this will have in commerce. I think that there will be impact on commerce, but there's -- it will come after the first service -- that service and workplace.

Yes. Interesting question on costs again. So GPT 4 is obviously the -- they used the word Ferrari of LLMs, which is interesting. And that the initial testing puts GPT 4 at 1,000x more expensive than regular search. Do you get the sense that it's really that much better than GPT 3 or 3.5. And then if the answer was we do have to use GPT 4, what does that do to margins?

Yes. So we don't have to use GPT 4, GPT 3.5 is super good for the use cases that we're supporting. And competition is also coming. So price will go down. Now the -- however, the -- I would say the additional variable that will need to be considered for enterprises is the cost of fine-tuning those models. So some large companies would want to have these models running internally in a fine-tuned fashion. So the early signal that we have is that at least for some of them, it will be -- it is currently super expense or close to prohibitor. So right now, our 3.5 is the right way to go for us. We may have customers who want to pay more for GPT 4 or to run a fine-tune model internally, and we'll be glad to support them if the needs happen.

Yes. And Laurent, just for folks that maybe aren't familiar, can you talk about what fine-tuning is and what that means and where we expect kind of that component to become much more critical in the future for customers that are using generative AI?

Yes. So fine-tuning means that instead of using a generic model that is updated a few times within a year or multiple times within a year, but each and everybody uses it. Fine-tuning means that you're going to recreate that model or rebuild that model using specific information gap is provided by the customer and both the fine-tuning operation is extremely expensive. It depends on the model. It depends on the provider obviously, but as a general rule, it's extremely expensive. So you can't really do that more than a few times in the year, typically, if you want to go there. And the second thing is powering the queries basically, generating the summaries is like 6x to 8x more expensive than with a generic model. So there will be players out there that will specialize in the fine-tuning. There will be competition and so on. But as of today, this is something that is quite expensive.

Yes. That's great. You talked about the SLA agreements with customers or with the 15% premium for enhanced performance. I think that's active, active, if I remember correctly. What prompted the introduction of this? And any commentary on customer adoption and potential uplift there?

Yes. So what prompted this? It was 2 years ago when [indiscernible] U.S. on the East Coast collect, and it was a major event. Netflix went down. Part of Salesforce went down. I mean, Roomba when the multiple part of the Internet went because of that was a major event. And this happens -- I don't know, it's like a big [indiscernible]. It will happen every 10 years, 5 years, 2 years, I don't know. But we decided that -- and I think that our agility and our design allowed us to come back in 45, 50 minutes, which was one of the best performance for those who were affected by U.S. East. But nevertheless, 45, 50 minutes, if this happens on a Black Friday, for instance, it's not good news. So what we decided to do is let's build an additional layer of protection in our architecture, let's go Active-Active on U.S. East. So if this happens again, it may never happen again. It's like an insurance policy, right? If this happens again, there will be a few hiccups less than a minute, let's say, a few during 30 seconds of problems, there will be a few hiccups and then this will move on to the Active-Active. So what do we see as customers? Well, first of all, it came from customers. They asked us for this additional protection and the way to commercialize that is to -- instead of selling Active-Active, we're selling a 5, 9 SLA, which is [indiscernible] well known by customers. So we expect this to grow down the road. We -- and yes. So it's fairly new.

That's great. Maybe talk Laurent a little bit about our partners and generative AI in terms of where are we at with folks like Salesforce and kind of how do we -- how does what we do with generative AI either complement sales, what Salesforce is talking about or is better than what they're talking about? And just kind of where do things stand on the integration front?

Yes. Yes. So Salesforce has huge ambitions around Einstein GPT and they are investing quite a lot in the AI cloud and in their data cloud. We have a different approach where instead of covering all the use cases, like as you know, Salesforce is looking at generating documents, doing segments and marketing cloud and so on. And yes, they will do some summaries also. Our view is that with all of the connectivity that we have, with all the security, with all of personalization and relevance that we have for the generative answering side, those customers who want to get to the best answers because they have a use case like case inflection, they have a use case like self-service support and so on, our customers will select Coveo for those use cases. We expect down the road that there will be some, I would say, additional coexistence because that's already the case between Salesforce and Coveo. Coveo is integrated in the Salesforce. We're sharing behavioral data with Salesforce. We are leveraging some of the behavioral data with Salesforce, so we expect that those systems will continue to evolve together and one will make the other better [indiscernible].

Yes. That's great, Laurent. Maybe just building on that. So otherwise, in terms of competitors that are out there today, I guess, who are we seeing so far? And then where do we differentiate versus those folks, including as it relates to cost and pricing?

So okay. So that's a great question. when I spend time with the CIOs of our large customers, they always come up with -- remember, I'm at the right side of my slide, those custom-made systems where they need to bring a vector database and there and build that on the side. So that's like that's one -- that's the one competitor to a certain extent. And we feel really good about our ability to provide the left side of slide with the connectivity with the relevance, personalization and so on and that makes a huge difference. So -- but at the same time, we need to make sure that we still invest and make sure that our semantic search capability scales and leverages all the greatness of what we already have on the left side. So that's one part of -- that's one area of competition. And then we have the classic fellows that some of them will be more designed for a build model. Others will be focused on certain use cases, and they will bring some generative answering in certain use cases. So it is basically at a high levels in competitive landscape than we used to have, build versus buy and when we buy, we differentiate it with our connectivity and it's with our security and personalization relevance and focus on service and commerce actually.

That's great. This is the last question that I see, so we'll see if anybody has any others. But do you see having the generative AI capability creating or enabling new use cases that weren't previously relevant based on the search functionality alone. And the one that I think comes to my morale is kind of what today is done in the conversational window. So maybe you can talk a little bit about that.

Yes, exactly. So and let me go back here because I think it's worth another look at this. I don't know if this qualifies as a new use case. But I think that if it brings more volume, more attention to the search box. Our belief is that search box is the universal way for users to interact with information. That's our belief. By making this conversational, there will be more usage, more value created from that search box. It translates more call deflection. It translates in more conversion translate in more productivity for the employees. So that's a new way to look at customer engagement to an extreme extent. And so -- and the [indiscernible] enabled with semantic search and large language model.

That's great. Well, Laurent, it doesn't look like we have any other questions. Folks, thanks so much for joining. If you have additional questions or ones that didn't get answered, I think we got through all of them, but please don't hesitate to reach out to Paul or myself, and we're happy to take those separately. But hopefully, this was informative and helpful, and we'll look forward to keeping everybody updated on our progress as we continue to have more folks getting involved in the beta program here and continue to roll this out across the customer base.

Thank you all.