Adnon Dow: And nobody's talking about that. That's the crazy thing. Everybody's talking about, oh, voice AI, voice AI. You wouldn't have voice AI without having an intelligent edge and a secure network and a robust network to process it. 'Cause that's the thing about large language models. As good as the agents, voice agents are getting every day, if there's any latency, if there's a pause, or if there's a disruption in the communication, or if there's background noise, it completely throws the agent off. I mean, I've noticed it all the time. Nobody talks about it.

Tom Arbuthnot: Welcome back to the show. This week, we are talking about AI in the operator network. We talked to Adnon, who takes us through what Pure IP are doing with AI at the operator level, what that means for customers, how they can leverage AI in the carrier network, work across different UCaaS vendors across different solutions, and how that works with their line of business solutions and also their UCaaS and their contact center. Really interesting conversation with Adnon. Many thanks for him jumping on, and also many thanks to Pure IP for all their support to the community. On with the show. Hey, everybody. Welcome back to the show. Really excited to have this conversation. Adnon's been talking to me a lot behind the scenes about what we've been doing at Pure IP with AI, but this is a chance to get some of it on the record. We'll see how much we can get out of you, Adnon. But do you just wanna start off by introducing yourself and your role?

Adnon Dow: Yeah, no, absolutely. So good morning, good afternoon, everybody. Thanks for having me, Tom. Adnon Dow. I'm the Chief Strategy and Product Officer, for BCM. I've been with them now almost, a year, a year in August, coming up. So, I'm responsible for our strategy, our go-to-market, and really, innovation and how we are transforming and pivoting to address, the trends that are happening, today, across the industry with AI, network infrastructure, and some of the services around communications, collaboration, and data services, which are our core services that we offer.

Tom Arbuthnot: Awesome. And, and for a bit of history for people, you've got a, a history at some pretty big players, including, Bandwidth before this.

Adnon Dow: Yeah, absolutely. Gosh, been in the industry for longer than I wanna admit, as you can see my, my gray hairs, but ex-Cisco, Motorola, where I've done networking, infrastructure, satellite, cellular, and then in the last 10 years been heavily involved on the voice communications and collaboration. Most recently, Sandy and I both, our CEO, worked at, Bandwidth, doing very similar things in terms of helping, wholesalers there and aggregators set up their networks to address, communication and collaboration services.

Tom Arbuthnot: Awesome. And, this is gonna be fun 'cause I spend a lot of time obviously talking to different operators and people in the industry, and, there's kind of a, a greater or lesser, acceptance that AI is going to have a material impact, and the way people are working is changing. And, I feel like you're on, on the side of, yes, it's definitely gonna have an impact, and we need to do something about it. But maybe you can give me your take on, our space and, and AI's impact over the last few years.

Adnon Dow: You know, gosh, it seems like almost every company that whether they release their earnings or making some sort of media release of anything they're doing, it all has to do with AI. And, and I always... I say this often to my wife when we're kind of sitting around talking, when she tries to ask me what I do, 'cause she's still trying to figure it out. You know, she asks me these questions, and a lot of them really don't understand what AI is and, and what it's actually doing, but there's multiple layers. The first layer that you're seeing, at least in, in our industry, and a lot of others as well, is, is they're using AI to better monitor, analyze and control and manage the core network infrastructure they depend on, whether it's an enterprise or a carrier. They're using AI because AI is much more effective. It's on 24/7. It follows instructions very well, especially if it's structured machine learning, anomaly detection type of instructions to, to manage and monitor the efficacy and, and the throughput of the networks. So you see that at, at the, at the base layer of a lot of these infrastructures. The second is really around looking at databases and, making sure that through algorithms and different processes, that those databases are most utilized and available depending on the constituency that it's gonna serve. So for example, if I'm a, a salesperson, how can I extract all the data out of my CRM, make sense of it, and be able to visualize it so I can make business decisions to drive outcomes? And then the third one, which we see heavily, is creating AI to enable, your service offerings in the market to make them much more applicable to the industries that you're serving. So, for example, recording, transcription, voice intelligence to be able to route calls and deterministically, be able to guide those calls to access the most important information and be able to be compliant based on the industries that you're serving. So those are really the three layers, and depending on how you go about it, different companies are doing different things. We're dealing with all three layers as we innovate, we rationalize, optimize, and, and really, roll out the next generation, what I call the intelligent, edge for communications and collaboration across the distributed enterprise that we're really going after to serve.

Tom Arbuthnot: Nice. I wanna dig into that. But first, interesting kind of, there's a, a theme through those three areas, and it's a lot of we had access to the data before, but we didn't really have the capacity to make use of it. So in operations, you know, us running a global network, there's loads of logging, loads of data, but you've only got so much time to make sense of it, so that's where AI comes in. And then on the, the, the usage side scenarios of, like, data in the CRM or line of business apps and things, same thing, like recordings. We could always have the data, but the, the... It feels like there's a big turning point here where we can actually use that and turn it into business value rather than just data.

Adnon Dow: No, absolutely. And, you know, for us, let's talk about services. When you look at the traditional carriers and the service providers, what they're doing out in the market, they really... It's, all over-the-top type of offerings. What that really means is they're providing their data and voice infrastructure for nothing more than transport. So the SBCs on the voice network will receive the calls, will set up the sessions and processes, and it's really about minutes and sessions, and the number of sessions and, and the amount of time that you're on there. And that's how really they make their traffic. Some of them kind of over the top will then lay on applications. But those applications are usually delivered on the far end, of that, particular delivery of that call. So, for example, call recording or some sort of voice bots or some sort of embedded, capability around voice paths and SIP prioritization based on their services, whether it's a contact center or a UC platform. So it's really about capturing, calls and delivering calls based on just reliability and throughput. And what you're seeing in the next generation type of smart, intelligent infrastructure is where it's really coming in to a deterministic type of approach where decisions are being made. Calls are being routed, recordings, transcription, sentiment analysis, fraud detection, all types of services from a, a voice intelligence is enriched at the edge when that call comes in, so you're delivering an outcome and a service to the end user through open APIs, through session layers, and through those capabilities that are integrated into the intelligent edge. And this is the difference between the traditional industry approach to the next generation, what I believe is the intelligent SIP edge approach that we're taking and how we're pivoting to service, you know, everything from SIP trunking to the UCaaS platforms, to the CCaaS platforms, to the, the mobile voice, the, the SIP-based voice that's coming across, WebRTC, name it. All of those have one common denominator. People wanna communicate, and they wanna communicate with the utmost intelligence for that call path, and that's really what we're trying to define what we're building.

Tom Arbuthnot: And that's quite an important thing there, the, the at-the-edge thing because you, you could, you could hear another podcast where a carrier will talk about AI, but what they really mean is we deliver calls to Teams, Webex, Zoom, name your UCaaS. We deliver calls to contact center, and the contact center vendor then picks up and does clever stuff. But what you're saying here is we're baking that into our network through the core to the edge and, and giving that kind of as a menu of options to customers. So and, and it is kind of a not an either/or. It could be better together. It could be combinations of things. But, but we're doing it on our network and, and it can be agnostic of whether it turns up at classic PBX, a UCaaS, a contact center or any combination of those.

Adnon Dow: Absolutely. You hit it, you hit it on the button, and I would break it down into kinda that call journey. You have the originating call that hits the access network, and that originating call could be all those things you mentioned, everything from a desk phone down to some sort of AI agent or client. It goes to the access network, whether it's an SD-WAN network, whether it's a direct connection, whether it's SIP. It hits the SIP trunks, and from there it's routed through the, the network. And then it hits an intelligent front end. And that front end, you really... We're, we're leveraging our investments as around a front end based on AI, our voice intelligent AI capability, where it's deterministically trying to put intelligence on the front end to be able to look at the context of the call, the enrichment and how it can be enriched, and then routed and orchestrated across the multiple capabilities that sit on the back end that are native in our infrastructure and, and create intelligent routing policy all the way across. So SIP normalization, least cost routing, tenant policies, numbering plans and anything that may be implemented. And then you hit the services. Those are the media services, the streaming, the recording, the transcription, all that being done natively. Fraud detection, for example. The call scoring, right? Everything from STIR/SHAKEN all the way out to the challenges that you see with a lot of these failed fraud controlled calls and risk scoring on the validity of that call and that caller's ID number. And then once it goes through that cleansing process, we're delivering it as a service, as an outcome. And this is where it's different. And then we're observing that entire quality of that call so that on top of it we're putting observability. So we're looking at the, the mean score, the MOS score, the, the lack of latency, the quality of service all the way through. Because as you know, you could drop a packet when you're sending an email, and it'll reassemble it and send it, and then when you get that email you don't know was there a packet dropped or not. You're either gonna get the email or you're not. You can't do that with voice. Even with AI and you're using it, the little bit of latency will impact how that AI agent is working, and it'll certainly, you know, impact that conversation. For example, you and I are talking, what if I drop every other word? That's what it'll sound like. And so this is why it's all done natively rather than over the top and we pass it through as a dumb call on a SIP trunk and hand it off to a PBX or a UCaaS platform or a CCaaS platform, and then that intelligence is then put on top of that platform. That's what we say over the top. And that's the difference between this next generation of that intelligent edge capability versus the traditional way that some of these carriers are doing it.

Tom Arbuthnot: Nice. And, if I was listening to this, I'd be like, well, so a, a Pure IP building their own LLMs, writing their own recorder, like, like doing their own data intelligence. Like, how do, how do we build, manage, and optimize those services for customers?

Adnon Dow: You know, a great question. So look, nowadays, why build it when you can partner or buy? Okay? There, there's just so many great companies out there that are doing a lot of these things. What we've done, Tom, is we've built a, an enterprise service bus, a middle layer that sits... You know, if you can imagine the three-layered kind of network architecture, you, you get our core network, our global core data routed network with the, voice SIP edges, the intelligent SIP edge. On top of that lays a microservices-based, containerized-based enterprise service bus, our integration layer. And northbound when we integrate with best-of-breed, best-in-class providers, whether it's recording, whether it's analytics, whether it's fraud detection, whether it's a UCaaS or a CCaaS, we power those and we end up integrating with them, and then we bring them in into the core network as opposed to just passing the call through and handing it off, let's say, to a Genesis or a Teams client. We enrich that data all the way through because we've integrated down at that enterprise service bus, and we're enriching that data and we're contextually providing deterministic routing based on the services that customer has subscribed to, and then we pass that call and that capability all the way through. So we don't build everything from the ground up. We've got the core network. We've got that integration service bus. There's some services that we build. Others we augment, and we're always trying to take advantage of the latest and greatest. And by being able to plug those in through open APIs, connectors, SDKs, we're able to enrich that data and be able to provide everything from real-time control to intelligent front end, to voice services and analytics all out to trust, security, and compliance in terms of ensuring those calls are not fraud calls, and that, that there's policy and governance in terms of being able to send that call through and record, for example, like in healthcare or finance, so that they're compliant in the way that they're receiving and, and positioning inbound or outbound calls.

Tom Arbuthnot: That's good to understand 'cause it's, like, there's, there's lots of options out there. I know you and the team spend a lot of time assessing different options and working with our customers and working out what the kind of the, the right portfolio of options are. And the, I guess from a customer point of view, we've assessed, chosen, and backed that outcome with that particular platform or provider.

Adnon Dow: Yeah. You know, it's... Listen, I call it outcome as a service at the end of the day. I mean, what are you trying to accomplish, right? As a, as a customer, we, we, you know, we pride ourselves on, on constantly seeking to meet the customers where they are, rather than have them come meet us where we are. And it's not about a rip and replace. It's about really looking at where they're going, providing a migration strategy, enriching their data, and providing them solutions to the business challenges that they're trying to overcome. That's really what it's about at the end of the day. So it's not like we're selling a service or we're trying to say rip and replace and our service is better than somebody else's. It's all about how we implement these technologies and how we take the customer on this journey of migrating from a traditional voice-routed infrastructure to the more advanced next generation SIP edge, where we're embedding a lot of these services and delivering an outcome to the end user. And that's where I think it's different than what you're seeing in the market today.

Tom Arbuthnot: And how do, how would Pure IP get the relevant data to the customers? Like, it makes sense, things like fraud detection as, as, as a service with AI, like the, the carrier's taking the workload there and delivering the right call or the right flags to the SBC. When it comes to, like, call insights or summaries or anything like that, how are we getting that data to the customer's CRM, or the customer's line of business app?

Adnon Dow: Yeah, you know, that's a really good, a really good point there, and some of the things that, you know, that we've implemented are LCR and LRN, infrastructure networks, that, that we're managing internally. So for our least cost routing platform and our location routing, capability, our LRN routing engine, we're able to take the... And ingest the calls coming in, and we do one of two things. We either dip a third-party, database to do a lookup, and we look at and validate that number, the authenticity of that number, if there's a fraud in that reported number. We cleanse it in that ecosystem that we belong to. We implement the policy and the routing based on the, what we see, either LCR based on cost routing or LRN based on location, and deterministically what kind of call is coming in. We score it, and in real time, it's dipping third-party databases to check it based on the trusted ecosystem, and then we route it based on that response. So it's not just our data. We're partnering with leaders in the market where we are going in and we are looking at, and then looking up that data in the infrastructure. We're looking at all of that capability in terms of the, the background information on that caller's number, and based on that, we are routing it to the end customer to deliver, a cleansed call, if you will, once we've looked at location routing number, and we've looked at that, that, that routing hops, how many hops based on the cost that's coming through. And different customers will apply for different things, right? Some may want, "Hey, I just want a least cost. I want a, a rate." Others who say, "Listen," let's call it, you know, a hospitality, a hotel chain, where they wanna make sure that when their numbers, when they're calling, that, or the customer's calling them, that that is not a fraud call and it's not blocked, and that reputation scoring comes through. And that's where you really create that trusted ecosystem and be able to leverage other databases to where you're cleansing that information as you pass the call through.

Tom Arbuthnot: Yeah, the fraud one is really interesting because it's, it's like AI is the solution and AI is the problem in this case. We're seeing a big uptick in, you know, robocalls and customized AI and, and fraud because the AI enables the bad actors in so many ways. So actually it's been... It's more important than ever before 'cause all that stuff is going through the roof and will continue to as these, voice agents and, get better and better, I think.

Adnon Dow: Man, you know, this is the thing about, voice AI. It's becoming so sophisticated that, you know, I think there are gonna be new technology that are gonna evolve every day. And, as new technologies evolve, then obviously there's gonna be bad actors and there's gonna be, impacts in the network that will then give challenges and arise to changing the way these calls are being processed, and we're seeing that now. You know, a lot of these robocalling and, a lot of these calls that are coming in that, that impersonate, you know, deep fakes, it's becoming harder and harder. So the only way is to partner with the best of breed, best in class, companies so that you enhance your capability 'cause nobody and nobody out there is gonna create everything. They can try—

Tom Arbuthnot: Yeah...

Adnon Dow: But they're never gonna be the expert at everything. And there are a lot of companies that are the experts in, in that, and that's what we have done. And so that is why we have leveraged LCR capabilities with such additional way to route based on cost and the least cost route that's out there. And then we've moved over to LRNs, where we've got NNIs and peering agreements with some of the top carriers so that when our numbers come across, they know it's coming across from us. There's not multiple hops, and they know where that originating number's coming from. We are the numbering authority, so they know it's a trusted number, and then we minimize those fraud calls. We put a, a high score on it, so when it's delivered, it's got high availability, it's got real-time analytics that it is a trusted number, and it's processed through. And today, AI is through the, you know, the RMD right now that's going on, which is the Robocall Mitigating Database here in the US, in North America. Like, the... If you get on that list with the FCC, it blocks you, and you gotta take 'em out of the service. And these are the people that are doing, you know, robocalling, and they've got AI doing it. And the only way to get out of that is in order to implement your own LRNs and create these trusted networks, and that's another one of those phases that we're creating on the edge. It's that, that next generation deterministic call routing policy that you're implementing based on that outcome that customer is looking to, you know, acquire.

Tom Arbuthnot: Nice. And I, I know we've got a, a different conversation coming up to talk about voice agents in more detail, but I just wanted to get your quick take as we wrap up on, where voice agents are, where the value is, and what we're looking at in terms of bringing that into the, the, the platform and the intelligent edge.

Adnon Dow: You know, look, for me, I look at it in, in three different paths. First of all, if you look at the omni-channel, and, and where, you know, communications are coming in from. You're looking at voice, you're looking at SMS, you're looking at chat, email, UCaaS, CCaaS platforms, PBXes, SIP PBXes or WebRTC, for example, and then you're looking at mobile phones. So you've got communications that are coming through either through open APIs, which are a lot of these agents, or you're looking at some sort of, SIP type of signaling that comes through. And the first place that it hits, it, it hits, you know, your edge voice network. And from that edge network, you gotta apply the security. We gotta figure out, okay, so is this a legit call? Number one. Number two, you gotta make sure it's secure as it goes through your network. Number three, you gotta look at the routing policy. Where is it going? What kind of services? And then you gotta anchor that media. Is it a recording? Is it a transcription? Is it just a plain phone call? And you gotta make sure there's high availability and you're monitoring that entire quality of the call. And then so there's services that you put on top of that, which are all the policies and analytics, everything from quality monitoring, the CDR analytics where you're enriching now the data, the tenant and policy management, the location, DLR location for 911, 112, and the compliance enforcement. So that's what we're doing there on the edge to enrich that data. Then on the outbound, it's either gonna be a live voice call, so it's just gonna go right to the end device, or it's gonna be a recorded path, so this is where you get your recording, the, the secure storage, the, the search and replay, the retention of the data from a compliance standpoint, how long do you hold it, so the legal requirements there. And then you start getting into agentic AI, which is kind of the third path, where you're looking at once you anchor that media, you're gonna look for a deterministic outcome. So everything from understanding the intent of the caller, making sure that you're transcribing, you're summarizing what they're saying. You're looking at sentiment analysis, scoring, and compliance. It becomes a multimodal insight in where they're coming in, what are the workflows and the automation that they want, and then you have to look at agent assist, where the questions and answers that come in, and someone like me could be asking, "Hey, can I get a hold of my bill? What's the last payment I made?" And you're actually in, in speech to speech, going back and forth with an agent, and you're using multiple RAG models and, and agentic models on the back end to be able to retrieve information and answer those questions. All of those things, Tom, all of those things can't afford latency. They have to be secure. You have to make sure that it is... There, it's not a fraud call, 'cause a lot of it are deep fakes, right? And you're analyzing, recording, and transcribing so that in real time, either through the call or once that call is done, the actual customer can look at these things, and if there's anything flagged, they can go in and retrieve that data, because we're housing it, and we're holding it, and we're retaining it, then we're providing a search and replay based on contextual words, based on the, you know, statements that are made, words that are said throughout that whole conversation. So this is the checks and balance and the intelligence that you're seeing, everything from just a regular voice call that takes a, a live voice path, all the way to the, multimodal, what I would say, multimodal AI and agentic AI workflow-enabled, work stream. It, it's a very, very, very complex and a very different path than a regular voice path. And that's where you're seeing things go, especially with AI.

Tom Arbuthnot: Yeah, no, super interesting. I think the, I like how you kind of paired there. Like, there's a lot of exciting things going on in the voice agent world, but if you want voice agents, you're gonna need to potentially have recordings of those sessions, make people know it's AI, treat the data in the right way, do those database steps. So it's, it's a really interesting time there, but you need the fundamentals of the, the network and the latency and the control and the security and then the recording before you necessarily get to being clever with the, the AI agentic stuff.

Adnon Dow: And nobody's talking about that. That's the crazy thing. Everybody's talking about, "Oh, voice AI, voice AI." You wouldn't have voice AI without having an intelligent edge and a secure network and a robust network to process it. 'Cause that's the thing about large language models. As good as the agents, voice agents are getting every day, if there's any latency, I, I think you've been on an AI agent call before where you've called and you've spoken to an AI agent. If there's a pause or if there's a disruption in the communication or if there's background noise, it completely throws the agent off. I mean, I've noticed it all the time. Nobody talks about—

Tom Arbuthnot: That. Yeah.

Adnon Dow: Right? Our, even our own agents, when we first started rolling them out, you, if you stutter or if you said a word or if there's background noise, it automatically throws that agent off, and then all of a sudden it doesn't know what you're saying or it takes some sort of, you know, hallucination and comes back with information that wasn't, really asked for, over and above or completely aside from what you've asked for. And so this is one of the things that we're really working on, and things are moving from just regular voice AI, which has been the big kind of, the big hype, and it's moving to agentics 'cause people want workflows. They want deterministic outcomes.

Tom Arbuthnot: You said outcomes, and outcomes is really interesting in AI as well because there's a lot of different options and potential costs and things. So going with an outcome model is really interesting 'cause it kind of puts skin in the game on, on both sides of the fence.

Adnon Dow: Yep. To me, that's, that's what it's all about, right?

Tom Arbuthnot: Awesome. Well, Adnon, thanks for catching us up on all that. If people want to find out more about the, the intelligent edge and the stuff Pure IP are doing, where can they find out more?

Adnon Dow: Yeah. So listen, they can go to our website and, look at all the capabilities. There'll be guides there. There'll be instructions on where to go. We're offering it across our multiple segments, and obviously we're, releasing a lot of WebEx, sessions where we're talking about it and, and informing people. But yeah, just stay tuned. There'll be a lot more coming out as we develop it and as we release, the different capabilities with our AI products.

Tom Arbuthnot: Awesome. Well, Adnon, thanks for catching us up, and, look forward to talking again.

Adnon Dow: Soon. Thank you very much, Tom. I appreciate your time.