This Week in Startups - The Future of AI Customer Support with LiveX AI’s Jia Li | AI Basics with Google Cloud

Episode Date: March 11, 2025

In this episode of AI Basics, Jason sits down with Jia Li from LiveX AI to explore how human-like AI agents are changing the game for businesses. Instead of clunky bots, these AI agents engage customers in real-time, helping with onboarding, support, and even stopping them from leaving before they churn. We’ll break down how they work, why they feel more natural than traditional chatbots, and how companies can train them to eliminate costly mistakes. Plus, we’ll talk about whether AI-powered support might actually be better than human agents.*Timestamps:(0:00) Introduction to startup and AI basics(1:00) Collaboration with Google and insights from industry experts(1:44) Introduction of Jia Li and LiveX AI's innovative customer retention(4:32) LiveX AI's AI agent demonstration and training methodologies(9:38) Personalizing user experience with LiveX AI*Uncover more valuable insights from AI leaders in Google Cloud's 'Future of AI: Perspectives for Startups' report. Discover what 23 AI industry leaders think about the future of AI—and how it impacts your business. Read their perspectives here: https://goo.gle/futureofai*Check out all of the Startup Basics episodes here: https://thisweekinstartups.com/basicsCheck out Google Cloud: https://cloud.google.com/Check out LiveX AI: https://www.livex.ai/*Follow Jia Li:LinkedIn: https://www.linkedin.com/in/lijiali/X: https://x.com/lijiali_vision*Follow Jason:X: https://twitter.com/JasonLinkedIn: https://www.linkedin.com/in/jasoncalacanis*Follow TWiST:Twitter: https://twitter.com/TWiStartupsYouTube: https://www.youtube.com/thisweekinInstagram: https://www.instagram.com/thisweekinstartupsTikTok: https://www.tiktok.com/@thisweekinstartupsSubstack: https://twistartups.substack.com

Hey, everybody, welcome back to startup basics. What startup basics you ask? Well, if you're a fan of the show and you've been watching it for a couple of years, you know that I get asked the same questions over and over again by startups. So we create a series called startup basics. This way, I can send somebody a link to this week in startups.com slash basics. And I can answer all their questions about legal, which we do with our friends at Wilson Sincini, accounting and cap table issues and all that kind of stuff, which overlaps. a little bit with legal, with my friends, Scott over at Cruise. And we started a new series on AI. Why? Because AI is just as important now for running a startup as legal and accountant. We're seeing amazing gains in our portfolio of over 400 startups that we've invested in over the years through using AI agents and all kinds of different AI tools, co-pilots, etc. So our friends over at Google collaborated with us on this new series, AI basics, for startups. And our friends over at Google created a report that's called The Future of AI, Perspectives for Startups. And it's got insights from 23 leading experts, including two of my besties, Shamath and David Friedberg.

Shamath talks about in this great report why a shrinking software market can fuel greater profits. And the Sultan of Science, he talks about regulation how AI is revolutionizing hardware, media, software, and biology. So get the report. We'll have at the end of the program, a URL for you to get it. And today, I'm excited to have Jolly. She's the president of LiveXAI. They're using adaptive AI agents for customer service, specifically customer retention. And, hey, they're a lighthouse customer of Google Cloud.

Google Cloud is an amazing product that we use here at my firm. Welcome to the program, Jolly. Thank you. Very nice to meet to everyone here. Yes, great to meet you as well. So let's talk a little bit about what you're building. Obviously, if you're listening to this week in startups, you know about AI agents, but you're doing it specifically for customer retention.

And I think we know what retention means you don't want to churn. Churn is the enemy of success in startup land. So how are you doing it? that exactly. So we believe human-like AI agent can create that empathy and trust between consumers and the AI agent. And through that process of having AI agent, guiding through the consumers, going through onboarding, going through product education, sometimes they need a little bit of customer support. And sometimes we also go into rescue moment. Somebody is about to turn and they are thinking about, oh, I didn't get much value of the product. So we can have a human-like

AI agent to guide them through the process. Hey, is it because some of the features that are not right for you or is it because you run into technical difficulties or is it because of pricing, et cetera, et cetera, right? So unlike the traditional approaches, it's almost like somebody is already standing on top of the cliff about to jump off the cliff. The traditional method, like pop-up window, is trying to save a customer after the jump off the, Yes. So the customer quits and you're like, why didn't you quit? And so this is, hey, the person goes to, I want to disable my account or they're going

unsubscribe. You have an agent start talking to them. Maybe you could show us how this works. Yeah. Happy to do that. We do that since the beginning of the journey. Somebody is already considering the brand or considering how to use a product and will be there. and also going through the process will be there, et cetera, et cetera.

So let me show you when the journey starts. Somebody is considering a sports equipment right now. Welcome. I'm here to help you find the perfect treadmill for your needs. Okay, so here's the agent in the bottom right. I'll sports cast it a little bit. You have a customer support rep who looks like a human, but this is obviously an AI, and it says I want a treadmill for intense daily work.

and the agent recommends this life smart power touch treadmill and why it's fantastic because it's got a great motor. And here you go. This is to engage your person at the purchase, yeah? Completely. Completely. Awesome. And so do you have any videos of it showing people quitting? I'm wondering?

Yes, I do. I do. Let's see the quit. I want to see the save. All right. Now let's see a save in action. All right. Somebody's canceling.

Hi, Jerry. I understand you want to cancel your subscription. I can help you with that. Could you give me some information on why? Okay, and the person can type whatever reason they like? I understand. Have they tried our new AI coding course? Just wondering because it has been very popular.

Minecraft, that's fun. Here's a short video on how to get started in Minecraft. Click this link to land there. All right. We're sorry you had issues getting started for your trouble. We would like to offer you a 20% discount. Would you like to try it out? Awesome.

Wow, this is great. I love it. This is what a person would do on the phone if you called and said, hey, I want to cancel. You say, oh, why are you cancel? And it doesn't work. I can't log in. And you're like, okay, well, I reset your password.

Would you like to try now? And would you still like to cancel? And then you do the really nice, gracious thing. Because sometimes people can get annoyed. They just want to cancel. But you're saying, hey, you went through this process with us. We'll give you a little nice discount here. Would you like to take advantage of that really well done?

Thank you. So as you can see, AI agent really tries to understand the needs of the user, right? So sometimes the cancellation is not because of pricing or not because of the users didn't enjoy the product. The kid grow up and no longer. needs the same product and there is the upselling moment, a recommendation moment. What truly the users appreciate is the AI agent understands their needs

and are thinking from their point of view how to guide them through using the product more efficiently and more happily. How long does it take to train that agent? Because everybody's software is different. In this case, it's a coding tool. I might have, I don't know, It could be like DoorDash and you're trying to get a return.

These are very different scenarios, very different responses. So how long does it take to train this agent to do it properly? And then how do you avoid hallucinations or mistakes, which, you know, if you're returning a product and you say something silly, you know, could be brand damaging. So I think sometimes people reasonably having used LLMs and seen hallucinations or errors on the margins, you know, they're like, I don't know if I want to put this with my customers. So how do you answer that question when you get the objection? Totally training time is scalable,

especially partnering with Google Cloud and Nvidia to make this much more flexible and smooth, efficient as well. The most challenging aspect, as I mentioned in the report, people often underestimate the power of data. So if you have the right kind of data and if you process data well, then training is much more straightforward. Right now, if you see many of the generic approaches, they're using data on the internet. Everybody is using the 10 trillion token, et cetera. So then the quality of the model, if everybody is using similar architecture,

and the training method. The depreciator is not as big, but data is really the core of the approaches. If I understand... So you can take people's like Zendesk or their FAQ or their website for customer support, you can just ingest that and have most of what you need, I guess? Actually, that's one approach to take

because traditionally those data have been useful for this purpose. But in the AI agent era, you know what's most important? It is the real world interaction from the users. What about using information on how I use the product? So if I was a DoorDash user, let's say,

and I had done, I don't know, 100 deliveries this year, and a different user had done zero, there's different contexts here. One is a very loyal customer, and let's say they tip all the time, and the other one, they're a first-time customer. Does this agent know my history with the product? Definitely, very important.

In order to make that personalized experience, as much as we can know about the users, either on the flight when they are expressing their interest or their needs, or some historical information on, hey, like what has been purchased, and what is the potential interest, that will power the AI agent to really simulate that human-like experience. Got it. When will these be indistinguishable from a human?

When will we basically cross the uncanny valley, as we say in AI, where a human and can't tell the difference? When will it pass the touring test? I felt that day will come soon. Right now, we already have, voice that is very similar to human. We have the reasoning that is very similar to human in the thought process and the digital human approach is getting better and better. And through the process, we actually realized the more it's human-like, the more potential possibilities that are using,

would tend to engage. No one wants to hear, hey, like press one, press two in the telephone engagement. I kind of like talking to AI agents now. You know, I was using Gemini. I was on the way to school with my 15-year-old, and she had read a book. So I said, hey, and they were going to review it that day, I said, hey, give us a summary of the characters and the main themes of the book, and Gemini's voice interaction is so good. You know, we were just talking to it.

And then I said, hey, and she's like, I can't remember. remember the character's name. And so I told it, hey, give us a quiz on the character names and some ways to remember them. And it went back and forth. And she loved doing it. It was like hiring a $100 an hour tutor. And we did it in a 15 minute car ride. It is amazing. Not only how fast this is all advancing, but I think as humans, how we're adapting to it and actually enjoying it, most people prefer to use the Starbucks app to order their drink because it's more efficient. there's less chance of a mistake. Most people, I think, even if we don't cross the uncanny valley,

I don't like talking to an AI agent because I can ask a question, get through it, and I don't feel like I'm burdening a human. You know, I kind of like it better. But, hey, we're getting close to crossing the uncanny valley. So that's amazing. Where can people learn more about your product if they want to try it out? What does it cost for a small company to use it? What's the entry price?

It depends on the volume of the interaction. So we are usage-based. If there are a small company, the cost will be fairly efficient, cost-efficient. Hundreds of dollars a year, thousands of dollars a year, something like that? Possibly, possibly.

Right now, we are mostly collaborating with successful companies with millions of users. So there needs to be much more, you know, customization and personalization. essential. Awesome. And where can people find out more?

Come to our website, livex. com. Okay. Everybody go to the website right now at livex. comaI.

L-I-V-E-X. Dot A-I. Jali, thank you so much for joining us. And thanks so much to our partner. Go to

G-O-O-G-L-E slash Future of AI for predictions, real-world examples and startup advice in that amazing. reporting Google Cloud's Future of AI.

Perspectives for Startups Report. Thanks again for listening. We will see you all next time on Startup Basics. Bye, bye.