NVIDIA Corporation (NVDA) Earnings Call Transcript & Summary

Good morning, and welcome to How Telcos Transform Customer Experiences with Conversational AI. Before we begin, we wanted to cover a few housekeeping items. At the bottom of your screen are multiple application widgets you can use. All the widgets are resizable and movable. If you have any questions during the webcast, you can submit them through the Q&A widget. We will try to answer these at the end of the event. A copy of today's slide deck and additional help materials are available in the resource list. We encourage you to download any resources or bookmark any links that you may find useful. You can find additional answers to some common technical issues located in the help widget at the bottom of your screen. An on-demand version of the webcast will be available approximately 1 hour after the webcast and can be accessed using the same audience link that was sent to you earlier. Now I'll pass it on to Lilac.

Hello and welcome, everyone, to our webinar. We're delighted to have you here with us today. My name is Lilac Ilan and I'm the Head of Global Business Development for AI-Powered Operations here at NVIDIA. Our goal for this webinar is to provide you a valuable insight and practical knowledge about how telcos are transforming their customer experience using conversational AI. We brought together a team of experts. They'll be able to address any of your questions and share their insights with you. I'm going to actually let the team introduce themselves first. So Anand?

My name is Anand Santhanam, based in London. I run the global strategy for the telco business unit of Infosys. I've been in the company for 28 years and specifically interested in the customer service aspects of how telcos are transforming themselves.

Awesome. Thank you. Ken?

Hi, everyone. Ken Goldberg from Talkmap. I run Business Development, including sales and partner management and been with the company and involved since the founding. Good to meet you all.

Good to have you. And Dhanya?

Hi, my name is Dhanya and I lead the GTM Strategy team for Telecom vertical at Quantiphi, focusing on the NVIDIA partnership. I work closely with NVIDIA team to help our customers solve business challenges using Quantiphi's cutting-edge AI/ML solutions.

Awesome. Good to have you here with me, guys. So we divided the presentation to different segments, each focusing on particular aspect of the topic. So here is what you guys should expect to learn and hear from us today. First, we're going to understand the telcos' key challenges and where they're going to -- where they're planning to utilize AI. We're going to define what is conversational AI. And we're going to talk about how NVIDIA empowers developers to use conversional AI. And then we're going to turn it into more of an understanding from our customers, examples from our ecosystem, hearing from Dhanya, from Ken and from Anand. And then all of us, we're going to have a short round of roundtable to talk about the future of conversational AI specifically and how it relates to generative AI. And last, we're going to have a quick session for Q&A that will give you the opportunity to ask us any questions you might have. So with that, let's dive into the first topic. Recently, NVIDIA asked 400 plus telco professionals about where are they going to plan to utilize AI to solve their business challenges. Customer experience and improving customer experience was very top of mind. In fact, approximately 50% of the responders said that they're planning to utilize AI to improve customer experience. Furthermore, Gartner estimates that by the year of 2031, there's going to be $240 billion of savings, thanks to conversational AI. So what is conversational AI? Conversational AI are really technologies that enable computers to understand, process and respond to human language. And has 3 technologies kind of part of it. The first is all about text-to-speech, text-to-speech, all in multi-languages, all in real time. That's the foundation. The second pillar is natural language processing. It really allows a computer and a user to interact in more human-like manner and the recommendation AI -- recommendation AI really provides real-time recommendation based on past preferences and other factors. Conversational AI, guys, is part of NVIDIA AI leading platform. Specifically, NVIDIA Riva is the GPU-accelerated multi-language speech and translation AI SDK and is really meant for building fully customizable real-time conversational AI applications. Conversational AI is really shifting the way business interacts with the customer. It allows for real-time analysis of customer feedback, automation of customer interaction and allows for assisting the agents with more accurate personalized recommendation and transcript. I'm now going to hand it over to our experts, Anand, Ken and Dhanya, so you guys can hear from them, how they're using NVIDIA Riva in their solutions to the telcos, helping them to transform customer experience. I'm going to start first with Anand from Infosys. Anand?

Great. Thank you for that, Lilac. I want to step back a little bit and look at the whole customer experience from the large enterprise perspectives. We have the fortune to work with a lot of leading telcos around the globe and I have been personally involved with them for about 20, 25 years now. And there are 2 perspectives I want to share, one from inside the enterprise, inside the telco enterprise and one from external. Now when we strip away all the language, all the jargon, all the terms around customer experience and distill it down to its most atomic expectations, what customers have, it comes to these 4 simple things. When customers reach out to telcos asking for help; know me, understand me in my preferred way and help me promptly. All the surveys that I have read eventually comes to just these 4 basic expectations. Know me is identification. And then I call in or when I chat or when I e-mail, in any way, know what identity is and my account is. Understand me is understand the context, understand the services, understand the products that I have. Interact with me in my preferred way and most importantly, resolve what I'm calling for promptly in that conversation. Simple expectations but distilled down to its most atomic aspects. Now from a agent and advisers who are helping the customers, I bring this down to, again, the most atomic expectations. Hey, educate me in your systems, in your processes, empower me with the right tools in my preferred way, to help my customer promptly. All agents, advisers, everybody wants to help the customer and deliver a great experience. It comes down to these 4 things. Simple enough and that's why I put that at top of the iceberg. Looks easy but now the perspective from inside the enterprise. Having worked with them, every large telco has these layers of complexity built in, starts with the way of working, encoded is the business processes, business policies, rules, procedures. And those are manifesting itself in the data and business rules which are there in the applications, the system of records, as we call it, all of these delivered to various channels and devices and these deliver the experiences. Now these layers of complexity and particularly larger the telco and longer they've been, it becomes far more complex that when you change anything in one of these layers, it goes and creates seismic uncertainty and unpredictability in the layers above. So we want to keep this perspective that AI and customer experience when we chat, a bot needs to think across all of these layers to be able to deliver sustainable result for the customer. Now external perspective, 3 large macro shifts that we are all living through right now. The agent and adviser workforce is -- up to 87% of this workforce have been born in the digital era. They are used to great experiences. They are used to working with technology far more easily. They work in a hybrid fashion. They expect from the work creativity, purpose and innovation and they want an understanding of the emotional quotient. This is the agent workforce and we need to keep that macro perspective. The second one is, while we have been investing in digital technology and digitizing the process, studies say that only 9% of all the customer interactions are wholly contained in the digital way. It eventually does land -- 91% of them land with humans. So we still need to work through enabling the human with AI in servicing their customers. And lastly, of course, what the topic of today is, which is the investment in AI through companies like NVIDIA and Riva product on what enterprises are going to do, expectation is that, that would increase about 4x in the next, 4x in the next few years. So definitely, as Lilac said, conversation AI is going to be the differentiator as we go forward. Now at Infosys, we have developed a platform called Infosys Cortex, to do this end-to-end, all the way from the expectation all the way down to the lowest aspects of the enterprise. Infosys Cortex is a platform with 4 modules. The learn module starts with everything to do with the agent, empowering the agent, in fact, starting with using AI to hire the right agent with the right skill set and the right mindset to be able to service that customer and provide them with the right education and continuous education in the way they learn today. What this helps us, is in hitting time to hire, effective hire and speed to proficiency, 3 KRAs, which customer service heads care about. On the top right is a set of functionality, which is about empowering the agents and this is empowering them real time as they are speaking and interacting with their customers through nudges, behavioral nudges, operational nudges, compliance nudges where -- what they need to be able to say and convey, guide them with contextual workflows through the systems. And remember, I spoke about the enterprise complexity, so guide them through various systems and policies and procedures, bring the search to them rather than they go search into documents as they are holding their customer call, bring the search in the right context on their screen in front of them. So capabilities around empowering the agent. Bottom right, post the conversation, taking these millions of records, millions of transcripts that Riva empowers and those speech transcripts are available for us to go mine into and study and derive value from. And the bottom left is actual, the experience through various technologies like avatars and augmented realities and help the customers through that experience. So we brought these 4 modules together. These are built into a LEGO block approach. So the architecture is microservices on the back end and micro UI on the front end. So any one of these LEGO blocks can be used and plugged in. So -- and it's available not as a monolithic platform but as a real plug-and-play that customers could build their own set of AI capabilities for influencing the customer service. So I'm going to now touch on one of these aspects, language neutralization, where we have partnered with Riva and what this looks like. So the scenario here is typical of global telcos where a customer calls in with a different -- with a language and the agent is of a different native language speaker. How we could neutralize this so that we are; a, able to have effective conversation and not limited by different native speakers. But most importantly, once we have neutralized that, we are now able to apply common algorithms to analyze through these conversations and understand what's really happening. So I'll play a quick demo and we'll then chat further. [Presentation]

As you can see, what -- we have solved a problem of utilization of adviser and agent resources, which is a big problem when you need to pair up different languages, different accents, to be able to service your customer base by neutralizing languages, we are now able to increase that utilization, which is a huge saving for customers. But the next advantage of language neutralization is using those transcripts, which by bringing down to a common language, we are able to run algorithms which allow us to extract value. And over here, I show value in 3 different forms: value in engagement value, which is the lowest part of the spectrum, where just improving the conversation that happened and advising the agent on how they can improve their sentiment, the choice of words, grammar, can be achieved through extraction of intelligence from these conversations. But beyond that is the process value, by looking at patterns across multiple calls, we can understand root causes. Why do customers even call, predict the intent and improve the performance of our team as a whole. But even beyond that is where we believe business value could come in, where deeply understanding likelihood of customers to buy, for instance, #34, predicting churn of customers, understanding where they -- what do they perceive about our brand. All of these are value that can be extracted from these conversations. So conversation data is the new gold, is how we see with conversational AI and our partnership with NVIDIA and Riva. So that's what I had, Lilac at this time. Over to you.

Thank you. And one of the other values that we should talk about for language neutralizer is, how it allows for companies and telcos to expand globally with much ease and allows for global support. So I. appreciate that. Ken, let's hear how Talkmap uses Riva.

Great. Thank you so much, Lilac. And sharing my screen. I'll tell you a little bit about Talkmap and then I'm going to share a demo as well. So if you listen to the top market analysts like Gartner and Opus Research, they call our platform conversational intelligence and they both recommend us to large service-intensive enterprise -- enterprises. And our focus is on large telecoms, utilities and financial services companies. And so today, our customers include 2 of the top 5 telecoms and 1 of the top 4 banks in the U.S. And our focus is really on understanding human-to-human conversations and providing visibility and business intelligence and enabling enterprises to be able to find and discover insights in that data and really tap into those conversations. And so we hear from -- most enterprises today are just unable to tap in and to really understand those conversations and there's a tremendous amount of value in that. So today, companies are really just looking at less than 1% of conversations. They're sampling conversations and they've got dozens of people, if not hundreds of people that are doing months of analysis and still just looking at under 1% and not able to see and have visibility into all of their customer conversations. And so there are lots of tools out there and there are some very good tools that enable you to look at some of these conversations but they really don't tell the whole story. They're very limited. Today, we generally sit alongside other vendors for things like speech analytics or customer surveys and we will sit alongside them and provide a lot more visibility into those conversations. And we're doing that at scale for these large service-intensive companies. So we offer an AI platform that can really give visibility into 100% of customer conversations. And this is based on -- we've got 7 patents awarded. We've got half a dozen patents on the way as well. And we've been working with GPT and large language models since 2018. So I'm going to tell you a little bit about the platform here. On the left side, we essentially ingest and process automatically transcripts coming from live chat or voice. And so those transcripts of voice are ASR, where speech-to-text engine of choice is NVIDIA, the best high-quality transcripts naturally provide the most value and the best outcome. And so in the middle, our platform ingests and processes data. We train a series of models based on each enterprise's conversations. And so with that, we will -- we are able to discover intents, customer intents or the reasons why customers are calling. And then based on that, we will then label, essentially enriching the full conversation down to the utterance level. And then we will analyze and provide trend analysis and even correlations among the intents. And further, we will visualize those conversations in the dashboard and be able to enable enterprises to see the dominant paths of those conversations and the flow of those conversations. And then you could also see trends and correlations within that dashboard. On top of that, we also offer full exports of that data, so you can actually tap into that -- what we call conversations as a data source. So you can take that data and embed it into your CRM systems and, for example, do a screen pop of call summaries, so agents have the latest information from previous calls with customers or you can leverage signals, such as churn indicators in your CRM or you can leverage things like Tableau or Snowflake, et cetera, or even use data science tools like Jupyter Notebooks and have your data science teams understand patterns and do further analysis. So again, all of this is done automatically using each enterprise's data. And also, we just launched our 8.0 release. We're very excited, and this includes -- this just happened this month, includes continuous intent discovery. So as your business changes, for example, if you introduce new products or new issues come up or there's changes in the market, it will actually discover new intents in that data, so you can always stay up to date. And this platform runs on NVIDIA for the fastest and most scalable platform out there. So we collected from our customers over 40 different use cases and each one of these can really deliver millions, if not tens of millions in savings to our customers. And so it falls into several different areas. One is around improving the customer experience overall. And in fact, trying to eliminate some of those -- the need for some of those contacts and reducing the call volume overall, things like improving first call resolution to eliminate repeat calls, do it -- being more proactive. Secondly, around agent performance, increased agent productivity, things like eliminating transfers, improving how calls are routed, improving handling time with these call summaries, improving compliance and best practices. And then digital self-service. So this data and this intelligence that you can see and visualize, you can actually improve your self-service tools, IVR, your chatbots. And in fact, using these exports of data, you can use those to train digital assistants and IVR. So -- and finally, growing revenue. So being able to protect revenue, looking at churn indicators is a big use as well. So that gives you a quick sense of lot of use cases and here's a customer example. And so within really a few months, I will -- of using Talkmap, this company was able to identify 20 actionable insights from using Talkmap. And so if you compare to what they were doing before and just to kind of go back to the problem, companies are spending months, as I mentioned before, analyzing data and then, again, looking at samples and coming up with ways to change the operation based on that and then spending many more months in order to implement and be able to monitor how those changes went. And so with Talkmap, you can implement that and within months, this large telecom company was able to identify a bunch of insights for tens of millions of dollars in savings. And they expanded from 1 call center within 90 days to 5 call centers. So things moved very quickly once they saw the benefits. And now they're targeting another $100 million as they continue to roll out the platform, addressing KPIs like FCR, First Call Resolution, NPS, reducing call volume and improving compliance. And so on the left side, this shows how quickly they rolled things out once they saw the benefits from support to collections to sales and service and loyalty. And on the right, just a sample of some of the use cases they're addressing. I mentioned things like call routing to improve and reduce call transfers, repeat calls, really, that cuts across all of our customers, is a very big use case. And then self-service is big, compliance and things like improving troubleshooting. So these are just some of the ways that customers are using the platform. And again, being able to save tens of millions, if not hundreds of millions over time. So with that, I'm going to show you a quick demo. And so you can see firsthand how this works. [Presentation]

So you can see that, that you can actually then visualize conversations at a high level and this is all done automatically. You can then even drill down on specific conversations to sort of test hypotheses. So happy to answer questions later and happy if you want to get in touch, to provide a live demo. So thank you. I'm going to turn it over to Dhanya now.

And I really like, Ken, how you're really, with Talkmap showing data mining of conversation, right, is really a powerful tool for the enterprises and for telcos to really understand their customer base and get value for their businesses. Great use case. Dhanya?

Let me know when you can see my screen.

We can see it and hear.

Great. Great. So before I get started with the topic of discussion, Lilac and folks, I would like to take a moment to quickly introduce Quantiphi. We are an AI-first digital engineering company. We started in 2013 when AI for machine learning made a shift from being an academic discipline to a more practical mainstream applied science. We've grown over the last 10 years and now are an organization of 4,000 plus AI, ML and data science practitioners. We cater to over 70-plus Fortune 500 companies. We are headquartered in Boston. We operate across the globe and have 2 large delivery centers in Mumbai and Bangalore. From a key competency perspective, we are heavily focused on applied AI solutions, including computer vision, digital avatars, twins, conversational AI, NLP, as well as data engineering, analytics and infrastructure solutions. From the industry's perspective, we work with -- we work across industries from telecom, health care, BFSI, et cetera and come with a lot of domain expertise. We have strategic partnerships with industry leaders like NVIDIA, Google and AWS. With NVIDIA, we're an elite partner and we've been awarded the Service Delivery Partner of the Year award for the last 2 consecutive years. We've also been awarded top partner awards with Google and AWS. With that, I would like to come back to our topic for today. And while I start, I would like to go back to the point that Anand had made during his presentation, right, about what essentially -- what or how do customers essentially look for in a service? How do -- what are the parameters that impact how they perceive a service, right? And the basic pointers or the key attributes would include being easy to use. So we need a hassle-free, a simple process to reach the agent. Is the process -- is the service empathetic? Does the agent understand the context of your problem and provides a response accordingly. Resolves issues quickly. Like customers today have very little patience, right and that really impacts. If there is any delay that kind of impacts the way they perceive the service. And finally, availability on demand, anytime, anywhere one should be able to get help from these. So keeping these parameters in mind, we at Quantiphi have developed our digital avatar solution that is powered by large language models. Our avatar is built using NVIDIA's Omniverse, Riva and NeMo tech stack. And how does our digital avatar bring or enhance customer engagement? So what are the key parameters? Primarily being able to portray the brand emotion and personality of the organization, right? So the avatar can be designed in such a way that it looks and feels like or is built as an embodiment of the brand itself, right? So looks, feels attire, everything can be designed accordingly. Human-like interactions. So not just in terms of looks but in the way it talks, in the way of mannerisms, gestures, expressions, tone and quality, all of those aspects can be brought in, which makes the conversation with the avatar almost human like, right? Symantec understanding -- this is where the power of LLMs come in. So the avatar is backed by large language models, which are able to basically take the request, understand the context behind it and provide a response accordingly. So there is a lot of training that is involved with large language models. There's a lot of data sets involved and that helps kind of build the model accordingly. Domain adaptability. Now every domain, every vertical -- industry comes with certain aspects that is very unique to it, right? Like for example, if we talk about telecom industry, there are a lot of jargons and keywords like, when you say, MBPS, it means network speed, right? But a normal -- I mean if it would have been a normal avatar, it would have just picked up as M-B-P-S as being different words, right? So that context, developing that kind of training is very easy with this model. Finally, continuous learning. So with every conversation, the model is learning and that enhances the overall experience for the end customers. Some other features of our solution include being modular design. So -- if I talk about the building blocks for the digital avatar, it would mean the overall personality and then the brains, which is nothing but the overall conversational flow. So each of these -- these blocks are modular in design and they can be run as separate microservices and easy to integrate. Our solution is also available in multiple languages. So you can use the same digital avatar and provide services in multiple languages. So this is, I think, very key, especially when it comes to customer support services for customers across the globe. And what does this mean for business, right? So it means higher customer engagement, where 90% of customers have rated their experience with virtual assistants as positive. Improved operation efficiencies. For one of the parameters, which is the average call handling time, we are seeing a reduction of 77% with digital avatars and finally, increased cost savings. So Gartner has said that there could be a reduction of labor costs by about $80 billion with conversational AI deployments by 2026. Now I'd like to quickly give a walk-through of how the entire conversation flow happens from the user placing his request to the avatar providing the response back to it, right? So here the user wants to know store locations near him and his location is Downtown Dallas. So this input is provided as a speech to the NVIDIA Riva ASR, which then converts it into a text format and feeds into the NLP unit of the -- NLP component of the NVIDIA Riva. This is where the intent identification and slot classification happens. In this case, the intent here is the store location and the slot identification is the Downtown Dallas, which is the key word here, right? This is then sent to the dialogue manager. The dialogue manager is nothing but a conversational bot which keeps track of the entire conversation flow in terms of what has been asked previously, et cetera, right? And from the dialogue manager, it goes into your information retrieval system, which is powered by large languages. So the retriever there looks for the corresponding response within the data -- different data sources available and provides a response back to the dialogue manager. This -- the dialogue manager then creates a proper response in the text format, which then goes back to the NVIDIA Riva's TTS module, which converts it into an audio file, which is played back to the avatar to get the animations, lip sync and everything in sync and finally goes back and is provided as a response to the user. So I'd like to now quickly play a demo for you. Here, the use case is that of a potential customer for a telecom company and he's trying to get different options of plans he can look for. [Presentation]

Okay. That's a great way to interact with your customers with avatars. Thank you, Dhanya. I appreciate it. So now let's go back to our agenda and talk about the future of [indiscernible]. So I'm going to have a question kind of to each one of you. And I wanted to get your opinion. So Dhanya, I'm actually going to start with you. You showed us in the flow, the use of automatic speech recognition and so forth. So where do you think the future enablement and the connection between conversational AI and generative AI?

Definitely. I think the main aspect of -- a conversation AI in fact plays a very important role when it comes to getting or enabling more from the generative AI, rightly. What goes in comes out, right? So if you do not have a good -- if the speech recognition is not of good quality, it's not able to pick up the right keywords, does not understand the jargons. I gave the example of MBPS as being a jargon, right? How does the ASR model actually [indiscernible]? How is it tweaked? Does it understand that concept because that helps in the response coming in from the large language models as well, right? Similarly in terms of noise quality, are the models able to get the noise, disable the noise coming in and get the content itself, right? So those are the aspects that conversational AI plays in terms of leveraging the large language models.

Makes sense. Perfect. I think what you're saying is, it's the ASR is being kind of like the foundation of the accuracy to train then the model or the generative AI. So we needed to be, things has to be right from the word go. So awesome. Anand, so there is the talk about generative AI writing. We're seeing the traction that it's getting in the market, where do you see generative AI even being used in contact centers?