Skip to content
YC Root AccessYC Root Access

This Startup Built AI That 80% of Callers Think Is Human

In this episode of Founder Firesides, YC General Partner Nicolas Dessaigne talks to Will Bodewes, founder and CEO of Phonely (S24), which just raised a $16M Series A led by Base10 Partners. Phonely is an AI-powered voice platform that answers phone calls for businesses — handling millions of calls a month across hundreds of verticals using custom LLMs that statistically optimize outcomes over time. They discuss how watching his dad struggle to answer phones sparked the idea, going from small business customers to enterprise call centers, and why by the end of this year most callers won't know they're talking to AI. 0:00 Intro & Series A announcement 0:18 What Phonely does (Voice AI platform) 0:43 Scale: Millions of calls per month 0:56 Customers: Call centers, insurance, home services 1:29 Voice AI boom & differentiation 1:55 Founder background & early PhD work 2:44 Optimization advantage (data-driven improvements) 3:12 Founder journey: athlete → AI PhD → startup 4:41 Origin story: dad’s business problem 5:06 Early product → iteration loop 5:49 Pivot to call centers (enterprise focus) 6:08 Building custom models vs using OpenAI 7:05 Architecture: modular models for latency & cost 7:54 Is latency still a bottleneck? 8:23 How good is AI on phone calls today? 9:01 Should AI disclose itself? (ethics + regulation) 10:10 Real-world use cases (revenue-driven inbound calls) 10:47 Inbound vs outbound strategy 11:13 Series A story 12:28 What’s next for Voice AI 12:59 Competition vs generic models 13:23 Future vision (50M+ calls/month) 13:49 Hiring (sales + engineering) 14:21 Founder lesson: reality vs perception 15:26 Advice for founders (who should start companies)

Nicolas DessaignehostWill Bodewesguest
Apr 16, 202616mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 0:18

    Intro & Series A announcement

    1. ND

      [on-hold music] Today, I'm joined by Will Bodewes, the founder and CEO of Phonely. They just announced a great 16M Series A led by Base10. Congrats, Will, and thanks for joining us today.

    2. WB

      Yeah, thanks so much. Very excited

  2. 0:18 – 0:43

    What Phonely does (Voice AI platform)

    1. WB

      to be here.

    2. ND

      All right, let's start by just the up. Can you tell us what Phonely is today?

    3. WB

      Yeah, so Phonely is a platform that allows you to answer your phone with AI.

    4. ND

      Mm-hmm.

    5. WB

      But that's, like, the tip of the iceberg. So really, what we do is we allow businesses to optimize their voice AI agents so that they can continuously get better on the outcomes that the customers care about.

    6. ND

      And, uh, what kind of volume are we, are we speaking about here? How many calls per day or per month?

  3. 0:43 – 0:56

    Scale: Millions of calls per month

    1. WB

      Yeah, so we, we do millions of calls every single month. Um, we do, you know, on a variety of different verticals.

    2. ND

      Mm-hmm.

    3. WB

      I think one thing that's exciting about our company is it's not just one industry. We've been able to figure it out across, you know, hundreds of different verticals.

  4. 0:56 – 1:29

    Customers: Call centers, insurance, home services

    1. ND

      And so who are your customers?

    2. WB

      Our biggest customers are people in the call center space, in maybe insurance. Um, we, we work a lot in home services. Um, but they run some sort of call center where they really care about capturing and optimizing AI agents for performance. So a lot of these customers, what they were doing is they had a call center that was qualifying leads, booking in appointments for very large organizations, and what we allow them to do is we allow them to make sure that the AI agent does a great job every single time, but also gets better statistically over time.

  5. 1:29 – 1:55

    Voice AI boom & differentiation

    1. ND

      So voice AI has kind of taken off in the past couple of years, right? And there ... We see so many companies doing that. Uh, how do you differentiate from all of these companies leveraging AI in their kind of like phone calls?

    2. WB

      Yeah, it's funny. We- when we started working on Phonely, voice AI, like the term voice AI, didn't exist.

    3. ND

      Yeah.

    4. WB

      Um, and so I think part of the way that we differentiate is just, like, having a ton of battle scars about how to get voice AI agents, you know, up and running in

  6. 1:55 – 2:44

    Founder background & early PhD work

    1. WB

      production.

    2. ND

      Sorry, remind me when, when did you, uh, start the company?

    3. WB

      Well, we started the company maybe, like, two years ago, like, officially, but we were working on this in our PhDs before that. So we were doing a lot of this experimentation, um, back in the early days. And so one of the things that, like, puts- sets Phonely apart is, um, you know, not just, not just the, you know, conversational quality of the models, but we started building our own LLMs. Um, and what that has allowed us to do is, you know, we care about optimization. Our customers care about how well the voice agent performs, not that it just answers the phone, right? And so we have the full comprehensive platform that lets you answer your phone, but we, we surface all the tools and all the data that you need to know to make informed decisions. And so for our customers, it's like we can statistically show them that changing one question, and we just did this the other day, changing one question can, uh, increase one of our customers' outcomes by 5%,

  7. 2:44 – 3:12

    Optimization advantage (data-driven improvements)

    1. WB

      right? And so-

    2. ND

      And you can tell them what question to change.

    3. WB

      Exactly. And so we surface through all that data. We look through all that because I think the exciting thing about voice AI is it's brand new, and nobody knows what to do about it. But it kind of feels like we just figured out you could put books on the internet, right? And nobody's realized that you can optimize that checkout, you know, that checkout conversion for, for an outcome, right? And so what we try to do is we try to take all that data and then service it back to the customers so that they can make great decisions on how to improve

  8. 3:12 – 4:41

    Founder journey: athlete → AI PhD → startup

    1. WB

      their business.

    2. ND

      Let's, uh, let's speak about what led you to that company. Uh, so because you have, like, a pretty unusual background here, like you started with an audio startup, then that AI PhD in Melbourne.

    3. WB

      Yeah.

    4. ND

      But you are American.

    5. WB

      Yeah.

    6. ND

      Like, how ... Come on. Like, how, how did [laughs] ... Why Australia? How did you come here?

    7. WB

      Yeah. It doesn't, uh, it doesn't really make a lot of sense, it looking ... You know, usually things make sense looking backwards, but I don't think that my, my journey necessarily does. Um, I was, uh ... Started way back, I was a college athlete. And so, um, I was a college athlete, and I was in engineering. And so I, I was-

    8. ND

      What kind of sport?

    9. WB

      Uh, cross-country skiing, funny enough, um, which is a real thing, and there's really college athletes out there that do that. Um, but I, I got pretty good. I was in, you know, NCAAs, and that was my whole life. That was my whole, you know, focus. And I got to the end of my, end of my career. I finally made it to the NCAAs, which is the highest that any-

    10. ND

      Mm

    11. WB

      ... collegiate athlete can make it. And my, my NCAA race was canceled because of COVID, and so I never got to, I never got to race it.

    12. ND

      Not good.

    13. WB

      Yeah. So I, I think part of it always had, I had this chip on my shoulder-

    14. ND

      [laughs]

    15. WB

      ... you know, that I had to go, that I had to go prove myself. And so I left college. I started my first company.

    16. ND

      Mm.

    17. WB

      Um, didn't end up working out. Um, but then I had, I was offered, like, a, a full-ride PhD, um, in artificial intelligence in Australia, and I took that opportunity. And I just learned as much as I could about AI, um, about the space because I felt like it was, it was gonna be the next big thing, um, and started working on voice.

    18. ND

      Yeah.

    19. WB

      And that's how we got here.

    20. ND

      And I think you said somewhere that you saw your dad struggling with answering the phone. Was that the trigger of that idea?

  9. 4:41 – 5:06

    Origin story: dad’s business problem

    1. WB

      Yeah. So, and my dad ran a, a small practice, um, as my whole life was just watching him-

    2. ND

      Mm

    3. WB

      ... you know, start this practice from our garage and, and build it up to, um, a decent sized business. And so I, I asked him what are his big problems that AI could potentially solve, and he was like, "Man, if somebody could answer my phones, that would be amazing." And so I was like, "Okay, well, let's look for a software that does this for you," um, so that he could use it, and nothing existed at the time, and so I started working on it.

  10. 5:06 – 5:49

    Early product → iteration loop

    1. ND

      Okay. And so from this first iteration of, like, an AI receptionist-

    2. WB

      Yeah

    3. ND

      ... just answering the phone to what Phonely is today, like, there's a huge gap.

    4. WB

      Yeah.

    5. ND

      Can you lead us, like, through kind of these iterations you had to make over the, the couple of years you've been working on it?

    6. WB

      Yeah, absolutely. It, it's ... You're, you're right. It started as, like, a very small business type of product, and I think the, the reason for that is because we needed the feedback from the customers about what they cared about, right? So we started in that small market where we could get customers on for, you know, 30, 50, 100 bucks a month, um, but they would give us really good feedback.

    7. ND

      Okay.

    8. WB

      And so what that allowed us to do is it allowed us to build a really good product and shorten our iteration cycle because we didn't have the luxury of having a ton of contacts into, like, call centers, into these big businesses.

    9. ND

      So how long

  11. 5:49 – 6:08

    Pivot to call centers (enterprise focus)

    1. ND

      did you do that before getting your first, uh, call center on board?

    2. WB

      Um, we, we did that for maybe four or five months or so.

    3. ND

      Okay.

    4. WB

      And then we got our first call center, and we realized that that one call center was paying us more than all of our small businesses combined. [laughs] And so we said, "Well, we should probably focus on this a little bit more." Um, and we, and we kept going on market.

  12. 6:08 – 7:05

    Building custom models vs using OpenAI

    1. ND

      Let's go back to the product. You also, uh, ended up replacing closed models. I think you were using some-

    2. WB

      Yeah

    3. ND

      ... off-the-shelf models before, right?

    4. WB

      Yep.

    5. ND

      By your own custom open source models. Like, can you tell us more about, like, why you did that change?

    6. WB

      Yeah.

    7. ND

      And what are the benefits?

    8. WB

      Yeah. There's a couple reasons. It's not standard, I will say, in, like, any other voice AI-... products that we've seen out there. Most people are just using, um, you know, OpenAI or, or other models under the hood. But we started working with Groq in the early days.

    9. ND

      Okay.

    10. WB

      Um, so they're the fast inference chip provider.

    11. ND

      Mostly for latency?

    12. WB

      For latency, exactly, because latency is one of the biggest problems that these voice AI models face. Um, and so we started doing that, and then we started building out, um, our own, you know, framework for being able to test and, and, you know, iterate. And, and based on our architecture, it just made more sense to, to have smaller models doing different tasks rather than one big model doing everything. And so we ended up breaking down smaller models, um, running them on fast inference hardware so you could reduce latency, um, and still get the same kind of quality.

  13. 7:05 – 7:54

    Architecture: modular models for latency & cost

    1. ND

      So how do you switch from model to model? It, it, does it depend on the question of, like, the, the, the time in the call? Like, how does that work?

    2. WB

      Not, not as much, um, around the question. It's, it's more of just around, like, the different components that are needed in voice AI, right? Um, so in voice AI you have, you know, maybe f- like storing a variable of a customer's name or email.

    3. ND

      Mm-hmm.

    4. WB

      Right? So that doesn't have to be the same model. That can be a different model.

    5. ND

      Okay.

    6. WB

      And so if you, if you break it up that way, um, you can have... You can save cost, for one, reduce latency, um, but still get the same quality off of that.

    7. ND

      So, like, when the people are going to share their name, you have a, a specialized model who knows how to recognize a name or something?

    8. WB

      Um, not that specific usually. Um, but it's more of, like, you know, you have some model that's meant for storing variables that has the context around these things. And what it allows us to do is it allows us to kind of, like, isolate and update those areas.

  14. 7:54 – 8:23

    Is latency still a bottleneck?

    1. ND

      We're speaking of latency. Is that still a bottleneck today in the, in the s- space of, uh, the field of OCI?

    2. WB

      It doesn't feel like a bottleneck. Um, it's definitely something that's important-

    3. ND

      Mm

    4. WB

      ... right? But it, it's one of those things where, you know, it's, it's kind of like oxygen at this point, right? You-

    5. ND

      It's good enough now.

    6. WB

      Yeah, it's, it's good enough now. Um, most, most people have pretty good latency. I think that there's still some rooms for improvement. I think more of it is on conversational quality and then accuracy.

    7. ND

      Mm.

    8. WB

      Like, the combination of those two things, um, I still think

  15. 8:23 – 9:01

    How good is AI on phone calls today?

    1. WB

      are probably-

    2. ND

      So how good is, uh, is AI now at answering the phone? Like, do you still need to hand off to a human?

    3. WB

      It's pretty good. For, I would say, about 80% of our customers, they have no idea they're speaking with an AI agent. Um, and, and I think that it's just, by the end of this year, I would say it's probably gonna be close to 100% of people won't know.

    4. ND

      You mean, like, 100% of the people who are speaking to on the phone-

    5. WB

      Yeah

    6. ND

      ... won't realize they're speaking to an AI?

    7. WB

      Yeah, they won't realize. They won't, they won't know.

    8. ND

      And how often are they going to speak to an AI versus a human? [laughs]

    9. WB

      At the end of this year, it's really hard to say. I mean, with the rate it's going, it might be a lot more than you'd think.

    10. ND

      Do you think you should disclose or your customers should disclose that it's an AI, or it doesn't really matter anymore?

  16. 9:01 – 10:10

    Should AI disclose itself? (ethics + regulation)

    1. WB

      I think for outbound calling, yes. Um, I, I feel like there is gonna be some regulations around that's gonna develop. I, I, I genuinely feel like the way that the world should be is you should disclose that this is an AI. Um, but I think that what's gonna happen is people are gonna start to realize when they disassociate the AI with press one for this, press two for that, and realize that it can actually be a really amazing conversation. And if you have the right tools, you can, you can make that conversation even better every single time, right? And I think that when people start to realize that on the consumer side, they're gonna get really excited to talk to an AI.

    2. ND

      Yeah. I mean, uh, even today, I prefer to speak to an AI-

    3. WB

      Mm. Yeah

    4. ND

      ... than possibly a human somewhere in the world that doesn't have any context about-

    5. WB

      Right, right

    6. ND

      ... who I am.

    7. WB

      Right. And, and there's also this, like, feeling... You know, sometimes you're talking about somewhat sensitive things, like you're asking... You don't want to ask a dumb question-

    8. ND

      Yeah

    9. WB

      ... if you're asking about finance, right? Um, but an A- you know it's an AI. You're like, "Well, I don't really care if I ask it a dumb question."

    10. ND

      Right.

    11. WB

      "I just want the right information," you know? And I think that that's gonna happen too.

    12. ND

      So anyway, if you know that's an AI, you may actually share more-

    13. WB

      Yeah

    14. ND

      ... about your problem or...

    15. WB

      Yeah. You won't feel bad for, like, wasting their time if you're trying to figure out your booking information and pulling it up. You're like, "I'm not wasting somebody's time. I'm just getting my information."

    16. ND

      What a world we are going into.

    17. WB

      Yeah. It's exciting.

  17. 10:10 – 10:47

    Real-world use cases (revenue-driven inbound calls)

    1. ND

      Speaking of customers, like, any, uh, surprising use case you've seen around? Like, kind of like, I don't know.

    2. WB

      We work a lot in businesses trying to make more money.

    3. ND

      Yes.

    4. WB

      Like, that, that's w- where our adoption is right now, is it's not as much of, like, the customer support and customer service time. It's usually, like, businesses, they have people that are calling in, like every single billboard out there-

    5. ND

      Yes

    6. WB

      ... you know, that has a phone number on it. Like, those leads, um, need to be kind of, like, sifted through. And they really want to surface the good ones, and they want to make sure that they're handled really well. Um, so I think it's not, it's not as much of a surprising use case, but just as the sheer volume of, of the phone being so core to so many-

    7. ND

      Yes

    8. WB

      ... businesses' revenue I think was the most interesting thing that we've learned.

  18. 10:47 – 11:13

    Inbound vs outbound strategy

    1. ND

      Like, most of your use cases are mostly inbound or outbound?

    2. WB

      Mostly inbound. Yeah.

    3. ND

      So that's these billboards. They see your number.

    4. WB

      Yeah. Exactly.

    5. ND

      People call-

    6. WB

      Yeah

    7. ND

      ... and actually you answer.

    8. WB

      Exactly. Yeah.

    9. ND

      And then you hand off to a specific, uh, customer.

    10. WB

      Depends. Sometimes we will, sometimes we'll hand off if, if that's what the customer wants. If it's, like, a licensed... Like, insurance as an example, you can't have an AI.

    11. ND

      Mm-hmm.

    12. WB

      You know, they have to be a licensed agent. If it's something like scheduling an appointment, we'll just schedule the appointment right with the AI. So it really depends on what the customer

  19. 11:13 – 12:28

    Series A story

    1. WB

      needs.

    2. ND

      Let's talk about the news you are announcing today.

    3. WB

      Yeah.

    4. ND

      Uh, your Series A with Base10. So, uh, how did you connect with them?

    5. WB

      Yeah. [laughs] It was, it was a funny story. So I have a, a background. I was an ultra... So right after I left college, I, I started doing ultra-endurance athletics. Um, and I would do these races, like ultra-endurance cycling races. And so-

    6. ND

      So you had to switch sport and... [laughs]

    7. WB

      Yeah. I know. I, I couldn't, I couldn't stop, right? Um, but to give you some context, the shortest bike race that I've ever done is 300 miles.

    8. ND

      Okay.

    9. WB

      Um-

    10. ND

      More than the longest I ever... [laughs]

    11. WB

      Yeah. [laughs] And so I would, I would do these where I would just basically not sleep for as long as I could and ride as hard as I could-

    12. ND

      Yeah

    13. WB

      ... for no real good reason, um, other than to like-

    14. ND

      What's the connection with Base10? [laughs]

    15. WB

      It's coming. It's coming. [laughs] Other than to push myself. And I posted about this on LinkedIn. I'm just saying like, "Hey, this is what, you know, I, this taught me about, like, committing to something and being a founder." Um, and Caroline from Base10 reached out about that, and that sparked up a conversation. A few months later, we had a few chats, and, and they decided to, to preemptively offer us a Series A.

    16. ND

      So you didn't even discuss with many other investors?

    17. WB

      No. I mean, we, we had some-

    18. ND

      They were excited enough. They made an offer.

    19. WB

      Yeah. Yeah. And it, we, we, we had a couple people that we were chatting with, but it was... We really liked the people there, and I'm very much, like, I pick people very carefully, and, and I wanted to pick somebody that I wanted to... I would be okay with

  20. 12:28 – 12:59

    What’s next for Voice AI

    1. WB

      hiring.

    2. ND

      Awesome. Uh, looking ahead, what's next for Voice AI? Like, is there a lot-Of improvements to make still?

    3. WB

      Yeah. I, there's, there's still, there's still work to be done around, like, interruption handling, around, um, end-pointing detection. Um, transcription is a really big one as well, as it's now we're handling the edge cases. Like, most of the words are solved, but when you have the phone, you have really garbled audio.

    4. ND

      Do you fear that kind of like the generic models are going to become good enough that it's going to be more difficult for you to differentiate yourself?

  21. 12:59 – 13:23

    Competition vs generic models

    1. WB

      That's a good question. Um, it's not my biggest concern right now because I think with voice AI, there's so much telephony that's involved with it. There's so much, like, um... There's a way that it should be done-

    2. ND

      Mm-hmm

    3. WB

      ... and it shouldn't be done, and I think that as we continue to go deeper and deeper into this, um, into this space, it's like we are statistically the best, m- the best at this use case that you can be, and I think if we keep going in that direction, um, it's the right

  22. 13:23 – 13:49

    Future vision (50M+ calls/month)

    1. WB

      place to be.

    2. ND

      And so where will you be, like, Phonely, like, in a, in two years from now?

    3. WB

      Yeah. Good question. Um, hopefully, you know, hopefully answering a lot more calls for a lot more businesses.

    4. ND

      Okay. So from 1 million f- a month to how many?

    5. WB

      Uh, I mean, hopefully, like, 50 million plus. That's our goal.

    6. ND

      Okay. That's cool. I like ambition. Um, and I guess because you, like, you just raised money, so you probably are, like, hiring too.

    7. WB

      Yeah.

    8. ND

      Like, anyone, uh, any job desk you want to share with the

  23. 13:49 – 14:21

    Hiring (sales + engineering)

    1. ND

      audience?

    2. WB

      Yeah, absolutely. We're, we're hiring a lot of, uh, sales folks right now. Really good engineers should be interested in reaching out. I think our team is unique in that we all, you know, kind of come from a similar background of, like, very low ego, um, very focused on just solving a problem, working hard, um, and we're just a lot of fun. So if you're in sales, we're hiring growth right now, we're hiring engineering roles, um, kind of the full gambit. Definitely reach out, um, we'd love to-

    3. ND

      All in SF?

    4. WB

      Yeah, yeah, all in SF.

    5. ND

      Okay. Um, before to conclude, uh, is there any one thing, uh, you wish you knew when you started

  24. 14:21 – 15:26

    Founder lesson: reality vs perception

    1. ND

      Phonely?

    2. WB

      I think the one thing that I maybe underestimated or didn't think when I started the company was how long it would take, um, and what success looks like on the outside versus what it feels like on the inside. Um, because I think a lot of founders, you... At least for me, I would like, you know, read the LinkedIn-

    3. ND

      Mm-hmm

    4. WB

      ... post and, and see the Twitter pages, um, and, and just be like, "Wow, that's, you know, that's so amazing," and they must have just got there right overnight. But I think the thing that I didn't realize was, you know, even at the top, the, the very, very tip top, you still have a lot of, like... You still fight. It's still a battle every single day. Um, a- and every day you gotta wake up and you gotta prove yourself, and you gotta say, "Oh, is there a new model coming out? Is, is our competitors doing this?" You know, it's always a fight, it's always a battle. So I didn't, I didn't realize that.

    5. ND

      Anything that would have been, uh, made, uh, your journey easier if you had known? [laughs]

    6. WB

      [laughs] It probably wouldn't have made it, uh, any easier.

    7. ND

      Yeah.

    8. WB

      But I think sometimes just setting the expectation of knowing you're going into... Like, you're going into war, right? And you're gonna have to fight, and you're gonna have to keep fighting every single day.

    9. ND

      And we're not going to stop for a while.

    10. WB

      Yeah, and you're not gonna stop for a while. But, um-

    11. ND

      Any, any advice you'd like to, uh, to give to founders based on that journey?

  25. 15:26 – 16:15

    Advice for founders (who should start companies)

    1. WB

      Yeah. There's a lot of people out there who, who think that they wanna be founders, and then there's, uh, people out there who have no choice but to be founders.

    2. ND

      Mm-hmm.

    3. WB

      Um, I think for the lot of people that wanna be founders, like, probably check out working at a startup. It's a great way to get your toes in the water and see if it's a good fit for you. Um, but if you have no choice to be a founder, it's all about, like, it's a roll of the dice, right? Like, every single day you roll the dice, and you just, if you roll enough dice, like, eventually it works out for you.

    4. ND

      Create your luck.

    5. WB

      You create your own luck, yeah. Luck is not something that-

    6. ND

      You need luck, but you can create some of it.

    7. WB

      Yeah. Exactly. Y- It's just like you don't know how well it's gonna work out, but you know if you keep trying enough, that you'll be able to do something that, that does work out.

    8. ND

      Awesome, Will.

    9. WB

      Yeah.

    10. ND

      Thank you so much for this advice and for joining us today.

    11. WB

      Yeah. Thank you. [outro music]

Episode duration: 16:17

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode ZxwYGbCOuDQ

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome