Kevin Wu, Leaping AI: Founder & CEO at Leaping AI (YC W25) | Voice AI agents that self-improve

Welcome to another YC Wisdom podcast. We have Kevin Woo with us. He's a founder and CEO at Leaping AI, part of YC Winter 25 uh batch.

Leaping is basically voice AI agents uh that self-improve. So had a couple of these YC companies on my podcast. I'm eager to learn more about Kev, his background, and how he's doing things differently.

I see here has past experience. you've been an investment analyst, worked at uh Amazon as well. So, Kev, can you introduce yourself?

Tell us a bit more about what you're up to nowadays. Yeah, my name is Kevin. Uh I'm the CEO of Leaping AI.

Leaping makes self-improving voice agents for businesses. We currently do around 100,000 calls per day on our platform whilst automating, you know, up to 70% of repetitive uh support calls on the phone. Personally, I'm I'm sitting in San Francisco.

I grew up in uh Germany and ran this company in Germany for a year and a half before moving to SF uh at the beginning of this year. So, who's your ICP? I see GDPR compliant there and you're from Germany.

Do you attack the European market a lot nowadays or it's pretty international? So, we primarily serve the European and American markets. Our ICPS are, you know, um companies with large call centers that um have a lot of repetitive phone calls and that want to uh you know, use AI to improve first call deflection rates, right?

And what's their alternative or what's their current solution and why is yours better? Yeah. So the current solution is either doing it with humans or outsourcing it to a um human call center or they might have um some basic IVR solutions such as press one for this, press two for this and we're helping them to make these conversations more natural.

What's your stack? Uh what has it uh how how have you built this solution? Is it uh uh Chad's voice?

Do you have your own model? Like what's uh what's under the hood? Yeah.

So we are an orchestration layer. So uh we co co collaborate with uh large language model providers like uh chbpt. What about sesame?

Uh as of late have you tested that one? What are your thoughts about that? Yeah, we have tested it's it's really good.

We're um kind of evaluating bringing that into our stack as well. Currently we're using um the following providers um Cartisia, 11 Labs, OpenAI itself and also Azure. And how many clients do you currently have and what do you charge them?

We have around 10 to 15 clients and we uh charge them like around a five figure six figure fee depending on the size of the client. And why did you pick that niche specifically? Right?

You were accepted to YC. you had all kinds of choices as perh what product to create. Why did you identify that niche specifically as very interesting to you?

I mean I I started in 2023 um even before uh doing YC and so I I used to be a consultant at BCG and um towards the end of my time there I was part of the Genai task force and we're looking at different verticals and one of the things that all the verticals had in common is that using Genai to automate contact centers and to automate customer support has a big potential uh across industries. So I had this top down view. At the same time I was connected pretty serendipitously with one of our largest customers now who who had this exact painoint.

So there was a very concrete business opportunity as well. Right. I'm reading some of your posts on LinkedIn.

You screenshot Reddit post of someone that works in call centers. So I'm guessing you must have heard a lot and uh spoke a lot about the topic of AI and replacing some jobs. You mentioned in that specific poll that working in the call center is uh is probably hell, right?

Like I started my career making 200 cold cold calls a day, not even answering some, but like uh getting five to 10 offs every day. Uh is is it a good thing that we would be replacing these uh these jobs ultimately? So I think that um the call center uh the call center work itself uh most of the times you know is pretty menial and pretty repetitive.

So I actually uh spent uh time living in South America and many people I met there uh spent time working in a call center uh to to earn money while they were students and most of them just absolutely hated it. So I think that um if we're able to uh you know have this done by AI in a similar quality, it would free a lot of businesses to take those same um employees that they have that do call center calls and move them to more productive revenue generating functions. Agree with that.

And yeah, let's talk also about the consumer side. Like you have a problem with your bank, you call your bank, you've been put on hold for like 15 minutes, right? And then you speak to an employee that it goes to very robotic type of processes.

Some of them are more competent. Some some others are just not competent. And that's really frustrating as a customer, right?

So what's the the customer experience like and how can an AI actually serve the customer better and get them to the results faster? Yeah. So I think there are uh three um steps in the call, right?

So basically AI can help in all of these three conversation steps and make it more seamless. So the first step is you have to wait until you speak to someone. Then the second part is um you tell the AI or sorry you tell the company what you're calling in for.

And then the third um part is actually the the person helping you. So right now um you know the first part is completely eliminated with AI because it it can you know answer the call straight away. The second part right now is typically handled by an IVR system where there's a lot of music and then you have to listen to some instructions and then press like a button like press one, press two, great experience, right?

I called to listen to some amazing music. Yeah, exactly. An AI can make this part even more natural.

And then the third part is um you know where you you actually speak to a human to get your problem resolved. Even then I'm sure you and your audience have had issues where uh like this person could not help you and lets you wait in the line for another five minutes right so actually AI with the full context and the full access to data is able to help in a lot of these use cases obviously there are more difficult uh parts of the conversation as well so AI might try to you know resolve this issue but if it cannot it also falls back to a um to an employee that that can to a human employee that can help out, right? Yeah.

I think there's it's just better at pretty much every single step. I think the main bottleneck for you is company adoption itself because the solutions is already there. It already exists, right?

So, how do you actually score new customers? Sometimes you're outreaching to these companies and you're you're like trying to convince an exac that is scared of AI that is like, "Oh that's going to replace me ultimately." Right? So how do you do your G GTM and how do you adapt your message so that big enterprises can actually um invite that in their company and not resist to AI adoption.

Yeah. So I would say uh regarding um you know attracting customers we have two channels one is outbound one is inbound. We surprisingly get a lot of inbound requests and so these customers find us on LinkedIn.

Uh they find us on on podcast they they know us from YC and then we have an outbound GTM motion where we strategically reach out to customers that are interesting to us. When we speak to new customers obviously we try to understand their like pain points what they're optimizing for and that might be different for different customers. Some customers might want to reduce costs.

Some customers might um want to actually have 247 availability because they're losing revenue. So we try to understand, you know, what they're optimizing for. What I started doing in the beginning, what I'm not doing anymore like as much now.

Uh and it it's also um going to your point of generating trust is I I try not to set unrealistic expectations, right? So uh I I don't say hey you know with this you're able to automate um like your entire human workforce right and save massive costs because at the end of the day right now AI is not there yet to be able to you know do 100% of the calls by itself right so if it's just a cost reduction play and this um company is thinking about reducing 80% of the cost right to me after having been in this business for two years with the current state of technologies this is not realistic right so I try to already from the beginning, you know, align on a goal with uh the client that I'm I'm confident AI can actually achieve, right? Um and so later on when we actually review the results um that hopefully does not lead to any disappointment.

And do you do cold email outreach? Like what are the titles that you're going after? And yet like what's the biggest pain point that has gotten you into the most stores as of now?

Yeah. So I'm I'm uh reaching out you know to VP of ops um people that think strategically about call centers and what has worked is you know if we talk about um companies having you know large call center volumes and us helping them to resolve many of these repetitive requests. I would say email is is not working that well as a channel for us because I think it it's also getting more and more crowded.

Yeah, for sure. Depends on on who you you deal with, right? Like I mean I do GTM for YC companies uh YCGTM.com and it works really well.

I think the infrastructure in email and getting in the actual mailbox is the real challenge nowadays probably 90% of the equation and then once you send a very thoughtful AI written email to the the right person, I think you get a a nice response rate. But yeah, in your case maybe like calls would be relevant, right? like you're calling product.

Maybe if you do outbound, but outbound's a bit uh it's a gray zone, right? Like on the the legal level, you cannot really do outbound. We do it to outbound, but we funnily enough, we do it with humans because um so our sweet spot is contacting people with AI, you know, if they have given previous consent.

And so obviously if I'm reaching out cold to VP of ops, he he probably has not given me consent. So I'm actually funny enough using humans to do the the cold calling out feature. Yeah.

And I don't think an AI would be like VP of ops level on the outbound. I think they're a bit clunky, you know, and still a great SDR probably from the the Bay Area would be better to to execute these calls. And yeah, like these VP of ops, uh, where where do you actually get them?

Is it like, oh, you're going to save that amount? You're going to increase that amount in customer satisfaction? And what does a pilot look like?

Like how can you get them to commit a little bit but not too much if in case it fails, right? Yeah. So pilot we we sometimes we do it for free and sometimes the the customer commits something up front also part of like an implementation phase uh especially if they want to connect the voice AI to some of their internal systems.

So we try to just find an arrangement that works for time being. We try to, you know, have a clear exit point for the pilot and then evaluate the results and um hopefully continue from there if the pilot's successful. Right.

So this year, what are your top goals? Is it to raise capital? Is it GTM?

Is it product? Like what are your specific goals? What do you want to achieve this year?

Yeah. So uh I would say both. Like you always have to build product and you have to sell it.

So uh on the product side we actually uh scaled a lot of um uh like our calling volume. We have more customers. So just to make the product more stable is something that we're working on uh every day.

There are a lot of uh pains on on you know increasing the volume from let's say 10,000 calls a day to 100,000 calls a day. So making the the product more stable is important to us. The second thing is making the product more user friendly.

So uh we're not like a self-s served product but still the experience does not have to be uh bad right even for some of our business customers to kind of use the the platform. A lot of our customers actually um you know they they create their first agent with our help and then they use our tools right um no code tools to create more AI agents for more use cases right um by themselves without us even knowing right so they might start with let's say automating customer support hotline but then they might create an AI agent that calls uh customers um you know that has signed up on the website but haven't really converted to being a customer right they run different campaigns to uh call customers and remind them to do something which is especially relevant in healthcare where you remind people uh to conduct certain actions right like uh take their medications for example. Uh so how we see our platform is is kind of giving these companies the tools and and nontechnical people at these companies the tools to kind of create their own voice agents.

Right. And what does the team look like right now? Like you're on the raising side of things and GTM side of things.

your founder is he the technical one. What does the team look like? So currently eight people uh I would say 70% of the eight people uh are on the technical side and 30% are on the business side.

So um you know we we try to keep a small team uh also because software engineers here in the Bay Area are pretty expensive. But also I feel like um we still have a lot of homework to do uh to set the foundation before we can even think about you know scaling further right and do you were uh or your startup was was born born in YCD winter 25 seems you scaled very fast you know like how have you seen that compared to your other peers like you're already with nine employees most of them are still like at seed stage sort of and low growth yeah I I'm not sure if the others are low growth. Um but uh I mean definitely the voice AI space is is pretty hot right now.

So I think it's just a general um pull uh that that we and others in the voice market feel from the market whereas some other YC startups might uh work in verticals or or in markets that that are not that don't even exist yet, right? And that have to be created. And you said a lot of traffic came from inbound.

Like what does your SEO look like? helped you attract that inbound. So, a lot of it is like founder content marketing from myself.

Uh, funnily enough, um, we did like a podcast uh during YC Demo Day with TBBN. I'm not sure if you've heard of that. And, uh, actually a lead I spoke to yesterday came came from that because they actually watched this podcast, right?

So, it's actually pretty uh, serendipitous and amazing. Um, I would say SEO is like something very important that we're starting to invest in now, but we haven't done so so far, right? Like TBPN is great.

You probably won't get the same view here. Uh, yeah, I believe the founders even a YC founder, right? Like the Soilent guy, John Kugan.

Forgot his name. Um, but yeah, it's it's a great podcast. I love what they're doing.

That guy's uh pretty next level. You also met um Paul Graham, I believe, a couple months ago. How was uh how was that meeting the startup gone?

Yeah, it was amazing and he even um you know uh agreed to invest in our startup. So, uh, yeah, we pulled up to Palo Alto, um, to his house actually and, um, kind of were invited, uh, to speak to him and I think the meeting lasted 45 minutes and he, uh, took our idea and, um, kind of put his own spin on it and we were brainstorming a lot and I really appreciate kind of the session and and getting his thoughts on on the space, right? because he also has a property I believe in the UK, right?

So he's somewhat bullish on Europe as well. Did that help unlocking some uh some paycheck there? Uh maybe I don't think so.

Uh actually we uh were making fun of Europe in the bit because we said hey um you know there's a lot of data compliance and regulation which we take seriously in Europe which we don't have to do for the US and so sometimes the AI is more effective if it can use more data and it's not bound by data regulation right uh so this is one area where um AI can actually be more effective in the US than in Europe because it's not bound by the same regul ations that even we as a company take very serious in Europe. And do you think that OpenAI could eventually launch in your category or are they focused on building the the infrastructure? So I don't think so.

I think they're building the infrastructure and we're using some of their models, but we're creating a lot of like the the tools and the platforms for enterprises to make AI accessible. And so that is not something that OpenAI is probably interested in the next two, three or five years because they have operator, right? And I wouldn't be surprised if you would have like your personal assistant answering every phone call, maybe doing a couple of outbounds eventually.

So sort of never know with OpenAI when they launched chat GPT, they'd like slashed a couple of billions if not trillions in value from other players. Uh like for example, I don't even read books nowadays. I just uh go very deep with chd and that's my my books.

uh they replaced that a whole bunch of other softwares right so you never really know what are your AI predictions um especially on the agentic side of things this year do you believe uh open AAI will improve operator and the market will will be uh will be shaken by that Q3 Q4 so I would actually um disagree a little bit with the premise so if we think of in terms of stack right uh openi provides kind of the LLM but they're not going to be the one talking to the enterprises, right, to build like a voice AI deployment because there's so much tooling involved. There's so much like back and forth project management even that OpenAI probably can and does not want to do. And so this is where someone like us comes in, right?

We're using improvements in the LLMs to kind of make our models and our product better, but then we are the ones talking to these enterprises like jumping on weekly calls with them like listening on the ground to whatever they they want to like to do and like helping them implement it. And so this is not something that like OpenAI is doing. So in terms of like improvements uh on the LLM side, I'm actually like a little bit bearish because um you know I you know I watched like when CHAVT got released and now it's almost two and a half years later and obviously models have gotten a lot better but they have gotten better in a in a slower pace than I expected to be honest and AI is still definitely needs to be controlled does not do things perfectly.

So I do not believe that actually like in the next 6 months we will suddenly reach AGI but I think that it will uh come slower and and and more gradually which is why it's important to just to just stay in the game right for as long as possible right and that would be a net positive for most uh people that are already building in AI could also um sort of break things in the process but I think founders uh like you have learned to move incredibly quick and adapt but what about the the rest of the market there, people that want to adapt to technology. Let's think of these call center people. What would happen to them once uh an AI replace them?

Do you think they would actually evolve or go get another low-level job or would they actually step their game up, be more entrepreneurial, and adapt to this AI revolution? Ideally, they partner with us, right? So we already work with several BPOS which are called business process outsourcing or like outsource call centers that employ humans that do calls for other companies and some of them are you know already feeling pressure from their clients to uh kind of show what is their AI strategy and they partner with vendors like us right to kind of give a holistic solution to their clients especially because um AI is not good enough yet to handle 100% of all the calls and so you actually uh you know need AI working together with humans to deliver a 100% solution to the problem.

Right? It's like the scale AI model in which humans trained AI models. But the question is like for how long, right?

Will we need humans? Are they competent enough to keep up with an AI like at a low level call center uh type of scenario? Yeah, exactly.

Mhm. I don't think they are. I don't think they'll train AI for long.

I think AIS will train. The question could you repeat? Yeah.

Um you know scale AI Alex Wang um hires a bunch of humans to train AI somewhat smart humans most of them like out of uh college or uni uh for specific prompts and reviewing the work of AI. My question is more a low-level human that is replaced by uh an AI in a call center. What's what's their future?

you answer to that that they would sort of train help train an AI, but is that really true? Do they really have like that intellectual power to train a an AI and to learn? I'm I'm just uncertain of their future.

You can picture like these call centers in Philippines for example, which I've been to and I I ran a call center with 200 employees about five years ago or seven years ago. And yeah, I cannot see these employees even training an AI, you know, like I I don't think they they have the skill set shortterm and we know that AI is uh is growing exponentially. I'm just uh very uncertain about these people's future.

Do you think they'll adapt or I disagree with this? So first of all, we want to provide the tooling on our platform to make it easy for uh these companies to use it. Right.

So right now we have the voice agents. We're going to release an eval feature where you don't actually have humans looking at AI calls, but you have AI looking at evaluating all the AI calls. The next step is going to be the optimization module where you have an AI that learns from the AI graded AI calls, right?

And you actually improve the first AI. Uh so basically it's like um it's like a human call center worker where uh I mean they do the calls and then there's like a call center QA person and there's like a call center supervisor. We use the insights from the QA person to modify the script to train people and so we want to provide this on our platform as well.

Our goal is to make it really easy for these companies and these um employees to actually use like tell AI what to do. So, um, we have a term called voice manager where, um, we designate people at our clients that know how to use our tools as like voice managers. And some of them were former call center employees, right?

So, it it's much more fun to tell an AI what to do using our tools than to like actually do the work yourself, right? Um, for a business owner, you still need to pay for that person. My point is, I don't think these people will be around.

You can AI can train an AI better than a human can. It's only probably one to 10 percent of the these call center managers that are smart enough to to do that job. But yeah, like may they adapt, you know, maybe they be more entrepreneurial.

I think I think again it's not going to be an abrupt process because uh even though there's a lot of hype being on the ground, I know that AI is not good enough yet to completely replace humans or something like that. So everything is going to be the process going to be a process and the question is do you as like a call center owner you know start the process learn about AI start to use it are open to experimentation right and have the right goal setting and and and see AI in the correct way right no it won't be gradual u nothing is right like there's regulations there's ethics there's companies not wanting to to have their reputation destroyed by destroying entire industries right including you right now what that which is why you're answering that way. Um, and most uh people that you reach out to these VP ops, for example, they don't want someone that comes in and says, "Hey, I'm going to replace your people." Because, yeah, they're not going to be down with that.

So, yeah, it's a it's going to be a gradual process. I think it's a better experience, like you said, for employees, for end customers, and also for scale for the call centers to take that much calls and resolves problem um easily and quickly. So I very much like that bullish on your company.

Kev, where can people find out more about you? Yes, definitely visit us at uh https www. leapingai.com.

L E A P I N G AI.com. So that was Kevin.