Jordan Dearsley (00:00:00):
We grew to about half a million in revenue over, like, three years, and we just gave all the money back to all of the users. Moved to SF and then kind of just was waiting around in the darkness for a good three or four months, me and my co-founder. It's really that I did not have it in me to put time into this and feel like my time was worthwhile anymore. I had convinced myself for long enough that it was worth my time, but I think it just finally became clear to me that I was lying to myself. And I think for the first time as a founder, I just had deep conviction on something very big because I finally just believe something that not that many other people seem to believe. Like my co-founder pulled me aside one time and was like, "Dude, if you don't get me a fucking infrastructure engineer, we are going to die." And I was like, "Oh, I need to stop coding now. Got it." So it grew quite quickly from where it was. It started at like $200/$300 a month for that sales training. To maybe six months later, we're at about $60K. An MRR, all self-serve, all developers. We're not really assigning any contracts, nothing like that.

Previous Guests (00:00:59):
That's product market fit. Product market fit. Product market fit. I called it the product market fit question. Product market fit. Product market fit. Product market fit. Product market fit. I mean, the name of the show is product market fit.

Pablo Strugo (00:01:11):
Do you think the Product Market Fit Show has product market fit? Because if you do, then there's something you just have to do. You have to take out your phone. You have to leave the show five stars. It lets us reach more founders, and it lets us get better guests. Thank you. Jordan, welcome to the show, man.

Jordan Dearsley (00:01:25):
Yeah, happy to be here. Was really excited to chat with you. Heard some other episodes, and yeah, I'm just honored to be here, honestly.

Pablo Strugo (00:01:33):
Well, I'm happy to have you here, man. You know, it's funny, like I think, especially during COVID, you know, there was a lot of talk about the death of the Bay Area. I'm like, it wasn't all that important anymore. And your story is kind of like you're in Toronto building, I might say struggling, for a few years. You move to the Bay Area. Maybe a year or so later, you launch a product. Year after that, you're $10 million in ARR, you visit $20 million Series A. So I don't know if the Bay Area is magical or not, but it certainly seems like it.

Jordan Dearsley (00:02:03):
Yeah, I think it is. Yeah. I think it's more to do with moving. I think it was the biggest thing for us. I think we got stuck in a rut of chasing a local maxima, working on what we worked on before, which is like a calendar app for three, four years. That moving was like the shock to the system that really did it. If we're going to move anywhere. Might as well move to SF, where gravity at least moves in the right direction.

Pablo Strugo (00:02:27):
So tell me the story. Like what? When do you start working on what became Vapi, and what's the team that's doing it? Give me that context.

Jordan Dearsley (00:02:35):
Yeah, yeah, yeah, for sure. So I'll kind of start from just before the pivot. So that would have been maybe July of 2023, I think, is when we decided, OK, his calendar app thing that we're doing, it was like a button for joining me super quick. We don't think that there's a big generational company to build here. We grew to about half a million in revenue over, like, three years, and we just gave all the money back to all of the users and then just kind of moved to SF and then kind of just was waiting around in the darkness for a good three or four months, me and my co-founder, just trying to find a problem to solve. We even went down to food banks and homeless shelters, just trying. as founders do, trying to find deep problems to solve. And we didn't find any there.

Pablo Strugo (00:03:26):
Tell me about that.

Jordan Dearsley (00:03:28):
Yeah, sorry. That's super random to bring up.

Pablo Strugo (00:03:30):
But I like it. What?

Jordan Dearsley (00:03:31):
Okay. Well, we need a really, really, really difficult problem to solve. So we're like, oh, obviously, world hunger and homelessness. Let's just go look at that. And we thought maybe there was something we could do there. Turns out it's a lot harder than what two tech bros can kind of come up with on their laptops. So. We kept pivoting around, pivoting around. I started to get personally. 

Pablo Strugo (00:03:53):
You know, just on that, I'll just go on a tangent on that because it's actually interesting. You know, one of the things. It's this weird balance. Because on the one hand, you want to solve, like, an important problem. And if you look at the. Like, getting a list of the most important problems in the world is actually not that hard. You know what I mean? Like, you know, climate change is one. Obviously poverty is another. Education is another. Like there are massive problems affecting hundreds of millions, in some cases, billions of people. But there's this other concept in economics called willingness to pay, which might as well be called ability to pay. And you need both to kind of come in because you can solve big problems. You can try to solve big problems, but if there's no ability or willingness to pay, you're not going to get the value or the value capture, and you're not going to get a business out of it. So there's a kind of a venn diagram, I guess, there that. But it's interesting, like a lot of times. First-time founders especially were either go to very niche, tiny little problems that are not that interesting. And maybe then they'll be like, "Ah, this is so interesting." Go all the way to kind of world hunger, kind of like you did. Hopefully, find that middle ground, but it's not uncommon.

Jordan Dearsley (00:04:55):
Yeah. I think the ambition was there, or like, oh, we just need to do something like super hard. I just think you're right. I think there's a bit of an overcorrection, or like, oh, and we just, like, solve the hardest fucking thing because, you know, we threw everything else out. So might as well try. But I think what also needs to be there is, like, some level of passion or determination for the problem. I think Michael Seibel. He was one of our group partners at YC. He told us quite early on, I'm skeptical that you guys are going to continue doing what you're doing for more than three months if it isn't working after three months. And he was right every single time. And every three months we would pivot, we would pivot, we would pivot. And so I think it does take some sort of like personal connection to a problem or at least some like long-term conviction just to hang in there long enough to see it through to something that's meaningful. And with the calendar app, we didn't really have any kind of long-term vision for the product. We were just like, OK, well, they want calendar links. We'll put calendar links in there. They want scheduling. We'll do scheduling. They want emails. We'll do emails. But there wasn't any long-term thinking. So, of course, we were never going to achieve something big long-term. It was just like, yeah, greedy algorithm.

Pablo Strugo (00:06:01):
How did you even get into the calendar app? You know, what is it? What was it, 2020 or whatever, when you started? Like, how do you even get into that?

Pablo Strugo (00:06:07):
We were just pivoting around looking for an idea, and then we built this like button to join meetings really quickly, and it had like 90% retention, and it just like started growing, and people just loved it. They were like, "Oh, I can join meetings fast now." Great, but it was like a very simple, tiny little. 

Pablo Strugo (00:06:20):
Feature right.

Pablo Strugo (00:06:22):
Yeah, but it was very high retention. So we're like, "Let's just, let's grasp onto this retention that we have, and I'll just build a bunch of other stuff around it. And, you know, that ended up turning into an AI note taker on it's later evolutions, where it's like, oh, well, they joined meetings really with high retention. Why not just, like, now just give them notes, and now there's more value with high retention. But I don't think it kind of worked. You know, half a million revenue is something, but it's not enough to build something generational. or a grander vision than notes. I think we can't realize, yeah, people don't like notes.
 
Pablo Strugo (00:06:54):
You know, give me some depth on the decision to throw that out and move because that's obviously a pivotal moment. And I think many founders are in that. Like if it doesn't work at all, it's easy. If it's crushing like you are right now, it's not that it's easy, but it's easy to know what to do. If you're in that gray, that's where you can get stuck.

Pablo Strugo (00:07:14):
So dangerous.

Jordan Dearsley (00:07:15):
It sucks. Yeah.

Pablo Strugo (00:07:16):
Walk me through that.

Jordan Dearsley (00:07:17):
And I hear it all the time. Because I posted about this story publicly on LinkedIn. And so I get a lot of founders reaching out to me. Like, how do I know if I should pivot or not pivot? At the end of the day, it just came down to something simple. I just did not have the will to continue. It's really that. I did not have it in me. to put time into this and feel like my time was worthwhile anymore. I had convinced myself for long enough that it was worth my time, but I think it just finally became clear to me that I was lying to myself. And that realization came from, I think it was a conversation with our advisor. Michael Seibel, where he was just like, "Guys, what the fuck are you doing? You don't care? You're working 9 to 5. What are you doing with your lives? And I was like, "Fuck. Yeah, right." Like, I have kind of not pissed away a large, you know, number of years, but, like, these are, like, the best ones. Shouldn't I be trying to do the craziest thing? And I'm just not. And so, of course, I'm not going to achieve crazy things. And so yeah, after that, I think we had a bit of decision paralysis. After we pivoted for those three months, we were like, well, it has to be really big. So, okay, this isn't really big. But this idea, the other idea we have about whatever, isn't very big. And so we just didn't really pick anything for a little while. And we were scared, I think, after burning ourselves. But that kind of led into me building. I was just grasping for ideas. I didn't know what to do. And so I built this AI therapist bot to go on walks with for, like, a couple hours of the day. And so I would call it and chat about my startup problems and all my ideas. I would help put it all together into something. And I would just keep going on walks with it every day, keep building it. Keep going on walks every day.

Pablo Strugo (00:08:49):
How did you come up with this in the first place?

Jordan Dearsley (00:08:53):
I think I tried PI or Siri or like one of these personal AI coaches and I was like I was just trying to talk to it I was like I don't want to fucking like chat with this thing I hate typing like I just want to like walk and think and have a fluid conversation with something so I'm like it must exist and I looked around and it was like wow it literally exists nowhere there is nothing at least pre like 4.0 audio this is like before we had audio models there literally was nothing that allows you to have like a fluid conversation And I was frustrated with that.

Pablo Strugo (00:09:22):
And so what happens after you build it?

Jordan Dearsley (00:09:23):
I walk every day, and I build every day, and I notice it's too slow. And then I notice it sounds too robotic. And I notice it decides to go on, like, long tangents and has to finish a complete essay of content before I can speak again. And I'm like, "Fuck, it's just like so annoying." At the end of the call, or sometimes the call just drops. And I lose the context of this like 20 minutes I've been investing in this like AI conversation. And it's just like, fuck. And so all these issues just kept compounding. I just kept building stuff into this therapy service that I was, you know, kind of building. I was like, oh, maybe there's a platform here. Like maybe I can now get us to other people. We found one guy who was like a friend of mine, Mo. He's also an angel investor in our company. And he was like, "This is the best thing ever. I called the, It's called Harman. I call Harman like 12 times a day." I'm like, "Whoa, okay." We have product market fit with one person. I think that was a bit more of a fluke. But maybe there is something to build there. Maybe it's just validating that voice is important. But I don't think it's validating that my thought buddy is valuable to people. I don't know how much you would have actually paid for that. And I couldn't replicate that again, even though I looked. And so. Back to pivot hell for a little while once again, but luckily I have my AI therapy buddy to help me through it. And I had spent a lot of time wrangling with these audio models to get it to the point where it was, and it was actually pretty good. Pretty good for what? I have no idea, but it was pretty good, and it felt fluid. And so when I stumbled on, I think it was Hyperbound. They were a YC startup, like two guys, Atul and Sriharsha. And they were just building this like AI sales role-play coach for their like YC startup. They had no revenue. They had just pivoted from an email client or something. And I talked to them, and they were struggling with the exact same thing that I had built, which is like the whole speech-to-speech interaction piece and changing these models and making it work reliably. And I was like, okay, why don't you just, like, use my thing? I'll rip off the therapy thing. It's not Harbin anymore. It's Voice AI Platform X. And I'll give you an endpoint to hit, and it'll set up like a WebSocket connection, and audio will stream in and out. Cool. And that was the first time I had a customer. And I was like, "I'll charge you, let's say, like five cents a minute." Sure. And I'll pass on the model cost at cost because I have no idea what you're going to spend. And so that actual pricing model actually ended up being, you know, the same pricing model we have today. But they ended up, after iterating for a few months, actually closing a deal with a sales team to do sales role-play coaching. And I was like, okay, interesting. There was a value here. Looked around, found some other startups doing the same thing. And then we were working with eight different kinds of YC or, like, not YC kinds of startups in the space for the few months following that first customer launching. So that's kind of the inception.

Pablo Strugo (00:12:17):
And what year is this? When exactly is this?

Jordan Dearsley (00:12:21):
This is maybe October of 2023. So it's a few months after that pivot and move in July.

Pablo Strugo (00:12:26):
Okay. It's a year after kind of ChatGPT comes out.

Jordan Dearsley (00:12:29):
Oh, I think so.

Jordan Dearsley (00:12:30):
What's happening in the voice? Because today there's so many voice apps or whatever. What's happening in the voice AI world at that point?

Jordan Dearsley (00:12:36):
Nothing.

Pablo Strugo (00:12:37):
Everything is in disparate. You have transcription models that are doing transcription large language models that are doing what they do. Text speech models are doing what they do. Nobody's really taken the time to put them together yet. And anyone who's tried has just, like, faced the same issues that I face and then solved it on my own. Nobody was really thinking about even speech-to-speech models. Which is now like a super popular thing that tons of researchers are spending time on. They did not exist. There was not a single speech-to-speech model in existence at the time, and there was no research that was far along, anything like that. I actually tried spinning one up and finding them myself so I could host them so I could solve my problem, but they didn't exist. So I had to build the best proxy I could, which is chaining the best-in-class pieces that were all over the place. But the pieces themselves, they were kind of slow. They were clunky. They weren't optimized for real time by any means. So I had to fight with deepRAM's support. People were like, "Please, real time is super important." I need your models to be faster. Please improve your streaming API. And it wasn't really that much of a priority for them at the time. Same thing with ElevenLabs. I was like, "Dude, the WebSocket streaming, I need it faster. I need it faster." And they just, like, real time wasn't as much of a thing.

Pablo Strugo (00:13:49):
Walk me through maybe the stack, like where is Vapi? Like there's so many different players that are part of that, right? Like, yeah, where does that become like maybe just today? Let's start there, and then we can go kind of from there.

Jordan Dearsley (00:14:00):
Sure, yeah, yeah. The way to think about us is like we're everything in between all these disparate pieces and like talking to a real voice agent on the phone that's integrated with your data and then scales to like millions of calls. So we are like the. I'm trying to think of a good analogy here. I would say maybe something like Stripe-esque, where it's like we kind of wrap together all of these, you know, complicated banking APIs, et cetera, into something that's like unified but still like super configurable and easy for developers and obviously works reliably at scale. So that's kind of what we've done for voice and all of these different model providers that exist. So now there's like hundreds of models and different providers on our platform that users can pick and choose between as if they're kind of building it themselves. So they kind of have a similar experience. What I was doing. Building my AI therapist back in the early days, where they're not sacrificing configuration or control, but they still like the ease of use and speed to deployment. So that's really like our value is just speed to deployment and then testing and then iteration and like that whole dev loop of improving your agents over time.

Pablo Strugo (00:15:06):
Maybe like a way to think about it in simple terms, like take ElevenLabs, probably the most famous out of the ones like you mentioned. Where do they stop, and where do you start? Like if I'm building a voice app, like do I, and I just want to build on top of ElevenLabs versus using you, kind of what's the difference?

Jordan Dearsley (00:15:21):
My answer would be different to like a year ago versus now. A year or two ago, they were just a text-to-speech API. And so you could use them, like you present text, and then they would send you back audio. And it's like, okay, great. But you can't have a conversation where you're texting it and then it's speaking back, right? And so you need to actually build an orchestration system that in real time knows, like, okay, you're done speaking now. Time now to hit their API. Get the result. Oh, you started speaking again? Okay, cut that off. And now let's generate the next statement the next time you're done. Like that kind of real-time orchestration is kind of what we did on top. Now they have a separate voice agent platform that they've started to release, but it's a bit more abstracted, whereas we're much more low-level. We expose all the raw pieces so developers can build entire products on top of it.

Pablo Strugo (00:16:10):
Got it. And then let's go back in time now, then. So you're doing this how, you know, because that's, like, voice has changed a lot, like in a short amount of time. How well did things work early on? And I'm trying to get a sense of promise versus reality, you know, for those first early kind of customers.

Jordan Dearsley (00:16:28):
Got it. And then let's go back in time now, then. So you're doing this how, you know, because that's, like, voice has changed a lot, like in a short amount of time. How well did things work early on? And I'm trying to get a sense of promise versus reality, you know, for those first early kind of customers. Yeah, I think there was some very early hype. There was this one player called Air AI at the time. And they were the first ones who were ever able to build a business doing AI phone calls. But they did fall into the trap of over-promising to users the capability of the technology at the time. Because the only models that could run at real-time speed. We're talking, like, under two seconds of, you know, to response time were like GPT 3.5 or like really shitty text-to-speech models. And so they were usually sounded cool in demos, but they weren't intelligent enough to actually, like, reliably perform the business objective. Like it would skip steps, you in a flow or reveal information to a user before it's supposed to or allow them to book an appointment for an invalid date. Those sort of things that make it so that you can't actually move the needle on business metrics reliably. And so we would see like 70% hallucination rates with 3.5 at the time. And so we just had to push on over probably a six-month period, like latency, latency, latency was our entire. My entire, I was gonna say our entire company, but it was just me. Me and my co-founder. But that was our focus was just like, how can we find ways to squeeze and squeeze more milliseconds out of this whole pipeline? We tried everything. We even started balancing requests amongst different regions of Azure, OpenAI so that, any one, moment when you finish your statement, we already know the fastest deployment to send it to and ping it to East US or ping it to West US. And these little optimizations over time all added up to something that got closer and closer, shaving it, shaving it to the point where it finally felt kind of real. And yeah. And of course, there were tailwinds as well. Like the models started to get faster. The companies like DPM ElevenLabs, et cetera, started focusing a lot more on real-time, smaller models. And so every maybe three or four months, we would see, like, a 20% improvement in latency or something like that. Of course, gains like that nowadays come much less frequently. But at the time it was like, oh shit, there's a new model. This is awesome. And it was like a step-change improvement in our product. And so, you know, we did our best with what we could do, but we were still beholden to the just general progress of tech at the time.

Pablo Strugo (00:18:51):
Did it feel inevitable at the time? Because you think about something like voice when you have text. I mean, it feels, you know, kind of obvious, like obviously people are going to want to build voice apps. I mean, it does now. I'm just wondering how much of that is hindsight 2020 versus at that time looking forward, did it feel like for sure there's going to be voice apps? Is it going to be us, or is it going to be somebody else that powers them?

Jordan Dearsley (00:19:11):
I mean, this is what was different from what we did with our last startup versus this startup. This startup, it was like, or this iteration. At this time, we were like, Okay, shit. If these models start getting closer and closer to human performance in this trend that we're seeing, this whole latency, cost, and performance kind of trends continue where they're going. Obviously we will hit human performance in like a year and a half. Like, it is absolutely inevitable that this will come. And I think for the first time as a founder, I just had deep conviction on something very big. I think that is what made it so I could continue working on this with it not working for more than three months. Because I finally just believe something that not that many other people seem to believe.

Pablo Strugo (00:19:54):
Well, walk me through that. Actually, that's a critical piece. So at that point, many people didn't think this was as inevitable as you thought it would be.

Jordan Dearsley (00:20:00):
No, I don't think anyone was focused on voice at all. Maybe ElevenLabs was still doing text-to-speech, but nobody was thinking about voice agents as a thing.

Pablo Strugo (00:20:09):
Man, it's crazy how fast things change. It's just ridiculous.

Jordan Dearsley (00:20:12):
Yeah, dude. And that was literally, what, two years ago? Yeah. Everyone was just so focused on text and what was going on with Eleven Labs that no one realized that, oh, of course, other mediums are going to be way more human and popular than the text medium. And so, yeah, I think that early conviction made me know, like vaguely, I know there's going to be this big shift that's coming. But what I didn't know is exactly what I would need to build to capitalize on that shift when it came. We actually thought for the longest time that what we were building was like the proxy to eventually becoming like the voice model company. Like, we thought we were going to raise $500 million. We were naive. We're going to raise $500 million, and we're going to train the master speech-to-speech model. I talked to everybody else. Nobody else was thinking about speech-to-speech models, at least publicly. OpenAI, I'm sure, was thinking about it a long time ago. But we thought that we were going to be an ML company and invest in that. But what we're going to do for now is chain together these models and raise money and get there eventually. But obviously, that didn't end up going the way that we thought. But we directionally placed a bet that roughly we know. We don't know how voice is going to come out in the world, but we do know that developers are going to want to build stuff to get it out into the world in ways that we just don't know yet. That was like a pretty clear bet at the time. So what we had conviction on, we bet on. And even if it was less than the full picture.

Pablo Strugo (00:21:33):
Walk me through, then kind of those iterations, like first quarter, second quarter, or months. I mean, I guess it was only 12 months, so there's not there's only so much. But like, you know what you built at first and then who built on top and what some of those step functions were, not just in terms of the models getting better, which I think everybody understands, but more in terms of the maybe the customer use cases or the unlocks in terms of the what you enable, you know what I mean, for your end customers over time.

Jordan Dearsley (00:21:59):
Yeah, I mean, at the start, it was, like, very experimental stuff. Like, it was really that one, like, sales role-play use case, role-play training use case. That was, like, the number one contributor to our volume. If I look at the chart, I have it in front of me. Like, October 2023, November 2023, December 2023, it's mostly just, like, that one company, plus, like, random folks trickling in from, like, hearing about us on TikTok or, like, whatever. Like, no, like, serious use cases. Yeah. We did have some like random, like tiny unfunded startups also building like AI interviewer. For like product management interviews or this random company in Sweden that is building this tech to help Ski resorts, like accept ML calls. Yeah, right, and so it's all over the place. 

Pablo Strugo (00:22:44):
So niche, yeah.

Jordan Dearsley (00:22:45):
Yeah, it was super random, but like this was representative of the pie of companies that were thinking about building voice at the time, and then if I fast forward maybe in like February, March-ish. We start to see a lot of growth. So I have in front of me. So in like January 2024, we had like 32,000 calls on the platform if I even zoom back further than that, and that's all just sales role-play training basically. 

Pablo Strugo (00:23:10):
Wow, so that business that you partnered with was doing really well, I have to assume.

Jordan Dearsley (00:23:14):
Yeah, they did decently well. They had these 32K calls. Then we started to get some other folks trickling in when we kind of hit 52K calls. We're looking at February. March was like 164,000 calls. We launched on product. So that was the first time we ever launched publicly. And so immediately we just tripled our volume.

Pablo Strugo (00:23:32):
Tell me about that, actually. What everything around, kind of like planning for it. The launch on Product Hunt. What it did. What it felt like.

Jordan Dearsley (00:23:38):
Yeah, the Product Hunt launch was. It was later than it should have been. I think we should have launched earlier and got more customers earlier, honestly, just to get more feedback. But I think we were just waiting to finish, like, oh, it needs to work on websites too. And it needs to work on not just Twilio phone numbers but, Vonage phone numbers and needs to have all this. So we kind of got in our heads about it needing to be bigger and bigger. But we did launch it once we thought it was ready, and biggest thing that we did was just like we posted a tweet and just got a ton of famous people we knew to retweet it or sorry not new but new of well enough to beg them and then we just had a cool video that some guy in our discord reached out and was like can I make a video for you for free we're like oh sick yeah come over and he did he made an awesome video for us and then we just got a bunch of friends with followings to retweet it. And that was literally just what started it.

Pablo Strugo (00:24:28):
And so a product on basically 4x, like a 4x call volume.

Jordan Dearsley (00:24:31):
Yeah, I'm looking here, like 50. No, I'd say 3x, 3x the call volume. Yeah, 165 that month of March. But honestly, you would have expected after that there to be a big spike and then a dip or something. But for the most part, it just, like, continued growing, but kind of at a slow pace. Like April that year, 279, then 464.

Pablo Strugo (00:24:51):
That's not that slow, man. That's like a double every month.

Jordan Dearsley (00:24:55):
Yeah, I guess it was. I guess it was doubling every month. At the time, it didn't feel like super fast growth.

Pablo Strugo (00:25:01):
Really? Okay.

Jordan Dearsley (00:25:02):
No, no, not at all. Because even though these numbers are trying to think of how to explain it, it was still just not that many people, right? And so it wasn't that noisy. And then I think maybe mid-year, we just started opening tons of Slack channels with everyone that was using our platform. We had maybe like 300, 400 Slack channels. And then it got noisy. Then, like, we just had, like, tons of people hounding us for support all the time, varying levels of technical ability.

Pablo Strugo (00:25:26):
How many people are on the team at this point?

Jordan Dearsley (00:25:28):
Me and Nikhil.

Pablo Strugo (00:25:29):
Oh, wow. Okay.

Jordan Dearsley (00:25:30):
I think maybe we hired, like, a couple guys out of Discord. That's, like, Sahil and Shubham. Like, they were in our Discord. We just kind of, like, brought them on board. But that was it.

Pablo Strugo (00:25:42):
And what's that generate? Like, that half a million call volume in whatever that was, June or whatever, like, what's that generating in ARR?

Jordan Dearsley (00:25:48):
So it would have been, on average, maybe like 12 cents a minute. So, you're looking at $500,000. You're about $60K and MRR.

Pablo Strugo (00:25:57):
Okay.

Jordan Dearsley (00:25:58):
Yeah. So it grew quite quickly. From where it was, it started at like $200, $300 a month for that. Sales training to maybe six months later, we're at about $60K and MRR, all self-serve, all developers. We're not really assigning any contracts, nothing like that.

Pablo Strugo (00:26:13):
What do you feel from a pull perspective versus what you felt originally with that button for the calendar app?

Jordan Dearsley (00:26:23):
Completely different. The more that, so once we did end up opening all those Slack channels with our users, we were like, holy shit, there is an infinite amount of stuff to do. And everyone is super noisy all the time to the point where, like, we actually had to shut down all of our Slack channels. We had to close them with all of our customers just because we could not focus on what was important. And actually, it had this weird rebound effect where you think it would hurt our brand or something, but people were like, wow, these guys are doing so well that they close all of their Slack channels with their customers. It got around that we were doing better. It was like, what? But that is why. We were just way too busy.

Pablo Strugo (00:27:02):
I'm going to ask you for a small favor, a tiny little favor. In fact, it's not even now that I think about it. It's not even really a favor for me. I'm actually trying to help you, do a favor for you. Just hit the follow button. You won't miss out on the next episode. You'll see everything that we release. If you don't want to listen to an episode, you just skip it, but at least you don't miss out. And in a sense, I mean, I would argue, like, the. I've seen this now a few times, like, obviously the first version of products are never, you know, great, but. And you don't want people to be complaining. That's not like a positive thing. And yet it's kind of a signal of a positive thing because, like, if people are shouting at you to fix this, to add that, whatever, it signals that the thing you've built is something they're actually relying on, right? Like if they don't really care that you don't have that new feature, then it's okay, their life is going. If you're like the reason they can't grow, the reason they can't do whatever they need to do, that's value to you. You know, that's how important you are.

Jordan Dearsley (00:27:57):
And this is, like, I think the key part. If I kind of gloss over the part where, like, we went from things not working to things working, what was very, very different this time is we had a lot of, like, I guess, relatively low-value users that had a 10 out of 10 pain. Like if we did not find a way to add auth tokens to API calls for tool calls, they could. Their business would not exist. Like it was just like existential for everybody. And so constantly everything was P zero all the time. We had an outage, like, even if we only had, like, eight users, they were all pissed. Like usually if you have an outage and you have eight users, like nobody notices.

Pablo Strugo (00:28:33):
That's right, you don't hear about it.

Jordan Dearsley (00:28:36):
Where it's like, oh no, like their customers, like the customers that they have, are pissed at them because they didn't get their calls answered. And so they lost money. And so then they got mad at us, of course, like rightfully so. And so we were constantly just facing all these reliability issues. And anytime we would want to move fast and push a change, it would break something, and then a customer would get pissed. And so it was this constant balancing act of just trying to push forward on what we thought was important. But for the most part, just listening to what people wanted and just doing it like immediately. If people would ask me for stuff, like within 10 minutes, I would have it up. Like a new API param or something like that.

Pablo Strugo (00:29:12):
How much easier is everything once you have that like burning need kind of driving you?

Jordan Dearsley (00:29:17):
It's not easier. Like, it's harder. It's there's a difference between, like, the feeling of uncertainty and that being easier hard and actually the work being easier hard, if that makes sense. Yes, like it went from things are hard because I don't know what the fuck I'm doing to like things are hard because there's just so much to fucking do all the time. But also in the back of your head, you're like, OK, but is this customer representative of the future customer? I don't know if this is important to do. So then there's also this little game you're playing. But it's at least better than sniffing around food. What do you call it? Soup kitchens and stuff trying to find a problem. 

Pablo Strugo (00:29:56):
Trying to find problems. Yeah, well, I guess what happens is at least you get clarity. like, obviously there's always questions and uncertainty but you do get a lot of clarity from that. The market, yelling at you to, like, dude, do this, fix this. Whatever you're like, okay I gotta. I gotta get these things done versus what are you trying. To like, find that pull, you know what I mean, you're like, I wonder if this next thing I'm building is gonna be the thing that tips us over, and you're just constantly in this motion of building and hoping, building and hoping, versus building and knowing that you're solving somebody's problem because of this thing you build, because they were yelling at you to build it in the first place.

Jordan Dearsley (00:30:30):
Yes. But also, if I zoom back to that time in April-ish, 2024, when things started to pick up, OpenAI also announced GPT-4.0 Audio. And remember, at that time, we thought we were going to be a model company that was going to be the one to build this. And so me and, actually, now I remember correctly, I think we had like six people maybe on the team in total. We kind of looked at each other and like, Oh, fuck, well, that was our whole vision and everything. Like, what do we do now? But it kept growing. We were just sort of like disheartened for a moment. And we were like, maybe we. Maybe this is it, and we talked to an investor of ours, and yeah, just, you know, pack it up like it's fine, like whatever, it happens all the time, and so we had a moment we were like, I don't know, it was like when one day of a moment, and they were like, let's just keep going. Let's just kind of see what happens. Who knows we just kept pushing, kept doing what we're doing, and just kept growing every week, like 10 percent, 20 percent. And we were like, I guess GPT-4.0 just, like, grew everyone's awareness that voice was important. It didn't really displace us by any means. It just made it more popular. And so then after that, like everything just started to grow extremely quickly. But we still had this feeling inside that was like, Fuck, OpenAI is just going to kill us. Or like something is just going to come and kill us because we're just like a rapper. And then eventually I met a sales advisor. His name is Mitch Miranda. He's awesome. And he was like, Look, guys, enterprises, they can't use OpenAI. It's just not physically possible. It's just way too fucking complicated. There's a constant refining and tuning and expertise that you guys are building in your team that these enterprises just don't have and would rather pay to have than build it themselves. both from infrastructure and model knowledge and all this other stuff that we've been building as a team this whole time. And he was like, long-term, that is your defensibility against all these big guys, is going into enterprise and solving that last-mile schlep of actually getting all these voice agents to production with all of the complexity that you've seen so far serving all these tiny startups. And so that gave us enough conviction that, like, okay, actually, like, green light, we're fine. Let's just keep growing our self-serve motion and keep growing its defensibility by making it more and more complex and featureful and start building out an enterprise motion that's defensible just because we get lock-in with these very large and slow-moving enterprises that have no way of accessing this technology without a partner. And that was maybe like June of last year or June of, yeah, 2024, when we signed our first enterprise customer, and we saw this happen, and it kind of worked.

Pablo Strugo (00:33:04):
Yeah, let's dive in on that because this is something, at least for me, I've been thinking about a lot, which is where can startups play relative to the big guys? And OpenAI being obviously one of them, but you mentioned enterprise, like Cohere is still a $5 billion player that's focusing just on enterprise. And then in voice, you've got like ElevenLabs. How do you think about that? Like, for example, take this enterprise thing. We double click on that. Like, how long how long does that last? Is it until OpenAI or like Cohere does more voice stuff or OpenAI does more enterprise stuff or Eleven Labs goes further into that last mile? Or is there a reason to believe that you're always going to have like that last mile space for you?

Jordan Dearsley (00:33:49):
I think as time has gone on, we've seen competition come from every angle. We've seen Twilio working on stuff and ElevenLabs working on stuff. And I'm sure Apple and Google, or sorry, Google and Amazon are going to work on some very large enterprise-facing versions of what we have. But I think what we have built slowly over time, working with that first engineer who needed a voice agent, to the startups to the enterprises that we gained experience with, is that that experience that we've gained working with them is actually very valuable and hard to, actually physically impossible to, shortcut. Every single parameter in our API is from us hearing from a developer, figuring out the core pain that they have, and shipping it and seeing how they did and then tuning it again and trying again with a bunch of other developers. And so our product is a representation of everything that we have learned so far. And so to think that, okay, maybe they're just going to copy our entire API as the solution. That's a lot of shit to build, especially for a really large company. Like, the surface area of our platform is enormous. So that's tricky. And then when it comes to actually working with enterprise customers, just internally, we have a lot of expertise that we have documented about, specifically for, like, prior authorization calls. We know the best way to tune a voice agent using our platform to make it so that it can navigate the IVR tree effectively. Like that expertise is very hard to learn without having done it yourself many times.

Pablo Strugo (00:35:17):
Let me ask you another question. Are you fearful or excited by the new OpenAI model releases? Like OpenAI is about to release a model tomorrow. Do you feel like, holy shit, what are they going to do now? Or are you like, yes, hopefully this is like a 10x improvement?

Jordan Dearsley (00:35:32):
I feel, like, 80% excited. 20% scared because I'm always paranoid. Yeah, I think 80% is like, oh, fuck yeah. We finally have, like, opening our real-time, and it's ready for our customers to choose to use on our platform. It's in there. Great. It also means that we have more explaining to do when a customer comes to us and is like, why can't we just use real time? And we're like, okay, here's 10 other customers that we saw try. Here's everything that they ran into trying to. Here's everything you're not thinking about, et cetera, et cetera. Where it does, as there's more awareness that voice is a thing, there is more simplifying of the problem that we need to educate the market on. So that is one thing. And the other thing is like, yes, I do sometimes have an existential, like, oh fuck, like, what if tomorrow there's like a new voice agent SDK and the infrastructure has like telephony integrations and they think about HIPAA compliance, and I don't think it's possible. But, you know, I do always like to be a little scared.

Pablo Strugo (00:36:22):
I think you have to be paranoid as a founder so that, and that goes for, like, way beyond AI. Like, that's for everything. But I ask that question because that's become my go-to question when I think about investing in new AI companies. You know, given the positioning of this company, would they be, because you used to have, like, you know, PDF summarizers as, you know, like a GPT wrapper, right? And then, obviously, that company is going to only be fearful of the next opening eye announcement. Because they're like, well, they could just do, you know, input a PDF, you're dead, and that goes, like, many different things, and you know that doesn't mean you can't build. Like, you know, think perplexity, but I think perplexity is probably scared of OpenAI. Now they got to search. What is even the difference today between them two sand it kind of starts to, like, you know, wedge, and you kind of can't tell them apart. And that's only value to the to the bigger player but they build a pretty big business already, but I guess my point is, like, that's become my big question. And that's what I tell founders to think about is whatever positioning you're taking, think to yourself how you feel about whether it's OpenAI. Pick your other big model. Like, how do you feel about them? If them launching a new model makes your product better and you deliver more value, that's good. If it makes you scared and withers like how much value you're actually creating for end customer, that's not a great place to be because you can only assume those models are only going to get better over time.

Jordan Dearsley (00:37:36):
I think where I thought the only place that I can be safe, or I still do believe the only place that we can be safe, is kind of covering that last mile because I feel like that's the trickiest part for a super, super large company to tackle. Last mile to enterprise, last mile to actually, developer getting live, and for the end use case. And then lastly, my assumption is there's enough meat there in that gap between models and production that something defensible can be built. And so that's why we do what we do.

Pablo Strugo (00:38:10):
So walk me through the second half of last year. So you went from what, $60K MRR to like a million MRR in those six months. Is that right?

Jordan Dearsley (00:38:19):
I can do the math super quick.

Pablo Strugo (00:38:21):
That's wild.

Jordan Dearsley (00:38:22):
Let me take a look. April 279, 464. July was 692. July was 1.1, sorry. August was 1.7, 2.1, like 2.6. Yeah. And so it just kind of, yeah, just kind of kept going, man.

Pablo Strugo (00:38:29):
And that's call volume, right? Those numbers?

Jordan Dearsley (00:38:41):
That's called volume. So if I map that to revenue, within six months, we were at about, like, $4 million in revenue annually recurring. Very small, small section of it was enterprise. Now that's larger, but most of it was just like self-serve organic. We did zero ads. We did zero marketing. We didn't have the capacity for marketing. We just had this Discord community, and, like, maybe some people make YouTube videos about us. That's basically it.

Pablo Strugo (00:39:05):
What does it feel like to grow that fast internally?

Jordan Dearsley (00:39:08):
You kind of take it for granted. Honestly, it's like you have a 4% week growth, and you're like, Oh wow, we're really, really losing this week. It just becomes your baseline. 10% becomes your baseline. And so it's very normal. The only difference in how it feels is more stuff is breaking more often.

Pablo Strugo (00:39:30):
That's the other side of my question. Does it feel like there's fires burning all over the place?

Jordan Dearsley (00:39:34):
It constantly felt like that. My co-founder pulled me aside one time and was like, Dude, if you don't get me a fucking infrastructure engineer, we are going to die. And I was like, oh, I need to stop coding now. Got it. And so, yeah, like, it's just pure necessity. Everything with that kind of growth.

Pablo Strugo (00:39:52):
How many people, like, so you were two people at the beginning of 24. How many people at the end of 24 and last year?

Jordan Dearsley (00:39:57):
At the end of 24, I think we finished with maybe 14 or so.

Pablo Strugo (00:40:01):
Now we're at 14 people and like $4 or $5 million ARR.

Jordan Dearsley (00:40:05):
By then it would have been higher. This is where I need to get a bit fuzzy with numbers, unfortunately. But today we're now like 26 people.

Pablo Strugo (00:40:14):
Okay.

Jordan Dearsley (00:40:14):
Yeah.

Pablo Strugo (00:40:15):
And then, you know, one of the things I, like, I think. One of the most. I mean, the pivotal moment is landing on this problem set. I don't know how much time you had to think, but, like, what advice would you have for founders that are looking for these kinds of, like, must truly. I think that's the key thing. Like when I think about, you know, I've done like over a hundred interviews now on people who found product market fit, and consistently, and frankly, even within these companies, most of which are successful, there's still gradients of the level of true need that their product, you know, really solves. But I feel like that's still the biggest thing. Like if you can land on a must-solve problem, then everything. And that was my question, again. Easier might be the wrong word, but certainly gets clear, like every next step just gets clearer and clearer, clearer, and you have this crazy velocity. So what's your, like, but then you could look at your story and be like, Ah, he got lucky. You know what I mean? Like, yeah, he moved, played around. It happened to be, you know, 22, 23, you know, happened to land on this voice thing, and it took off. You know what I mean? Like, what's your thinking around that? And what's your advice to founders that are looking for these kind of must-solve problems?

Jordan Dearsley (00:41:26):
A few things. I think it's quite important that you just find someone who has a 10 out of 10 burning pain. When I say 10 out of 10, I mean literally ask them, on a scale of 1 to 10, how painful is this? I know it sounds like a silly question, but just ask them because then you'll hear six, and you're like, wait, what? It's very important.

Pablo Strugo (00:41:42):
It's true.

Jordan Dearsley (00:41:43):
That people actually believe that they have it and then get them to convince it to you, like a 10 out of 10. Like, is it really 10 out of 10? Like, isn't this other thing you're talking about 10 out of 10? Like, why this specifically? And then they'll be, like, Oh, actually it's 8. And you're like, okay, now we know. And so it's about pushing on that and getting them to prove to you that it's as painful as it is. Because you deserve to know before you fucking spend four months building on something that might not solve the heart of a problem. And so what we did is we found a 10 out of 10 pain for this one specific AI sales role-play training company where it was existential that they figured it out or their company would die. We just so happened to be in a place where lots of other people were going to start having that problem too. I would have been a slightly more intelligent at the time, which I kind of became like a couple months later as I saw the growth. I would think, okay, latency is improving. Cost is improving. Performance is improving. Like, QED, there's going to be a lot more stuff that people want to build with voice in 18 months. I think if you can just have a high-level conviction on where the world will be, like realistically, you can at least directionally point yourself in the right. Like following gravity, right?

Pablo Strugo (00:43:03):
But to be clear. Only after having found that 10 out of 10 problem, right? Or- 

Jordan Dearsley (00:43:07):
I don't know if there's an order that's good or bad. Maybe the other way around.

Jordan Dearsley (00:43:12):
The other way around, you can really, I mean, there are proven founders who know spaces so deeply that it's fine. But I'm going to argue that for 95% plus of founders. The other way can lead you a stray very often because it's very easy to convince yourself the world's going to go a certain way.

Jordan Dearsley (00:43:25):
And you're like, well, in the future, it's going to be like this that you build. And in the meantime, nobody's got 10 out of 10 problems. So you got no bar.

Pablo Strugo (00:43:32):
Yes, that's true. I think it's fine to start with a hypothesis, but just like very quickly after, you need to find the person who has the pain to prove that it's worth pursuing. Like, one way or the other. Right.

Pablo Strugo (00:43:41):
Like you guys need to be motivated to have a handful of 10 out of 10 problem. And think there's going to be more of these in the future. That's fine. What's scary is to say people will have a 10 out of 10 problem. They just don't yet. And then it's like, OK.

Jordan Dearsley (00:43:56):
Yes, yes. No, no, no, no, no. You're, yeah, you need to find at least, like, one person.

Pablo Strugo (00:44:00):
That's right.

Jordan Dearsley (00:44:02):
And that's a realistic path too.

Pablo Strugo (00:44:03):
But having, you know, the other question I have is, having been in these kind of two worlds, right, of like not 10 out of 10 and now 10 out of 10, what would, like, let's say, because what I go back to. A point I made earlier. Like, there are a lot of founders that, if they're honest with themselves, they're doing sub-a-million, maybe they're doing a million or $2 million in ARR, maybe they're doing $100K, right? But somewhere around there. And if they're honest with themselves, they would kind of realize that they're not really solving a true, like, 10 out of 10 problem. What's your advice to them?

Jordan Dearsley (00:44:33):
Oh man, at a million, that's so tough. At $100,000, it's a bit easier. At a million, there's a lot of investment. You've got a full team behind you. That's a tough call.

Pablo Strugo (00:44:42):
But let's say at $500K, at $100K.

Jordan Dearsley (00:44:45):
At the earlier stages, one, it's okay to throw everything out if you have to, but you also have earned exposure to $100,000 worth of value. And so you should probably talk to those people and find out what else you could do in their daily lives. What are the top five pains in your life? What are they? And they'll be like, one, two, three, four, five. You're at five. And it's like, well, why didn't you get me solving one?

Pablo Strugo (00:45:10):
That's right.

Jordan Dearsley (00:45:13):
And then there's ways to expand and kind of work with those users to then pivot to something that's higher value. But I think what's important in that situation where you have revenue is you need to find a way to build conviction. Otherwise, you will just continue doing what you're doing. And so that's why you need to be talking to customers, running experiments, and being willing to let that thing go if the data pushes you in that direction.

Pablo Strugo (00:45:38):
I have specific examples of people on this show that did exactly. One that comes to mind is a company called Noibu was doing something in e-commerce for two years. They're 3K MRR. And they ended up very similar to you, saying, "Screw this. I'm not going to another day." Some catalyst happened. Doesn't really matter. And then they went back to everyone, like, ask for the real problems. And they got a list of real problems. They took the first one. They found a way to solve it. Two years after that, they're doing $300K MRR, right? And like a very simple story, but it happens. More often than you think. And you can really, you really can ask customers, your ICP, what is the biggest problem for them? And they'll have an answer.

Jordan Dearsley (00:46:12):
Yeah, definitely. For the one where you're like, You're already at a million. That's tough. That's like, I mean, we threw away $500K, right? So I somehow, you know, and that was like a big life decision for me. Like it wasn't easy to do that. But we kind of just put the business on autopilot for a couple, months and just went exploring. I did say we gave the money back. We actually gave it back four months later. We took some time to hedge a little bit. I thought that made it easier.

Jordan Dearsley (00:46:38):
When you moved, what kind of personal runway did you have? Oh, we could have kept going. We were like, breakeven at that time. Yeah.

Pablo Strugo (00:46:44):
Oh, but you had, like, cause you had your existing business or what? You would have kept. You were breaking on that. I see.

Jordan Dearsley (00:46:50):
Yeah. So, so we, we were already default alive. We spent a million over like three years. Like we were super slim. And so we could just keep going.

Pablo Strugo (00:46:59):
So you didn't, right? It's not like you shut it down, went to zero, and then had a personal run where you moved and then had this thing break even. And then when you figured it out, you kind of cut it out.

Jordan Dearsley (00:47:07):
Yeah, but what was most important is like we just had to completely shut off all thinking about the past product, like 100%.

Pablo Strugo (00:47:12):
You maintained.

Jordan Dearsley (00:47:13):
Yeah, we had, like, any support emails that came in. It was just like, Here's a refund. Like we didn't, exactly. We didn't. We spent zero time on it. And so we just needed to make that mental space to, like, make an informed call. But I think we were quite certain, like we were all just burnt out on it. So I don't think we could have continued doing the calendar thing even if we wanted to.

Pablo Strugo (00:47:32):
I mean, I think you mentioned some, but any that come to mind, any points where you thought you're just going to fail? Like, this was just not going to work. You're just going to quit and maybe go get a job?

Jordan Dearsley (00:47:42):
During those three months, there were a couple times when Nikhil almost ran off. My co-founder was like, oh, we should go to India and help those kids. That's my end goal anyways. Might as well do it now or almost ran off to, like, join a crypto gang helping tribes in Africa like, trade, and it's just like, yeah, it got really weird for a sec. And so I just felt the need to just, like, grasp at straws and just, like, make something. And so that thing was the AI therapist that I kind of started all.

Jordan Dearsley (00:48:11):
By the way, I guess we didn't touch on this. What did you do before Vapi? Like, what was in your co-founder?

Jordan Dearsley (00:48:17):
Before Vapi or before, like, the superpowered. Like, the AI no ticker.

Pablo Strugo (00:48:20):
Before, yeah. So, yeah, before your original product.

Jordan Dearsley (00:48:23):
Oh, yeah. We dropped out of school. We dropped out of school to do all this. Yeah, we met at the University of Waterloo. Dropped out in like our third year to start this. And then just like, yeah, then all this happened after that.

Pablo Strugo (00:48:35):
Awesome, man. Cool. Look, let's stop it there. I think we touched. I mean, I have a few questions. I was going to end on, but I think we touched on failure. We touched on the moment of maybe I'll ask it anyways, actually, like because you could talk about a few moments. When was the moment? Where you personally felt, like, you had reached true product market fit?

Jordan Dearsley (00:48:53):
That's really hard. Probably maybe even like after we were at like a couple million in revenue and we had an outage and everyone was just like so fucking mad that was probably that was probably it. But still I wouldn't. I don't know if I would call that product market because I always thought it was like this big shiny thing that was like spiritual and whatever. It just became, remember, I was chasing it for like three/four years and never had it right. And so I always, but I think it was clear looking back it's like, oh, yeah, when everything went down and everyone's businesses went down. It's like, oh yeah, that's, yes, we have product market fit.

Pablo Strugo (00:49:29):
And then last question, what would be your biggest piece of advice to an early-stage founder that doesn't yet have product market fit?

Jordan Dearsley (00:49:38):
Don't think it's a really big thing. I think product market fit can be achieved with one person and one user. I think it's as easy as that. You're going to have product market fit immediately with an extremely, extremely, extremely small market as, N of one. And then you just need to achieve product market fit with two, and then four, and then five, and then ten. I think that's a much better mindset to have than to think it's like off in the distance and like a magical thing that you just don't understand because you're not good enough.

Pablo Strugo (00:50:00):
Perfect. Well, Jordan, thanks so much for spending time, dude. It's been great.

Jordan Dearsley (00:50:03):
Of course. Have a good one, man.

Jordan Dearsley (00:50:06):
Listen, when you go to, like, a restaurant, you eat a nice meal, maybe a fancy one, maybe not. Do you leave a tip? I assume you probably leave a tip. You probably leave a tip 100% of the time. Well, guess what? A review is just like a tip. And I know you haven't been leaving one. So just like the waiter that doesn't get a tip after hours of great service, I'm getting a little frustrated. So take your phone out and leave a review. It helps the show move up rankings. It helps us get better guests. It doesn't just help me. It helps way more founders. Thank you.