AIMinds #042 | Tom & Adrian from Canonical AI

About this episode

Tom Shapland is the cofounder of Canonical AI. He loves working with Voice AI developers to help them analyze and improve their agents. Before Canonical AI, he was the cofounder of an agriculture technology startup.

Adrian Cowham, a seasoned developer with 20+ years in R&D, specializes in voice AI and analytics. He created a platform to help developers enhance voice interactions by analyzing success rates, relevance, and user engagement.

Listen to the episode on Spotify, Apple Podcast, Podcast addicts, Castbox. You can also watch this episode on YouTube.

In this episode of AI Minds, hosts Demetrios and guests Adrian Cowham and Tom Shapland, co-founders of Canonical AI, discuss the transformative impact of voice AI on communication and business. They explore how Canonical AI aids developers in refining voice applications by providing robust analytics to enhance user interactions and call quality.

Adrian and Tom highlight challenges like user reluctance from poor past experiences and the technical complexity of delivering nuanced voice analytics akin to web platforms. They share how Canonical AI tackles these issues with advanced metrics, including call mapping and custom alerts, empowering developers with actionable insights.

The founders also reflect on their journey from agricultural tech to voice AI, driven by a passion for innovation. They envision voice AI replacing traditional interfaces, emphasizing the importance of community support in advancing this fast-growing field.

Fun Fact: In an earlier venture, Adrian and Tom experimented with creating an AI-enhanced reading companion for book publishers, which didn't succeed at first but helped pave their path toward Voice AI. This sort of "talking book" concept was both innovative and amusing in its inception.

Show Notes:

00:00 Corporate reorgs led to entrepreneurial success journey.

04:02 Adrian crucial in culture; CEO maintained order.

07:27 Started without a plan, explored AI possibilities.

12:44 Voice AI community fosters growth and collaboration.

16:13 Voice AI revolutionizes call handling in businesses.

17:54 Voice AI enables new interactions and experiences.

22:19 Technology's trajectory: Voice AI enhancing consumer experience.

26:21 Analyzing user flow and behavior via analytics.

28:53 Levels: Latency, Understanding, Execution, System, Bytes/Bits.

30:26 Deep analysis of signal processing and conversation.

More Quotes from [Speaker]:

Transcript:

Demetrios:

All right, folks, welcome back to the AI Minds podcast. This is a podcast where we explore the companies of tomorrow being built AI first. This episode, like every episode, is hosted by myself, Demetrios, and it is brought to you by Deepgram. The number one speech to text and text to speech API on the Internet today, trusted by the world's top conversational AI leaders, startups and enterprises such as Spotify, Twilio, NASA and Citibank. We are joined by not one, but two very special guests, the co founders Adrian and Tom of Canonical AI. How you all doing today?

Adrian Cowham:

Great, great. How are you doing today?

Demetrios:

I'm excellent, now that we get to have this conversation. So Adrian, let's start with you. I know that you've had some vast experience learning to code in high school from your brother in the 90s, coming out and living the dot com era. You're working at HP those days?

Adrian Cowham:

Yeah, yep, working at Hewlett Packard.

Demetrios:

And then you started to go the startup route. What made you want to go into startups?

Adrian Cowham:

No, the large enterprise world is interesting. You go through every six months there's a reorg, another reorg, another reorg and you're kind of just left dangling a lot of times. And I was kind of an ambitious engineer. I wanted to keep going, keep learning and keep getting pushed and, and it was good at HP in the early days and that was there. But then things started to slow down and kind of started to hit a ceiling a little bit and then they're like, oh, you should go into management and da, da, da, da. I'm like, I'm a gearhead, man. I gotta be pushing pixels and, you know, pushing bits and bytes and so. And there was a little cohort of us at this, at HP that kind of felt the same way and they ended up leaving the company and starting their own and recruited me out and then we had a successful startup, they recruited me out to the next one and so it's just been kind of snowballed from there.

Demetrios:

So this group is either a good influence or a bad influence on you, depending on how you look at it, 100%.

Adrian Cowham:

It's just, it's all about perspective. For me personally, it's been an extremely positive and life changing experience. So I'm eternally grateful for the cats that kind of helped me get into the startup world. For the people around me. People close to me. Nah, they just, you know, just, hey, do a nine to five, you know, log off at 4:30 and then maybe 3:00 on Fridays and let's go hang out. Let's go do this, that and the other thing. Man, I got too much ambition.

Demetrios:

Yeah. So you spent 10 long years of your life, a decade of your life at a company, and it was in the agricultural space. You were CTO there too, if I am, if I recall correctly.

Adrian Cowham:

Yeah, yeah, definitely. That's right. Tom asked me to be CTO about halfway through that experience. And we built some really cool stuff. We rebooted the company. We got it into a position that we were really happy with. And that's kind of one of those things where it's like, okay, how much more can we squeeze out of this company? And we got approached with the acquisition offer. We're like, okay, let's do this and let's assess the situation.

Adrian Cowham:

Let's see where we can go from here and now.

Demetrios:

Tom, this is where you two met, right?

Tom Shapland:

Yes, that's right. Adrian and I met at the agriculture technology company that we built for 10 years. We started that company. So it was actually my co founder was from a different co founder for that company, and he had worked with Adrian previously at other startups. And it's been really fun working with Adrian. He can build anything and always has a great attitude. He was the cornerstone of our culture. At our last company, I was kind of a square, right? I was the CEO and kind of a square, keeping everyone on track.

Tom Shapland:

And then Adrian made sure that there was glue among all. All the people there that were and made sure everyone was enjoying the journey.

Demetrios:

And so the last company was in agriculture. It. It was an agtech company, Correct? What. What was it exactly?

Tom Shapland:

So the company helped farmers make irrigation decisions. It was based on an atmospheric turbulence innovation that I developed during my PhD work at U. I'd always been interested in business. I didn't want to see this technology that we developed just be another paper that nobody read. In fact, nobody has still read the paper. I can go on Google Scholar and go look it up like, no, nobody reads that. I occasionally threaten family members to like that. I'll quiz them on what's in my paper and that they have to read it.

Tom Shapland:

And they're like, okay, okay, we'll back off. We'll stop making this request of you. You're pulling the read my papers card. And so anyway, we started this company and we were lucky enough to get into Y Combinator and that really was the start of the journey with startups. For me, it was. It was quite the culture shock to go from academia and the glacial pace of academia to being in Mountain View. Among all these startups in yc. But it was.

Tom Shapland:

It was amazing. And yc, I'm very grateful to YC and how it really kickstarted my journey in startups.

Demetrios:

Well, it seems like you took to it quite nicely because after a bit of time, a decade, I guess, you decided that you've done what you needed to do and you had a successful outcome with this. And then you said, I haven't had enough of the startup life, and you guys wanted to do it again and re immerse yourself into it. So talk to me about the inspiration behind Canonical and why you wanted to get into this space that you're playing in.

Tom Shapland:

Adrian, I'd like to hear it from you. Why did you want to start a company with.

Demetrios:

Yeah, you're probably very scarred, right? Like, you. You had a lot of scar tissue. I can imagine. And. And so now, going back to the starting line and deciding to do it all over again, what was the impotence for that?

Adrian Cowham:

I'm a glutton for punishment, I guess. No, it's.

Demetrios:

You should read Tom's paper.

Tom Shapland:

It's a barn burner.

Adrian Cowham:

Man. I just love the drive, the push, the excitement, the environment. Like, you know, it's just like a personal philosophy for me to fight complacency. And so we always gotta. We always gotta grow. We gotta keep pushing. And so Tom was like, hey, you want to start a new company? I was like, I've actually never done that before. Let's go.

Adrian Cowham:

And so we. We didn't know what we were gonna build. I was like, is this what people do? They just start a company without knowing what they're gonna do? He's like, oh, people do all the time. Don't worry about it. And so we just kind of took a look at the landscape and AI was just coming out. I think chat GPT3 just got released. Well, AI has been around for a while, but, like, it was really starting to take off and we're playing around with that mess, around with a couple ideas and just started, like, just goofing off and we've. It took us a little while.

Adrian Cowham:

It took us a few months to kind of. To get to where we're at. But, yeah, I mean, that's. That's why. I mean, just, you know, keep going.

Demetrios:

Excellent.

Tom Shapland:

I think for me, part of the philosophy, there's. There's some driving forces to it. Like, one was I wanted to work with Adrian again. This is like the. In the show notes for this, this episode. It's like, hey, do you want to Listen to a bromance. It's like. And talk about voice AI and this is the episode for you.

Tom Shapland:

but also, you know, I think one of the things that you realize when you. When you build a company and. And get to a successful exit with it is you. You. You viscerally experience how fleeting victory is. You know, it's. It's there. And this thing that you're aiming to achieve for years and years happened, and then it's gone, and you're like, well, wait a minute.

Tom Shapland:

That wasn't really that great. The victory wasn't all that great. It was the journey that was so much fun. And it's the journey that really matters. And the journey. I wanted to go on that journey again. I wanted to experience getting to build a company again and hopefully this time enjoy it more. Because last time, it was my first time and it was really stressful.

Tom Shapland:

Yeah. And I let that stress get under my skin and. And then looking back, I thought, like, oh, that was actually really cool. I should have been more open to what a neat experience that was. I'm hoping to experience that this time, going through it.

Demetrios:

And how did you all land in the voice space? How did you see that there was something here? And recognize. All right, let's pull on this string a little bit more. I know, Adrian, you mentioned AI was hot, so that probably gave you a bit of a compass. But where was it that made you say, there's a product here?

Adrian Cowham:

Right? And so we started build. Working with. Trying to work with publishers, book publishers. And we had this idea of like, okay, imagine sitting down and you're about to read your favorite piece of reference material. I'm a nerd. I love programming languages and I read programming books, and we thought it was super cool. You know, irrespective of the reference material. You get to sit down and then you have a companion.

Adrian Cowham:

And you can use this AI to chat with the book while you're reading it, ask questions and get examples, ask clarifying questions. This, that, and the other thing. And so we built this and we were working with publishers, and we were really trying to improve latency and cost. And so we built this caching layer, the semantic caching layer, and that worked. And, you know, we were talking. Publishers were moving really slow, and so we couldn't get traction. And so we were thinking about moving on to something else. And we're like, you know, those caching layers actually kind of legit.

Adrian Cowham:

Let's see if it can stand on its own. And we did, and we got some initial Users signed up and using it kind of at scale, which was pretty scary. And that's how we came into the Voice AI space, because at that point, at that time, latency was a huge deal, and a lot of these guys were, like, willing to try anything to improve latency. And so that's how we got into Voice AI. And then from there, talking to the community and just keeping our ears open, we're like, dude, nobody's doing evals and assessments. Like, nobody knows what's going on inside these calls. They're a black box. Yeah, maybe they listen to one or two calls a day, but.

Adrian Cowham:

And, you know, customers are calling to complaining, and then they're actually, like, really listening. Right. And so we're like, what if we developed a, you know, a metrics and analytics platform that does it for you automatically? And then you can just. You have all the. All the metrics tracked over time. You can see what's happening. You can add your own custom metrics. And so that's how we got into the Voice AI space.

Tom Shapland:

The Voice AI space. I feel so lucky that we found it. It's just such a. There's two things I really like about Voice AI. One is, and this is, like, the primary reason that I love the Voice AI space is the community. There's a real sense of community in the Voice AI world. You can't go on Twitter and reach out to somebody who's building with LLMs and be like, hey, what are you building? They're like, there's so many of us that building the LLM space that nobody will, like, get back to you. And it's.

Tom Shapland:

There's no sense of, like, affiliation or camaraderie, but if you see something cool that somebody's building with voice, you can just message them and be like, hey, what's that about? How did you build that? And you get to meet these cool people and we're all figuring out the space together. It's. It really feels like a community. I've made a lot of friends through it. This has been the part of the journey that I've really enjoyed this time, is being a part of the Voice AI community. And then the second reason that I think Voice AI is a great space to be in and why I'm so lucky that we kind of happened upon it, is that it's clearly about to explode. Like, it's already growing really fast. And I think we're still just in the early days of how it's going to grow and what we're going to see come out of voice AI.

Tom Shapland:

So it just, it feels like we're getting the best of both worlds right now, like community. And we're in early for getting to ride a bigger wave and get to experience a cultural phenomenon of this, of voice AI eating the world.

Adrian Cowham:

Yeah, it's almost like, if I can just add on to that with respect to what Tom said about community, it's been a huge surprise to me at how, like, kind of warm and helpful everybody, everybody has been early on. The guys at vapi, Jordan and Akhil were just so helpful to us and like, gave us a chance and worked with us. And we're eternally grateful to those guys into all the other co founders that we've met in this space. And like Tom said, you know, definitely there's that sense of community and it's like, oh, cool. We kind of feel like we belong here. Whether or not that's the truth, that's another. That's a different story.

Demetrios:

But it's almost like the. And I don't know because I don't have a Harley Davidson, but I've heard when you have a Harley, you are instantly part of a club and you can give like a hand signal as you ride by somebody else that has a Harley on the highway.

Adrian Cowham:

Yeah, that's totally true. Tom used to own a motorcycle and he said I was. As soon as he started riding it around town, he was always just throwing the peace sign to every other motorcycle and they would throw it back and he's like, dude, it's so cool.

Demetrios:

Instantly part of a community. They don't screen the members, but I guess that doesn't matter at the end of the day. So now you talk to me a bit about the product itself, but I want to know more, really around what you've seen people get the most value of and how you've seen people engaging with Canonical AI and the product you've built and how that's driven your design decisions to build more.

Adrian Cowham:

Right. So what we've seen is we have customers logging in and they're taking a look at the metrics they want to. So no, okay, I have this. Here are my assistants. Let me look at yesterday's data and let me see, you know, how often was this assistant able to meet its objective?

Demetrios:

And you're talking actual. You're talking AI voice assistant.

Adrian Cowham:

Yes, AI Voice assistant. Like, how often were they able to meet the objective? Yesterday.

Demetrios:

And the customer is somebody that has a voice AI assistant app that they've built or it is someone that is an end user. Maybe they have a, like a phone studio or a number that you can dial. Who are, who's the customer?

Tom Shapland:

Yeah. Let's zoom out and talk about what's happening in the voice AI world before diving into how people are using our product and how we're solving problems for them. There are three ways that we're seeing interactive real time voice AI experiences. Catch on. Those three categories are first, existing call volume. So you can think of anytime somebody calls a business that's an example of an existing call that a voice AI assistant or voice AI agent can take on. And examples of that are things like people calling a dental practice to book a meeting. Or another example of it might be something like, I don't know, there's so many like a in after hours agent at a plumbing company that answers the phone when you know it's after hours and nobody's, it's not during normal work hours.

Tom Shapland:

So that's existing call volume. Then there's this other category that's really interesting, which is like call volume that's new and enabled by those sorts of voice AI infrastructure that Deepgram has built. It's really cool to see the creative things that people are coming up with. One of my favorites I always like to talk about is my friend Andrew D'Souza's Borde app. So Bordy is a social networker. It's a voice AI assistant that you call and tell it about your interests and it finds people that it is talked to that have similar interests. Um, and generally it's oriented towards professionals meeting other professionals and it's like a way of scaling up referrals. It's a really neat app.

Tom Shapland:

So that, that's a great example of like voice AI is also there's this whole category around new call volume that is enabled by this, this entirely new medium. And then there's a third category that I think of for interactive real time voice AI and that's voice user experiences. I think we're going to see more and more examples of people using voice first as a way to interact with the computer in a way that they, they previously couldn't. Because things like ASR weren't good enough until the stuff that you guys have built at Deepgram came around. And so I just think like, look, I look back on like you know, the 90s and my dad carrying a briefcase and I look back, I'm like, what was in that briefcase? I'm like, why did people carry that? And I think my kids are going to do a similar thing very soon. Like they're they're in late childhood and they're going to be teenagers soon. They're going to be like, I remember growing up, you used to be like, tapping furiously with your thumbs on a phone. It was so bizarre.

Tom Shapland:

It was so like 2020s. And because we're going to instead have voice UI interfaces. So that, that's generally how I see voice AI eating the world.

Adrian Cowham:

It's.

Tom Shapland:

It's going to eat existing call volume. It's going to. It's going to generate all this new call volume and it's going to eat existing user interfaces and user experiences. We're going to. It's going to do away with a lot of keyboarding and clicking and tapping.

Demetrios:

Okay, so that being said, now you have the product itself that you're building, right? Like Canonical AI, and how does that look so that people can interface with it?

Tom Shapland:

Yeah. So the problem we're solving for people, or I should say it this way, we're solving a problem for the voice AI assistant developer. So people who are developing voice AI assistants to take on that existing call volume, you know, to do things like qualify leads, they'll build these voice AI assistants and they'll test them as much as they can, or test them even using sophisticated simulation methods before deploying them. But once they deploy them in the wild, they don't know what's happening in their calls. They're manually listening to calls, a sample of their calls every day, or they're waiting for their customers to complain and then listening to those calls. And it's just not. I mean, we've been building digital products for a long time now, since the 90s. Right.

Tom Shapland:

And even before that. And we know that you have to see what your users are doing with your product if you want to improve it. And like, manually inspecting a subset is not going to get you to the type of product quality that consumers expect. And that's the problem that we're solving for people. And the way we're solving that problem for people is we're. We giving them.

Demetrios:

Can I. Can I interrupt you real fast? Because there is a little story that I want to tell you about this. Last week I was talking to a friend about how we encounter voice agents in the wild. And we instantly try and get a human on the phone as fast as possible just because of the. And so, so that's because of all this scar tissue that we have or these bad interactions that we've had. It's. We don't necessarily give ourselves the patience to see if the voice agent is good yet. And so we were laughing because he asked me, how do you, what's your like go to word to get the human on the phone? And I just repeat human, human.

Demetrios:

And he said, oh no, you know what works for me? He goes, I yelled help, help. I'm drowning. He says, humans come on real quick when you yell help. So anyway, that's a little bit of a tangent.

Tom Shapland:

Yeah.

Adrian Cowham:

You don't know. I'm actually like going through some of our customers transcripts. You don't. You'd be surprised. Like it's surprising how many times people do that. Like for the folks that know it's a voice I voice AI agent, they're like person, person, person. Or they'd be like emergency, emergency, emergency. So it's, it's pretty interesting.

Tom Shapland:

Well, that's a really, I mean I think that the, there's a really interesting point to be made about the trajectory of this technology, where it's coming from and where it's going there right now. Consumers are, we've been trained on these terrible IVR and rule based systems. It's, there's only a small subset of people out there who have called, who have made phone calls and actually encountered a voice AI agent in the wild. So they haven't had that magical experience yet of experiencing what it's like to talk to a voice AI that's backed by an LLM. And for as many times as we, I mean, as much fun it is to talk about like, you know, ways to like get around a voice AI agent right now, like we also see a lot of calls where people are like, hey, I want to talk to a representative. And then the representative, the voice AI will say like, sure, I'll connect you to a representative. But first can you tell me who you want to, what this is about so I can like pass that information along. And then the caller will say what it's about.

Tom Shapland:

And then like right there, the voice AI agent will just solve their problem and the person will be like, oh, that was great, thank you. And the call, the call ends. And, and we're going to see more and more experiences. Consumers are going to have more and more experiences like that with calling random businesses that they need to do business with or I think with generative Siri on their, on their phones and they're going to come to just expect voice AI agents. I think we're going to see people saying not representative, help. I'm drowning, I'm drowning. I need to talk to a representative. They're going to say, you Know, connect me with a voice AI agent.

Tom Shapland:

Like, I don't want to talk to. This person's not getting this. Like, just give me my voice. AI Chin.

Adrian Cowham:

Yeah, what? One of the things that we've noticed is with respect to humans asking to talk to a representative is sometimes they immediately know it's an agent and they just want to get a human on the phone. But other times they discover it's an agent because the agent barfed, the agent repeated itself or just said something nonsensical. And then they realize, oh, I'm talking to a robot, I want to talk to a human. You know, let get me to a human. And so, you know, we can identify that and we can help developers fix it. They can just add a custom, add a metric that says, hey, let me know when the human asks to speak to a representative. And then you can track that over time and then you can improve your prompt and you can improve your product and then, you know, leads to revenue.

Demetrios:

So, so that is basically you. You're able to track custom asks that are said, or you're able to define within Canonical what you want to track. Is that how I'm understanding it?

Tom Shapland:

Yeah. Let's talk a little bit more about how we're solving this problem for our customers. There's two things that our product provides. The first is a map that shows our users, the voice AI agent developers, what's happening in their calls. We map out the stages of every single call. And what you end up having is you can look at this map that shows you all of your calls for the previous day for a particular assistant. And you'll see something like 90% of the calls are going down this happy path and everything's working just like you prompted it to work. All the third party integrations are working, but then you'll see these branches that come off those happy paths where like, something weird is happening.

Tom Shapland:

And you can't ahead of time ask questions about what that is, and you can't really simulate what that could be because humans are so unpredictable. You really have to see what the agent is doing in the wild to find these sort of edge cases that really need to be fixed. So that's one way that we solve the problem of helping voice AI developers know what's going on in their calls. We map their calls for them and give them an easy way to surface the calls that need their attention. Um, then the other way that. Sorry, go ahead.

Demetrios:

That's like, that's almost like when I'm looking at my Google Analytics and I See page views and I can see the user flow. They look at the homepage and then they click onto the blog, and then they click onto this blog and then maybe they bounce or they go back to look at the blog and look at something else. And so you're mapping that out, but in the voice flow, so the user is asking for something. You get an integration. That is the agent or the agent calls on some kind of a tool, it is able to answer the question. And then at some point you see, hey, there's some funky behavior going on here. What is this? And you can click into it and get the transcript or get the recording, I presume, and then say, oh, it didn't actually do what the person was asking for or it gave them something completely wrong.

Tom Shapland:

Yeah, that's exactly right. I love that you brought it back to websites because right now we live in a world where we're used to interacting with digital content through websites. And I think that's going to change. And we're going to see more and more interactions with voice AI agents and voice AI assistants are basically just websites. And right now the state of voice AI is that there is no Google Analytics. There is no way for people to know what's happening inside their voice AI assistants. And we're building that for people because clearly that's needed. Right.

Tom Shapland:

We know from our experience building the Internet that, like, we need to have product analytics and what's going on when users are interacting with your digital products. And that's what we're building for voice AI assistance. There's a second part to how we are helping voice AI developers. So the first part is the call mapping the caller journey mapping. The second part is the stuff that Adrian was just talking about, where we provide audio metrics that are specific to the challenges of voice AI agents. Things like latency and interruptions and that sort of thing.

Demetrios:

Oh, yeah.

Tom Shapland:

We also give the developer the ability to ask custom things about a particular agent. Like, here's a question I have. And every time this agent talks to a caller, if this question is true, show me that call so that I can see what's happening in that call.

Demetrios:

It's almost like a flag. Like, hey, if this question comes up, we gotta see it, because there's something there to make someone ask that question.

Tom Shapland:

Yeah, that's right.

Demetrios:

So you highlight it and what are the other levels? I know there's. There's the latency level. What are other levels that builders can look at and monitor? And this makes me think of something like a data Dog or an observability platform where you're looking at the system in that level. So there's a fascinating thing happening here because you're building the voice interaction and you really have to look at things on the level of understanding and ability to execute tasks. And that's one abstraction layer almost, or that's like one level. And then as you said, there's another level that's happening, which is the system and the actual bytes and bits, as you referred to them earlier, Adrian. So what are some things that you have in that layer?

Adrian Cowham:

Right, Good question. So the way I think about is every phone call, every conversation that happens with a voice AI assistant represents infinite possibilities. 100% it's going to be infinite possibilities. You don't know what's going to happen. You don't know the quality of the audio of the phone line. You don't know if they're calling from a laptop, laptop, their web browser drive through. How far is the mic? Is there echo, is there a crosstalk? Is he talking to the passenger about what he wants to eat? So every call is a system of infinite possibilities. And so what we try to do is approach it from two angles.

Adrian Cowham:

The signal processing, the highly technical stuff like what you were just referring to, and also the conversational part. So with the signal processing data, I mean we go pretty deep down the stack. We can tell you how much with the SNRs, how much background noise there is, we can tell you the pitch of either caller because sometimes what we've experienced, sometimes the assistant pitch is just. Is completely shrill. And so people hang up immediately. We don't want to listen to that. We can we calculate the amount of silence during a user turn, which is indicative of something of the assistant failing. So we have interruptions both if the assistant is interrupting the user and vice versa.

Adrian Cowham:

Oh, interesting. Yeah, we go pretty deep down stack on signal processing. There's other metrics that we capture as well.

Demetrios:

Well, I love this, fellas. This is really cool to think about. And you have done something that is very new to me, but it makes complete sense. It's very obvious that something like this is needed. We need it in all software development and especially when it comes to using AI, we very much need this evals and observability and just the ability to monitor what's going on. And so it is very logical that we need it more than ever in voice, especially because we want to get to those magical experiences with voice AI and not have people instantly asking for a human or hang up.

Adrian Cowham:

I mean we look at some of these call flows. You had 100 calls today, and 50. 50% of them went from start immediately to the end because they hung up. Because they knew it was a robot and they hung up.

Demetrios:

That's the equivalent of bounce rate. Yeah.

Adrian Cowham:

100%. Yep.

Demetrios:

Wow.