theCUBE + NYSE Wired: Physical AI & Robotics Leaders QA2 | Brian Peterson, Dialpad

Clips
More from theCUBE + NYSE Wired: Physical AI & Robotics Leaders QA2

Brian Peterson

Co-Founder & CTO

Dialpad

play_circle_outline Brian Peterson: Transforming Business Communications at Dialpad with Integrated Voice, Video, Messaging, and AI Technology

play_circle_outline Deployment of a modern, cloud-based communication platform adaptable for hybrid work environments.

play_circle_outline Dialpad's AI journey began eight years ago with a focus on customer insights.

play_circle_outline Integration with various CRM systems enhances real-time communication and customer interaction capabilities.

play_circle_outline Transforming Communication: Embracing Diverse Modalities and AI-Driven Insights for Enhanced Customer Engagement at Dialpad

Info
Transcript

Brian Peterson, Dialpad

Brian Peterson

Co-Founder & CTO Dialpad

Brian Peterson, co-founder and chief technology officer of Dialpad Inc., joins theCUBE’s Dave Vellante during theCUBE + NYSE Wired: Robotics & AI Infrastructure Leaders 2025 event to explore how AI is transforming business communications. The conversation highlights Dialpad’s unique journey from its roots in Google Voice to a deeply integrated AI platform.

Peterson shares how Dialpad leverages AI to extract insights from more than 10 billion minutes of real-time conversation data, reshaping customer engagement in sales and support. Unlike bolt-on sol... Read more

explore Keep Exploring

What motivated the founding team to start the company, and what were their initial observations and goals during its early days? add

What has the company developed to enhance customer communications, and what makes their platform distinct in the current market? add

What has been the experience of a company heavily involved in AI communications over the past eight years? add

What features and integrations does the AI-powered communication platform offer to enhance user experience and facilitate customer satisfaction? add

What is the vision behind the development of the communication platform mentioned in the text? add

bolt Powered by CUBE AI

Brian Peterson, Dialpad

search

Dave Vellante

>> Hi, everybody. Welcome back to Palo Alto. My name is Dave Vellante, and we're here with theCUBE and NYSE Wired: Robotics & AI Infrastructure Leaders series. We're going to take a little detour from the whole robotics theme, and we're really happy to have Brian Peterson in here. He's the co-founder and CTO of Dialpad. Dialpad integrates voice, video, messaging, and of course brings in AI. Brian, thanks for coming in.

Brian Peterson

>> Yeah. Thanks for having me.

Dave Vellante

>> So why did you start the company? Let's go back to the early days. What did you see? And let's talk about how things have changed. I mean, you guys have raised a lot of money, you're moving markets.

Brian Peterson

>> Yeah. So our team's background was, we were the Google Voice team. So we were back at Google around 2010 doing Google Voice to 10 million plus consumer users, not necessarily business users. Then at the time we thought this technology, advanced communications for cloud, all the different advanced routing that it made so much sense for businesses. So we said, "Hey, we should take what we learned there, build what we think is the next generation communications platform." And when we did it, we thought there was two pieces that were the most important piece to It was, one, it should have everything you need for communications. Really video, audio, texting, omnichannel, all of it should run through the same thing because communications aren't just one form. They are omnichannel. They are multimodal or multi-medium. And then on top of that, it was very obvious early on, and we got into AI about eight years ago that AI just makes too much sense and that's where the world was going. And you have all this data with communications, but you can't do anything with it. So for example, we have over 10 billion minutes of just 100% relevant sales and support conversations that a lot of insights we're turning into live coaching, we're learning off of those. We're figuring out what customers like, what they don't like. That in combination with the communication platform lets us do amazing things, even real-time AI.

Dave Vellante

>> Interesting. So you come from the consumer world where you learn that scale. I mean, that's where most of the innovation starts usually in the consumer world, and then you bring that to the enterprise. I want to talk about your data because that seems to me a real competitive advantage. But before we get there, what was it like going from that consumer world to the enterprise world? What was that challenge? What were the requirements that you had to really understand from a customer perspective?

Brian Peterson

>> Yeah. We had a good advantage because when we were at Google, being at Google Workspace, which is what it's called now, was becoming a big priority for Google. So we already were overlapping a lot with B2B stuff with Google Workspace. And because we had so much overlap, we could see every application of this consumer product and how it made sense for businesses. I wouldn't say it's, this always works this way. You don't just go from a consumer app to a business app, but it's just, oh my gosh, it just fit. It just so happened that the thing we were building for a consumer, which is not always the case in B2B, was the writing was on the wall, this is perfect for businesses and they were struggling for it. They wanted the same functionality, and we kept hearing that at Google because we built this amazing consumer product and they've got all these business users. They're like, "I want that for my enterprise." So for us, we're like, "Let's just take the same formula." So we were super lucky that we could just take that same formula and apply it. Then once we knew all the extra data we need the insights. I hate to say, I know sometimes you should say, "We did all this research," but we were pretty lucky and it was obvious, but it's not like that for everyone. You could imagine consumer.

Dave Vellante

>> Yeah, totally. So you have 10 billion minutes, all this data, so how did you get there? What were the use cases that got you there? And you mentioned know sales support and live coaching. What was the journey to get to these?

Brian Peterson

>> Yeah. Well, the first thing is we built the most modern communication platform, like customer communications platform. We think in the world we're in over 50 countries with customers. We're at scale. I think I said before this millions of conversations a day, our main product, Dialpad has been out over 10 years now. So it is something that we've grown into, but now obviously you can imagine after COVID and stuff is exploding. So we went with build the best communication platform and something that was pure cloud, super easy to onboard, deploy anywhere, which became a big deal. Obviously with hybrid, we're going against a lot of companies who were mostly desk phone and you being at your desk, we knew that Wi-Fi and internet and your computer was going to get better. So we built Pure Cloud. So that already differentiated ourselves to the point where we've been able to get to the scale. But when we got heavy into AI about eight years ago, it's just been a constant journey as people buy us... In the beginning, mostly they bought us AI native communications platform. So they already were buying us not just for the communications, they were buying us for the AI too, which gave them the confidence to let us use their data to get better because they saw the vision of where these communications was going. And then over time with just again, that growth and word of mouth and we've put a lot of emphasis on user experience and super easy to use, over time they started using the AI more. And one example is you'll see with a lot of contact centers out there or other communications platforms, they added AI maybe in the last two years. Well, if you look at their adoption, it's really low because it's bolted on and you don't really know what it does. And the ROI is not really there yet. Well, we've had it for eight years. One of our most used features is real-time assist, where you automatically detect the question for your sales team, for your support team and give the response. Well, that turns a junior agent into a senior agent on week one, and it's accurate. So we do manager intervention. If a call's going wrong, we notify the supervisor and then they can get and see the live call happening and jump in and whisper. This was something that they saw the benefit and the return on investment on from day one. So it's just grown and we've been lucky to have obviously as much usage and customers as we have.

Dave Vellante

>> So Brian, to your point, I mean a lot of unified comms vendors do bolt on AI. We've seen that in the last couple of years. And of course they'll say, "We've been doing AI forever." But I'm interested in how you designed Dialpad's AI capabilities eight years ago, how you designed it in, and maybe you could talk a little bit about the architecture?

Brian Peterson

>> Yeah. So unlike a lot of the traditional communications providers, because we knew it was at the beginning, we had all the different modalities, so video, audio, messaging, but we also knew that we needed to get access to that data to give them what the value they need for automation, for coaching, for insights. So we built our own AI team, we built our own NLP, natural language processing team, our own machine learning team. We have about 30 PhDs in applied science and AI. So we even do our own transcription, which, and most people would say, "You're crazy. Why would you do your own transcription?"

Dave Vellante

>> It's a commodity. Why are you doing that?

Brian Peterson

>> Why would you do that?

Dave Vellante

>> Why are you doing that? Something that-

Brian Peterson

>> It turns out, one, the models for something like even the generative AI models are built for the whole World Wide Web, of good and bad, of hallucinations, facts and fake news. And we knew that we had this moat of 10 billion, not just any data, but 100% relevant data. These are real support conversations. These are real sales conversations. These are verticalized. We know which one of our customers are in the legal space, healthcare, retail. So we can customize all of this. And we knew we had to do it because the only way you can really train on all of this data is if you own it. And it's been a benefit for security reasons. You go to enterprise, they already trust us with their most sensitive data to know that most of their AI is in-house too, running in the same platform. That's been a big advantage. Cost has been a big advantage. Availability has been a big advantage. It's getting better. But in the last four years, even if you wanted to run just summaries, generative AI summaries on your conversations, you wouldn't be able to do it at our scale. There was not enough bandwidth, there was not enough availability. And even if you could, it'd be so expensive, you'd have to charge $100 a month just for AI. There would be a showstopper. So when we are able to build our own, we built our own LLM, we call DialpadGPT. We've been able to run over 400 million generative AI summaries in less than two years, which I've been claiming is the world record. I don't know if it is, but no one's called me out on it on LinkedIn and it's mainly because we had availability. We had volume, and we had availability at a good cost. And it turns out it's also more accurate because if you train all your AI, even the transcription on exactly certain conversations, you get better out. So better data in and better data out. So the future is moving to these small language models that are more verticalized, and that's where we are right there, front and center on, because we have the team who knows how to scale it and to customize it. So we knew we had to do it.

Dave Vellante

>> So very high fidelity data. And you've got this verticalized data. So you're building vertical small language models. Is it based on one mainspring model or do you have a collection of models?

Brian Peterson

>> Yeah. So it starts with a bunch of definite open source stuff too, but with our own recipes, we call them, with our own training of our data on those open source models, which you couldn't do at the volume we need to do, if we just used a ChatGPT or anything like that. We need full access to be able to train and customize the models. So it started with that, and that's what DialpadGPT is. It's a common model used for our type of data though. So it is very specialized, not to the World Wide Web of whatever you might want to ask it, but it's a large language model specialized on sales support conversations, and recruiting conversations. And then that's our first step and what we're moving towards now, what we're investing in is we're going down to the use case even more so support, sales, recruiting as just examples, and then the next phase will become even more small language models focused, which is verticalized, so healthcare, legal, that sort of thing. And that's even what Nvidia says the future is. If you want to have accurate age agentic AI even, you need to have a very specialized AI. And the only way to do that is if you have the whole team.

Dave Vellante

>> Indeed. Okay. So eight years ago you started on your AI journey, and I'm presuming it was deterministic AI. How did you weave in layer in probabilistic AI? How do you use that and how do the two relate to each other?

Brian Peterson

>> Yeah. I mean, the future basically, I think is what you're asking is also where's it going? How do we know and how do we adapt? Agentic AI become the hot topic. If anyone's been following anything, it goes from ChatGPT and generative AI to all of a sudden now how do we use this? So it goes to now the new hotness is a agentic AI, and that is, we believe strongly in that too. There's a reason why there's a lot of investment in it is that this generative AI is so special, but no one knew what to do with it yet other than to maybe summarize some things and write some emails. Now, we're realizing that this stuff is going to be really amazing at automating things, and that's the next frontier. We just announced our agentic AI platform that's coming in the fall that's going to let these customer communications things be automated. And by that, we know that, again, back to the better together of we own the entire communication platform coming from Google Voice and having all that expertise on scale and being worldwide, but we have the AI pipe, the connective and native to it, we then can have humans and bots training themselves. That's the future. And I'm not a believer that the entire world's going to be automated. I'm not one of those people who says 100% of customer support interactions are going to be automated. I don't see that happening in our experience. I think though it's going to go from 0 or 5% to a very large percentage are automated, but they're all easy stuff. But the bots know also the conversations happening with customers, so they know where they have answers or don't have answers, and you get feedback. Then these humans on the agent side also have really good conversations, and they're usually better experts, especially the more senior ones. They then can train the bots or we can find gaps where, hey, you're doing all these things that are very tedious, that are very basic, that you don't need your really good human agents to be doing. It's wasting their time. They should be working with important customers instead of doing basic stuff like change my address. Well, that's where we can determine that stuff and automatically have it learn from each other.

Dave Vellante

>> And you've got the human in the loop.

Brian Peterson

>> The human in the loop is a big thing that I think a lot of the agentic startups are finding is a gap.

Dave Vellante

>> And customers are demanding that.

Brian Peterson

>> They have the gap. And it goes from bot to human back to bot to human, and-

Dave Vellante

>> Your agents can learn from those exceptions and the reasoning traces of humans. How does that work?

Brian Peterson

>> Yeah. It's all reinforcement learning and because you have that data, the perfect gold data, and you have real conversations happening, our future is even down to the individual business.

Dave Vellante

>> You know what the right answer is?

Brian Peterson

>> It's down to the individual business. We already know the right answer because that answer has been given 1,000 times in the last two years to your call center by, and we know that it's the right answer because one, it's repeated a lot. And because we have other things, like we have our own proprietary AI CSAT model. So it's customer satisfaction with 90% accuracy on every call. Well, we know that when this question is answered, this particular way that the customer's happier and when it's answered this way, we know they're not as happy. They don't like that answer. So now again, it's a loop.

Dave Vellante

>> Real time as well-

Brian Peterson

>> In real time, it then even tells them this is not the right answer.

Dave Vellante

>> Don't say this.

Brian Peterson

>> Yeah. So it automatically generates to even the human, "Hey, here's what you should answer for here, because this has the best customer satisfaction results."

Dave Vellante

>> I presume you integrate with tooling upstream CRM. Maybe you could talk about that.

Brian Peterson

>> Yeah. So we consider ourselves an AI powered communication platform. So we integrate with everything in and out. So we have hundreds of different integrations to every single CRM to every single data lake there is. We have the ability to feed data in as well for rich caller ID. We have an integrated sidebar. We have an enterprise customers building custom components in our sidebar that can automatically enhance the active conversation. So that is a big part, and that's a big part of the age agentic future is age agentic is all about taking action. Well, those actions, a lot of them will be in your own product, but if you're a SaaS product, you are going to have to interact with a lot of other SaaS products. So the future of this is taking natural language and saying, "I want to do this thing," and having it automatically be able to interact with a third party system and take the action correctly and be able to take that action. And that is what is moving very quickly with this MCP world is people are still figuring it out. The technology is really good, but you're going to find out that there's a lot of extra pieces around this AI that you need to invest in. And that's what we've been doing, which is it's a lot of curation. It's a lot of checks and balances. There's a lot of tools around this amazing generative AI and making sure that it can't screw up. You can't say authentication has to be very specific. You need to authenticate the person. Then you need to know what do they have access to, and then you need to know, hey, I want to schedule my appointment. You can't mess that up. It has to work. And I think people are finding that it's very difficult and a lot of hands-on the bolted-on solutions are lots of professional services for two months of let me tweak this for you and this for you and this for you. And still it will make mistakes.

Dave Vellante

>> Because everything changes.

Brian Peterson

>> Because everything changes. And that's why we're putting so much effort into this platform for agentic AI that can be very controlled, very secure and reliable. So that's where things are going too.

Dave Vellante

>> And it comes back to the data and you've obviously... Well, you've got your own LLMs, you've got your own transcription, presume you've got your own analytics as well.

Brian Peterson

>> Yes. Big data. We have unlimited data. We've always said voice is the last offline data set, and that is the AI future. There's so much data and you hear it, the data becomes the moat. The data is where the value becomes, and the world is out of data. If you want to look at these third-party models, they're out of it. They can't find more data. And the data they have isn't good data.

Dave Vellante

>> Yeah. They're using synthetic data, but that's not enterprise data. Enterprise data is real. Not synthetic.

Brian Peterson

>> Exactly. And it's very specialized. That is the future. You can't use a ChatGPT on a sales conversation and have it say things correctly because it's going to use the data it has over again, the entire World Wide Web of who knows what.

Dave Vellante

>> Well, it's going to give you different answers every time too.

Brian Peterson

>> Or different answers.

Dave Vellante

>> That's killer based on what you just described. It has to be this answer, not that answer or not some derivative of the right answer. It's got to be the right answer.

Brian Peterson

>> Well, and imagine you take that LLM and put it into the face of your end customer, how scary is that? Because now you're like... I think there's some story about an LLM sold someone a truck for free or something they turned on, so-

Dave Vellante

>> They gamed it.

Brian Peterson

>> Yeah, they gamed it. Exactly. So you have to be really careful with the checks and balances of these systems. It is one small piece is the LLM. The rest of it is all this redundancy and self-training around it and controls, checks and balances for sure on all of it.

Dave Vellante

>> So it's data and of course engineering connecting to all these systems.

Brian Peterson

>> Yeah. And it's obviously very complicated. We connect. We are in communications, so almost every single product needs to connect to us.

Dave Vellante

>> Well, Brian, what's next for you guys? What are the things that observers should pay attention to in the future of Dialpad?

Brian Peterson

>> Our dream is when we started is that communications was broken, especially customer communications, and the future of communications is all kinds of modalities. It's not just a call, it's not just a text. There's more and more ways to communicate than ever, and those conversations are the most important things you have for your business. So we feel that, and this is where all our investment is going, is have the best communication platform in the world, which we think we have because of our background, but annotate it with the best AI tailored exactly to that communication data, which gives us real time, which gives us post-call insights, gives us live coaching that no one else can do right now. We're just doubling down on that. But the next phase, like I said, is in our launch in October, is going to be agentic AI explosion, and everyone's going for it. There is still a lot of too much manual work going on. I'm not a person who thinks AI is going to take over everyone's job. I don't think that's going to be the case. I think it's going to be similar to cloud. That didn't mean less jobs. It just means more innovation. Well, people expect that they shouldn't be doing these mundane tasks, these simple things. And I think agent AI is going to take that 5% to 80% even maybe, but you're still going to need the 20% human.

Dave Vellante

>> Different jobs.

Brian Peterson

>> Different jobs. Spend time on the focus on the people that matter, the customers that matter with real people. Let the bots automate the rest.

Dave Vellante

>> Brian Peterson, congratulations and thanks for coming on theCUBE, really appreciate it.

Brian Peterson

>> Yeah. Loves being here. Yeah. Thanks for having me.

Dave Vellante

>> You're very welcome. All right. Keep it right there, everybody. This is Dave Vellante for John Furrier and the entire CUBE team, theCUBE plus NYSE Wired: Robotics & AI Infrastructure Leaders. We'll be right back right after this short break.