Can AI Be Trusted to Run Critical Networks?

Artificial intelligence is firmly in the heart of the world’s most critical infrastructure — the massive networks that keep our digital lives running. In this episode of the AI Proving Ground Podcast, two of our top trusted advisors to the world’s largest network operators — Dave Clough and Yohannes Tafesse — break down the high-stakes reality of applying AI at scale, the often-overlooked work of preparing data and building trust, and why the lessons emerging from telecom will shape how every enterprise approaches AI in mission-critical environments.

Support for this episode provided by: WEKA

More about this week's guests:

Dave Clough is a seasoned Mobile Solutions Architect with over 30 years of experience at the intersection of networking, business strategy, and innovation. He has been directly responsible for several billion dollars in sales and is recognized for his rare ability to blend technical expertise, market insight, and leadership to deliver impactful results. A proven leader in designing and deploying breakthrough solutions across 3G, 4G, 5G, Wi-Fi, virtualization, voice and routing, Dave has helped shape the evolution of modern networks.

Dave's top pick: Four Pillars of AI Success for Service Providers

Yohannes Tafesse is a technology enthusiast with more than 12 years of professional experience in the field of mobile telecommunication systems. Prior to joining WWT he worked as a mobile network planning engineer at U.S. Cellular. Prior to that, he worked at AT&T as a specialist RAN engineer. His interests include mobile communications, cognitive radios, machine learning and scalable data centers.

Yohannes's top pick: What NVIDIA AI Aerial Means for Telcos Seeking to Optimize Wireless Networks and Deliver New GenAI Experiences

The AI Proving Ground Podcast leverages the deep AI technical and business expertise from within World Wide Technology's one-of-a-kind AI Proving Ground, which provides unrivaled access to the world's leading AI technologies. This unique lab environment accelerates your ability to learn about, test, train and implement AI solutions.

Learn more about WWT's AI Proving Ground.

The AI Proving Ground is a composable lab environment that features the latest high-performance infrastructure and reference architectures from the world's leading AI companies, such as NVIDIA, Cisco, Dell, F5, AMD, Intel and others.

Developed within our Advanced Technology Center (ATC), this one-of-a-kind lab environment empowers IT teams to evaluate and test AI infrastructure, software and solutions for efficacy, scalability and flexibility — all under one roof. The AI Proving Ground provides visibility into data flows across the entire development pipeline, enabling more informed decision-making while safeguarding production environments.

Chapters

• 0:00 Setting The Stakes For Telco AI
• 2:24 Operations Efficiency And SON Foundations
• 5:27 From Alarms To Churn Reduction
• 9:50 Making Money Vs Saving Money With AI
• 13:40 Five Nines, Trust, And Human Oversight
• 18:20 Data Readiness And Breaking Silos
• 22:24 Prioritizing Use Cases And AI NOC
• 27:05 Capturing Tribal Knowledge At Scale
• 30:00 Proving ROI And Efficiency Gains
• 34:06 AI For Cyber Defense And Device Posture
• 38:15 Crawl Walk Run And Investment Strategy
• 41:00 Future State Of AI-Driven Networks
• 46:15 Key Lessons And Closing

Transcript

SPEAKER_02: 0:00

From Worldwide Technology, this is the AI Proving Ground Podcast. Today, it's easy to forget what really powers this digital world of ours. The calls we make, the messages we send, the streaming, the commerce, the critical services we use daily. Everything depends on a vast invisible network managed by the world's largest service providers. They're the quiet backbone of modern life, and they're under pressure like never before. Margins are shrinking, customers' expectations are rising, and behind it all sits a level of complexity that's almost hard to fathom. Hundreds and thousands of stealth sites, millions of alarms, petabytes of data, all of it demanding to run with near perfect reliability. Now, enter AI. In this episode, we're talking with Dave Klough and Johannes Tafaste, two field CTOs here at WWT, about what happens when artificial intelligence is asked to manage some of the most unforgiving systems on Earth, networks that can't afford to go down. You'll hear how service providers are using AI to improve efficiency and effectiveness of their operations centers, predict outages before they happen, reduce customer churn, and even defend against AI-driven cyber attacks. But you'll also hear why it's not easy, why trust, safety, and governance matter just as much as innovation, and why the lessons these operators are learning about data readiness, ROI, and the human side of AI apply far beyond telecom. So without further ado, let's jump in.

SPEAKER_01: 1:42

Doing great.

SPEAKER_02: 1:45

Excellent. We are talking today about AI in the service provider space. And I am curious, it seems to be an interesting landscape. You know, the operator business is under immense pressure. Margins are shrinking, expectations of consumers are rising, revenue streams are having issues.

SPEAKER_03: 2:15

So one of the key areas that we're seeing AI helping operators today is allowing operators to become more efficient in their operations and get deliver a better service. And this actually started a few years ago, actually using some uh machine learning systems with uh SON, which is self-optimized networking for radio networks that allowed us to have much better uh radio communications. Uh, if you're familiar with uh the 2G to 3G migration, the 3G to 4G migration was really nice and it had a lot to do to do with that technology. We're also seeing the technology starting to come into these very, very complex um maintenance and support areas from their NOx and network operations center. Uh most operators have a tremendous amount of alarms. AI is able to come in and actually run through a lot of those alarms, figure out the ones that are meaningful and those that need to be worked on, as well as keeping uh interaction between the different systems up to date. And this is really, I can't stress this anymore. Um every operator is moving along and having a number of changes and upgrades and solutions happening all the time. And they're buying new systems, they're taking down new systems, and keeping that all correlated and up to date is extremely difficult. AI has some capabilities to actually understand those documentations, those changes, and be able to actually put forward a more efficient uh interaction and connectivity between those systems, as well as diagnose for a knock when those systems are out of phase or out of out of sync or need to be upgraded. And that's been one area that we've seen a tremendous amount of value on that. Um, we've also seen it from being able to help with uh churn reduction. So being able to identify users that have not had very good experiences. So being able to actually target those people and then tell marketing, uh, well, we really need to give uh Johannes or Brian over here some notoriety and give them uh a coupon or something along those lines to actually get those systems up and operational better. So that's actually been pretty interesting. Um, and I I think we're gonna we're gonna see other areas here as well. I I've yet to see really operators take AI and bring it to the mass market. But Johannes, I mean, you you've you've got through a lot of this stuff. I mean, you've done a lot of earlier saw on systems, and I think you and I have talked a lot about this, and we we've just seen you know a number of different areas dealing with information as well as as uh support, but going over the top to top line growth, let's say where they make new money is we're still finding our way, but I think saving money, being able to optimize, has been rather uh rather good. And we found a number of areas we can do even more work on.

SPEAKER_01: 5:13

Yeah, Dave, you're absolutely right. Um, in terms of just if I if I could start with um the making money aspect of it, um I think I think a lot of businesses are still kind of figuring out, right? AI as we has been, Gen AI has been around for for a couple of years now. Um and the service providers, I mean, they serve hundreds of millions of customers. So there is some real opportunity there to say, hey, what can we offer to our customers? And there's still some deep exploration uh that's happening, especially leveraging some of the infrastructure they already have, right, that they have invested in. So, and you have to remember they're looking at both uh consumers as well as businesses as customers. And there's a lot of opportunity to add value to their uh you know business customers through AI. So that's an ongoing discussion. Uh, but on the flip side, um, there is this huge infrastructure that service providers, telcos, manage, right? So if you're in the uh wireless space and you you manage 3G, 4G, 5G networks and you have um 100,000 cell sites, that's huge, right? Distributed all over the country. You have the core network and you provide um all sorts of services. So being able to manage um this huge infrastructure and optimize it and operate it, uh, there is huge opportunity for AI there. It's it's been very manual intensive. Uh and uh that is an area where we're seeing some success with some service providers, you know, AI plus automation, and now you're adding Gen AI for um you know interaction with the network using human language and being able to create some intent. So uh we're seeing a lot of um success there. And of course, um, you know, tying that uh with uh you you you pointed uh churn reduction uh with customer experience, right? So being able to define what's the customer experience, you know, as as they uh get to experience different parts of the network uh during the rush hour, uh, and being able to identify where there are opportunities, maybe prevent churn or maybe upsell. Um so there is this uh real synergy going on between how the network looks at customers, how the marketing looks as customers, marketing team, right, um operations team, and being able to deliver an end-to-end value. Uh so there's a lot of opportunity where AI could could help uh service providers.

SPEAKER_03: 7:42

Yeah. I should just I should just go for some of our listeners here, churn is uh it's really a it's a mechanism that uh Johannes and I are very close with. So um churn is reported by every operator. And really what that is, it's a it's a score of how many subscribers they lost versus how many subscribers they brought on to their network. And if that continues to get to lose and gain and not have an overall gain, it's not really good for the operator because it costs them a lot of money to get new subscribers onto their network. Because nowadays, most of like the radio, the the wireless networks are are very complementary. They all they all are uh similar in capabilities and coverage and and um and uh capacity as well as as price. So just to just a little sideline on the background uh on that one. Um, but I I think the that is something that's been really interesting to see in the market, uh modernizing the network and being able to offer more and more capabilities and really seeing the operators start to focus on uh experience of the user, where in the past, I think it was just about coverage capacity, rolling systems out. Now AI gives them some really good information back. I think the other thing that's really interesting to me is that there's a tremendous amount of data, and I think you and I were just talking about this a little bit. There's a tremendous amount of data that subscribers have, not subscribers, but but operators have in their networks, that has really been untapped. And it's to the point where not any language models have been able to see the insides of that either. And so one of the things that I think you and I have been talking a little bit about is there might be a day where operators will use and build out their own language model based on a lot of their data and the way that they want their network to run. It's almost like uh building a board member, if I would describe it as uh another person in the organization that has the persona of the operator, has all the information of the operator, and can use in conjunction with the employee to make sure that they are able to do a better job, understand information better, and then deliver potentially better decisions and better outcomes to the market. So you know what? That's just something I think we talked a little bit about.

SPEAKER_02: 9:56

But yeah. Yeah. Uh Johannes, I I wonder, just so um we can level set with our listening audience out there, that might be in another industry um and not necessarily the the service provider or telecommunications industry. What lessons do we think can be taken away from this conversation as we dive deeper into the uh, you know, the the telco networks? Um is it just a matter of they're doing all these things at a bigger scale? Or what types of lessons should enterprise leaders be taking away from how uh the service providers, the big, you know, these giant, biggest of the big companies are are treating AI and trying to advance AI?

SPEAKER_01: 10:32

No, so that's a really good question, Brian. Um, so you know the service providers are looking at AI from two, you know, two perspectives, making money, saving money. And even in the saving money aspect of it, right? So you have your mobile network, this gigantic infrastructure that you have to manage. It's considered a critical infrastructure for the country. Um, and then you have your IT office and you you have marketing, you have HR, and there's a lot of opportunities in that space as well, which is um you know common across the enterprise uh world as well. So one of the main things that you know looking at the size of the investment that the service providers are looking at is being intentional, right? So there's a lot of hype sometimes in terms of the value you can get out of um an AI system and what it means. And we all use all sorts of AI systems in our day-to-day lives personally, and you know, um, we get some benefit out of it. Uh, but being able to be intentional, look at your workflow and where you add value and design accordingly is something that that you know the service providers have to do because they're dealing with this critical infrastructure, right? So, and then in terms of the adoption, even though a system adds value, if it disrupts um the way you do business significantly, all of a sudden, the adoption tends to be a little bit slower, right? There's a lot of caution. So making sure that, hey, how do we adopt this in a way that does not impact uh our day-to-day operations and does not actually lead to loss of efficiency and loss of confidence is extremely um important. The other thing is pulling in all the stakeholders into the discussions, right? So you have uh the executive sponsors or decision makers that are looking at what value are we creating. You have maybe your data science team uh that's looking at this, um, you have your IT organization. Uh, maybe you have to bring in GPUs and you have to look at the infrastructure you have, the cooling and the power and all that. So everybody coming to the table looking at this. Um, and after identifying that, going through, hey, let's test it out, let's prove it out, and how do we think about scaling afterwards is something that I think everybody should should look at. That would help with um adoption and getting value out of that that uh you know implementing AI at a at an organization level. Yeah.

SPEAKER_03: 12:56

I just add to that one. I I think you the difference is really that an operator is held to higher regard in terms of reliability and performance. But otherwise, the the actions and the approach are are very similar in terms of looking at your data, looking at business needs. The the difference is that with an operator, I've got a I've got to achieve about five nines of operation. And then if I do have an outage that's caused by, let's say, uh an AI system that doesn't make the right decision, or I'll let that happen. Um, you know, I have the federal government coming after me or the FCC or have uh other things that come back behind me.

SPEAKER_02: 13:42

Well, Dave, I'm glad you mentioned that five nines um, you know, type of scenario. Service provider networks, you know, they are held to a higher regard. There's a lot of critical services that run on these things. There's a lot of businesses that rely on these things. And, you know, we are talking about the need to have that five nines reliability. Um it's not a buzzword. Uh Dave, what makes AI safe enough and reliable enough for that five nines environment? What do organizations have to do to get to that point?

SPEAKER_03: 14:12

Well, really good question. So one of the things that we've seen um is uh the kind of the trust but verify, run through well-known AI approaches, look at with the output, and then help train that particular model or that system. And one of the things I talked about a little bit earlier is using a lot of the uh the data that the operator has to actually potentially build models. And it's actually within reach now for a lot of operators to actually start building your own models. And that will allow for basically first-hand knowledge, first-hand knowledge or first-hand data of the network to help fuse into a data model that allows for uh, you know, potentially better behaviors, less hallucinations, and potentially more accurate uh transmission of data. It also goes back to the training systems we talk about, and then making sure we have human in the loop until you get to a sustainable, measurable outcome where potentially when you do see uh an AI system that gets some sort of insight or idea that you're able to have enough trust in that that it can basically take it to an automation system or a change in the network. But almost almost within the next you know five to ten years, there will almost always be some sort of human in the roop roop on those systems for the critical infrastructure, I would expect, until uh those systems become can become more viable. Um we we have seen very good um uh approaches and capabilities within SON. And that was that's not necessarily a it's a kind of if you will, early it's more machine learning where SON was able to take measurable measured outlets and then do automatic changes, and that allowed for very, very reliable uh increment increments of support for for wireless networks. And Johannes and I have talked about this a lot, and uh, you know, Hannes, what's your view on that one? I I think it's it's it's still there, but it's kind of you know, it's one of those systems. I think my experience has been that I've watched somebody like Google use AI to an incredible degree um and have amazing output. But the only thing I have with that solution is it's I see it from the outside. Google hasn't allowed you to go inside and see exactly what's being done and how they're doing it. But they're indicating that there's some pretty good stuff here. So there's something, there's a lot, a lot of stuff there that could be interesting, but it's a trust but verify approach just because of the reliability required.

SPEAKER_01: 16:47

That's that's exactly right. Um, you know, on the on the flip side, the infrastructure that the service providers deal with in general are very complicated. You have a lot of data points. Um, so an argument could be made that you know these AI models could actually make uh more optimal decisions than than humans, given the size of the data, right? So um so that's an opportunity, uh really. But from a 5.9 perspective, uh Brian, like you said, I think there's an opportunity to make sure that uh the infrastructure, even the AI infrastructure and everything around it, redundancy meets that 5.9 requirement. I think as time goes on, uh we see the you know um the telecom world and IT world kind of merging a little bit, right? So the way we think about infrastructure, you have disaggregation and you have um network services being provided as as virtualized or containerized software, so things of that sort. But when you make that happen, you still have to keep in mind that you know delivering this uh 5.9 requires efforts from um the underlying infrastructure from the hardware, from the power perspective. And that goes to the AI as well, right? So just keeping that in mind to make sure that none of that is compromised is important.

SPEAKER_03: 18:04

Yeah, that's a really good point, Johannes, because I I find today a lot of decision making in a knock, or just even from an executive level, is not always 100% because they're not able to absorb all the information fast enough to make a really a better decision, in my business. And I think AI in in those particular cases could be much more informed and more accurate and allow for these big organizations to deal uh much much faster. Because as we know, uh operators typically, when they work on something, it's it takes a long time because it's very complex, it's big, and anything that can speed up the decision making is a huge advantage for them.

unknown: 18:51

Yeah.

SPEAKER_02: 18:51

Yeah, I want to I want to get into the AI-powered knock assistant that I know that we've helped um develop and and offer to to our clients um in and outside of the service provider organization. But before I do that, Johannes, I want to pull on that data string a little bit more. You know, certainly operators are sitting on mountains of data from telemetry to uh you know network performance, usage, and so forth. Um how and that's not a situation that's exclusive to them. We hear that from clients across the board that you know data in silos is an issue. How are we advising or how are we helping our clients, you know, move from those silos of data um into more actionable insights that are kind of you know ready to be consumed by an AI? Because you do need to do that first.

SPEAKER_01: 19:37

Yeah, no, absolutely. Uh in you know, I think in um in AI, and prior to that, we called it machine learning, and prior to that data science and statistics, whatever you want to call it, you know, the the the famous adage is if you know garbage in, garbage out, right? So um, and then truly, if you want to take advantage of AI, uh, whether it's you know workflow automation or optimization, you need um uh a good way of making sure that you have your data that's accessible, right? You need your data needs to be clean, structured, and reliable. Um, and that is really a culture within an organization of okay, identifying what data do I need to capture, right? And then how do I uh make use of it and how do I keep it up to date, right? So if the data is you know out of date and you get information from AI on that outdated data, then you lose trust and the you know it's gonna be um a problem to move forward. So I think focusing on your data that you need, and telcos do some you know a good job in terms of that data retention um to some level, because there are some regulatory requirements in that space, right? So uh that's that's extremely important. Um, but also when you think about hey, how can I really leverage AI by bringing in this data, even if you look at you know the big service providers that um um deliver services all over the country, you have a very distributed engineering team um that, for example, in in in a given part of the country has figured out how to optimize the network. So figuring that detailed configuration and how to do that for this complex network and making sure that that knowledge gets through um to the other engineering teams all over the country, but gets optimized or tweaked to adapt for the local settings is a challenging process and takes a long time, right? So, you know, by the time you figure out hey, there is this feature or parameter that I can optimize to uh the time you get it rolled out in your entire infrastructure, it could actually take months or more than a year. So that's not that's not surprising. So I think AI, even at this level, could play a huge role to um optimize the deployment of solutions that are already understood as well. Um, and that means a ton of money from if you look at it from an experience, a customer experience perspective. If you look at it from man hours that is spent on that perspective on the network, if you look at it from hey, I've already purchased this feature or this capability from my vendor, and usually costs me millions of dollars, but I haven't used it for a year. A lot of times, actually, features don't get used for multiple years. So that's investment you've already made. You just haven't had the opportunity to get to it. If you leverage AI and get, you know, get to it, then you're you you're saving a tremendous amount of money, right? So you're making good use of the capital investment you've already you've already made.

SPEAKER_03: 22:44

I'd just add that I mean, we've just found there is a treasure trove of different data. As as Johannes is describing, there's different pockets and different pockets of expertise. So it's really being able to make sure we get the data put in the right uh organizational structure, clean it well, um, so that we, you know, we get rid of some of the bad data coming in. And then the operational capabilities above it, as is Johannes described, replicate that throughout the organization. So it's it's a both end. So really we've we've done with the AI NOC is really we've started one area. Um, and then as we've cleaned and gotten really good data, let's say on the IP network, now we're getting into the access network, now we're getting into the security network, now we're getting into you know security operators, now we're getting into the enterprise side. And being able to kind of go by lead by example and move that forward. Um it's a it's a very, very large problem. Um, all of the operators are kind of grappling with how they have different formats and different systems. Um it's so it's twofold. It's it's organizing the data and then bringing the right AI uh practice or hygiene uh to it and using it.

SPEAKER_00: 24:00

This episode is supported by Weka. Weka helps enterprises and research organizations achieve discoveries, insights, and outcomes faster by improving the performance and efficiency of GPUs, AI, and other performance-intensive workloads.

SPEAKER_02: 24:18

Yeah. Yeah, certainly treating data as that strategic asset and giving it the you know the attention and care that it deserves is is is one of the key pillars that I know that we talk about in the WWT research that we've released to date. And by the way, for any of those listening out there, WWT.com has a lot of great uh research as it relates to AI, specifically um in the service provider industry as well. So go check that out. Um, you know, Dave, once you have that kind of data estate in order or you're on the right track, another key pillar is prioritizing high-value use cases. So walk me through some of the value add use cases that we see emerging right now. And I do want you to touch on in your answer, the AI-powered uh knock assistant and why that is such a high value use case at the moment. Is it just like that's an easy area to apply it and you can start to build momentum towards the larger things, or is it just a providing um immediate impact?

SPEAKER_03: 25:12

Well, uh so I'll I'll kind of take those couple questions, Adam together. So we have a number of really interesting uh approaches to the market, the AI NOC. We would like um to actually help for uh optimization of knock operations and more efficiency there, which I'll get into a little bit. We also have um the churn reduction solution that we run through that's actually been pretty interesting and has actually run really well. Every operator has a churn problem. We've also looked at documentation that feed into those particular systems, and then we have other solutions we're developing right now for security analytics and and systems behind that to allow for more secure solutions, having AI look at that as well. Um, I would say the uh the the AI knock is not trivial. It's we've seen a lot of people try it and not do well because they haven't really understood the full uh gravity of what you need to do, which is you just can't go in there with a tool and just lay it on the bench and then start running it. Um it's a series of tools, a series of approaches that you bring together and fuse together, as well as interviewing um the customer and understanding the way that they need to do business and how you need to tailor it for that particular area. And that's often overlooked or it's brushed by or it isn't optimized in a way that make those systems viable. So it's really interesting to see a couple of studies have come out um that have shown that a lot of AI work has failed. Um, and that has to do, I think, with not really putting the right rigor and the right uh discipline around it. When we've come in and worked on uh the AI knock, we've really looked at, okay, what is the problem? What is the business case? We start there, and then by identifying the business case, now it allows us to go and find the right tools, the right architecture, right solution, rather than starting with tools and solutions first, um, and then trying to figure out how you're gonna fix that use case. And that's I think that's a really good, great part of WBT and the way at least we work together with customers. That's really allowed us a lot more uh flexibility to go to go through those particular functions um and really start solving the business problems and adding software where it's needed. It's it's been interesting in this journey that we've been going along. We initially had to go out there and build connectors to ServiceNow. We've had to develop connectors to Splunk, had developed connectors to other data retrieval systems. Um it's been the system's been out there for a while in a couple of accounts we worked on. And since then, those providers like ServiceNow, Splunk and Up have actually built APIs or actually connectors. So we don't have to develop those systems in the past. So it's uh so it's been there's a recognition in the industry that this is a needed area, it is of great value, it's not trivial, but I think as you move through it, you get relatively uh uh easy to measure results. And from my experience, that's that's that's been excellent. And your house has been right in the middle of that too. I I think you put some some comments there if it makes sense.

SPEAKER_01: 28:30

Yeah, no, absolutely. Um so just having a very focused discussion we found with the right business group within an organization to identify value is um is extremely important. Hey guys, real quick, is my video blurred?

SPEAKER_02: 28:47

Nope. You're coming in coming in good. Yeah, feel free to I would just uh Johannes, just start your answer over. Um but yeah, you're coming in good loud and clear.

SPEAKER_01: 28:55

Okay, great. Um, so I think having a very focused value discussion with the right stakeholders within an organization is extremely important to prioritize value, right? Say, okay, what are we gonna get out of this? Is this better customer experience? Is this uh operational efficiency? Is this we're gonna we're gonna make money uh by uh you know going through this this um AI deployment? And we found that to be very useful just to ask those set of questions. That way everybody's aligned uh and and right you can go after some of those use cases. The other thing that's you know something that we've learned is um there are um you know reasonably achievable use cases that you can you know go to success uh rather quicker, right? I would say um low hanging um uh fruits. And then there is the um difficult use cases, uh, right. So um you want to go after some of the easier ones early on to build confidence and get credibility uh so that you can go through. Some of the more complicated use cases that may take a little bit longer, right? So that there is the willpower and continued investment to make sure that you get the result. And the other thing that we looked at in terms of achieving this, and Dave mentioned that AI NOT is not a trivial solution, is for any successful AI deployment, I would say 50% is a good understanding of the space you're working in, domain knowledge, right? And 50% is an AI capability, the tool sets, or I'd say I'd call it the computer science aspect of the business. A computer scientist that's extremely skilled in AI cannot come up with a solution to treat patients if they don't know anything about the medical industry, right? Or they don't understand about the human anatomy. So that's where you have to look at the same as the same the telco space. You need to understand the space very well. You need to understand how this networks operate and what matters to subscribe providers so that you can have the same conversation. So that's that's extremely important. And in the NAC case, um, you know, what we've realized is there are a lot of tickets and outages that come up that uh consume a lot of time. Uh and we saw that uh this AI solution uh really created a lot of efficiency. Um so um that made that made a lot of sense. Um and over time, what we're trying to do is call, you know, what you know, catch what's called a tribal knowledge. As you come through and resolve issues and tickets, um, you know, some of the more seasoned technicians, engineers could look at an issue and and you know, fig figure out what's going on fairly quickly because they've been in the industry long enough, they know that particular network, you know, whatever the the operator that they've been working with for quite a while, right? So they know that network like the back of their hand and and they do well. If you bring in more of a junior um expert um that you know, or or engineer to come in and look at the system, it's gonna be a different experience. However, as the senior guys work through this problem, AI is able to capture it. Now you can have your you know junior or mid-level guys perform um at the level of the seasoned technicians, right? So I think that's a value that that's usually um underestimated, but over over time could create um could add a lot of value to to um service providers.

SPEAKER_02: 32:42

I want to shift gears a little bit to um ROI and something that so many of our clients struggle to articulate. Um what do we how do we knowing that service providers are under that pressure to constantly on a quarterly basis address churn, address revenue streams, and so on and so forth, these AI investments are not cheap. How do how do we make the case or how should we advise um our service provider clients or any organization out there to think about how they articulate ROI back to their stakeholders, whether it be you know Wall Street or uh their executives or their boards? Like what is the what is the equation here, Dave, with uh as it relates to ROI?

SPEAKER_03: 33:25

Yeah, well, I think it's it's ROI is I I would I would kind of change that to a little bit so return on your investment, but it's also a return on your experience in that network. So if you have a really good experience in the network, you're gonna have a more uh a better consumer attitude and you're gonna have more people who want to consume your service. So that's that's one aspect that I think operators are starting to go after because if you look at the operators today, they're all competing pretty much on the same technology. And in the past, they've been uh competing on different technology, different coverage. One had the iPhone, one didn't, and that's how they differentiated. Now they've got to really go back and say, okay, how am I offer a really good experience? And then make sure that drives lower churn then. So that means more people are gonna stay in the network. Churn is one of the number one um line items in a quarterly report that everybody looks at, and that can make or break a stock from you know, from a landline perspective, from a wireless cable, any of those systems. That's that's a huge, huge area. Um, the other one to look at is really how can I make my operations more efficiently? And I think that's the other metric, and that goes back to how can I take um and use my workforce more efficiently? Not necessarily am I going to reduce them, but how can I have that that person that's an enabled to make better, faster, more more meaningful decisions? And that's where AI can really help as well. So we we know the first things we did when we worked with with service providers is actually brought in RAG models just for documentation. So documentation from Cisco, from Ericsson, Nuke, it's extremely exhausting, extremely difficult to go through. And being able to get accurate information allowed those planners to build much faster, much more accurate decisions, and then get that system in place faster. If you can in place system faster, you can deploy something faster. Now you can be much uh more competitive in the market to deliver a better service. That's another system. So I've talked about quality experience, amount of rig workforce uh productivity and efficiency. And then the other one that's really uh impl implemented and and needs to be further implemented is the uh network efficiency uh capability. So we're actually seeing gains of 20, 30 percent on let's say radio optimization or network optimization. Networks today typically um there's no interaction between the application and the network. Basically, the application guns and sends something, and the network just has to absorb it and deliver it. There's no feedback mechanism. With AI, with those systems, we're gonna be able to have a much more dynamic, much more capable network and leverage things like segment routing and other functions so we can actually make sure that delivery is much more accurate and much more efficient, so I don't have to have just a tremendous amount of capacity and expense put into the network that I might not necessarily use all the time. So I can use it much more uh uh judiciously and support. So I kind of look at there's a lot of efficiency messages there, Brian. Um we are looking for some top line growth messages as well to show that forward. But for instance, our AI NOC can can you know cut alarms down by 50% huge differentiator. Um, it can actually help, you know, return back maybe two to three uh heads so that they can actually do more predictive maintenance rather than reactive maintenance. Uh so there's some those are some of the areas that we're we're putting into the market, but it's it's an evolving, it's an evolving approach.

SPEAKER_02: 37:33

Yeah. And Johannes, I'll I'll shift gears on you a little bit too here. I uh we're at the you know getting close to the bottom of the episode. I I do want to make sure that we uh touch on security here. We haven't necessarily brought up uh cyber, but of course, these networks are probably you know amongst the most highly targeted networks in the world. So what can we do from a cyber perspective um to protect with AI and protect against AI? And then what does that mean for uh you know the broader you know non-service provider industry? Are these trends that they should be picking up on as well?

SPEAKER_01: 38:07

Yeah, I know absolutely that that's a great question. Um AI, right? So your um the attackers, you know, whether it's you know your your casual um hacker or you know um state agents, which are becoming more and more of a problem, right? So that they're leveraging AI. That means they can um exploit um uh you know um attack vectors quicker, uh and they can come up with new ways to um attack the network and you know whatever infrastructure you want to call it. Now, and they you know it really democratizes the level of expertise that you need. You have an AI system that can come up and show you what how you can um inflict a maximum damage, uh, which is a problem. Um, the only way to encounter counter that is to leverage AI on the defense side as well, right? So make sure you have um the right partnerships. And one of the things that we're looking at is the cybersecurity space is very crowded, uh, right? So um it's it's um it's not uncommon for uh whether service providers or um you know enterprises to have multiple tools uh that they're using as part of that their their cybersecurity strategy. That by itself could create um a gap, right? Because now you're juggling multiple tools, multiple vendors. Um, I think getting to a more organized, you're having a good understanding, making sure you have a good platform that is AI aware, and there are some tools out there that we work with that constantly learn from threats that are discovered all over the world, whether it's network specific or enterprise specific, is extremely important. Um, the other thing you have to look at from a service provider space is just because you are leveraging AI, it doesn't mean you have to relax the security requirements that are around your infrastructure, the network. So if somebody did not have access to the data or to the network prior to AI, they shouldn't be able to access information or interact with the network through the some of the AI tools that you're building. So making sure that some of the um access controls that you've you have uh right flow through to the AI systems and doing your homework from that perspective is extremely critical. Um also making sure that you procure uh um some of these AI tool sets from trusted source, right? So securing the supply chain is extremely important. If you get you know models and you know AI tool sets that um you don't trust, uh so it's like it could be an attack from within, uh, which is a huge, huge risk. So that's some things that that to look at, um, Brian.

SPEAKER_03: 40:57

But I would I would ask you to add to that, Johannes, and really good points. I think one of the things you just touched on is a persona of the devices, the persona of the user, kind of the devices, and then an understanding, what AI gets allows us to do is look across the network and kind of get it. We we use we overuse this term, but a digital twin of what the network's doing. As well as we can start to use AI to prototype the behavior of devices, and then obviously look and then interrogate those devices to say, what is their what's their origination? What's what are they doing? Um, and then what are they doing now? And that has been one of the largest game changes because we do have AI coming in. We have state actors that are not coming in through malware anymore, they're using backdoors and things that they're exploits that they've delivered. And you have millions of devices in these networks that are very difficult to inventory and understand. But AI has the ability to do a lot of that for us, yeah, give it base level. And I think this is what you know we've been talking a little bit about is being able to use that. And then when we see abnormalities, we we zero in on that area and interrogate, either remove or remediate or update. So I think that in and of itself, just being able to control devices and monitor them more efficiently is going to be one of the big deals. And there's gonna be a number of tools, as Johannes talked about, that will fuse into that. But that I think will be the the bigger, bigger opportunity for us to secure our networks.

unknown: 42:34

Yeah.

SPEAKER_01: 42:34

Yeah, hey Brian, if if I may, real quick, just go back to that ROI discussion, because I think that's that's important. Yeah, we get a bunch of questions on on that one. Um, you know, the the way I look at it, and and um, you know, David and I we've talked about this for for a while as well, is um I I don't think anybody, you know, particularly the service providers can afford um to take the risk of being left behind, right? This is something that they have to do, um you have to you have to do, you have to play with, and you have to experiment with. Um and you can't just sit back and let this play out. Um, you have to get your feet wet and and you know, make sure your organization understands it and build a culture. So that's if you don't do that, then um, you know, the chances that you'll be able to control your operational cost um is not gonna be feasible, um, whether it's as a single service provider or as a service provider, you know, as an industry as well. That means you're not going to achieve the multiples that you know the street expects, and it's gonna be um a challenge by a challenge by itself. So, how do you balance having to do this with the um you know huge infrastructure investment it could require? Um, I think what we have been saying to our clients is to make sure you look at the uh crawl-walk-run approach, okay? Identify the use cases, make the minimal investment that you need to prove out to prove out those proof of concepts and use cases, get some benefit out of it, build confidence, um, and then make additional investment to scale those operations that that makes sense for you, right? So that's one conversation we're having. The other one is service providers have invested a ton of money over the past you know decade or so on you know data collection tools, AA, automation tools, things of that sort that are not being utilized. So you you've bought this this you've made this investment, you're paying these licenses, but you're not utilizing them because they're too complicated and you don't have the resources to take advantage of it. So if you put an AI there to leverage off all this data and all this tool set, it's really protecting the investment you've already made. So you have to look at how much additional investment do I need to make in AI to get the benefit out of those tools versus how much am I already spending or have already spent? So that's a different list to look at it. I just wanted to share that quickly, Brian.

SPEAKER_02: 45:00

Yeah, no, I love that. I mean, let's just, you know, we can wrap the episode here after this question. But, you know, given, you know, Dave, we'll start with you. Given what Johannes just said, where, you know, these service providers are investing in those data tools, those data collection and you know, analytics, what does that mean for the service provider of the near and long-term future? What do they look like? How will they interact with their end users? Just give me a little bit of a future state as it relates to AI data and uh and these operators.

SPEAKER_03: 45:28

Yeah, really good question. So as we move forward, I think operators are going to be able to offer a better user experience by leveraging their data. They have a tremendous amount of data that they can unleash and help uh customers offer better have have a better experience. So and be able to use that data to, let's say, you know, you have very good location information, you can feed that data with applications and services to offer better experiences moving forward. And what we've seen is that data behind a wall and not really exposed. And there's a lot of fear of doing some of that by the operators. I think as we use AI and we get better understandings of that and understand how to fuse that data with applications and services and other systems that are out there, um, we could see operators offering new experiences, so top-line growth capabilities for, let's say, differentiated data delivery. We could see we're hearing some things about, let's say, AR and VR glasses that uh that mobility can can leverage and offer you know really interesting experiences of being able to look at something and having your mobile network look that up quickly and deliver you uh some sort of insight on that, or use that as some way of recording conversations if you wanted to, or or or being, let's say, you know, going through Paris and looking at something and saying, hey, I wonder what that is, and quickly referring back and saying, Oh, yeah, it's the Arc de Triumph. This is the feedback behind that. There's a number of areas that can go in, and we're seeing the possibility of fusing a lot of what a hyperscaler does much more closely with what an operator does. And hyperscalers have this amazing treasure trove and innovation and ideas and systems and really amazing things that we use every day. And you have an operator that has all this amazing data of where the user is, what their likes, what they're doing in the network. And and when you start to be able to fuse that together, we're you know, that the sky's the limit on new, really interesting things you can do. And I think potentially AI and fusing in that data will allow that to start to happen in a more meaningful manner. Whereas in the past, it's not been a it's been a clunky relationship, it's kind of been uh adversarial many times, and you know, potentially in the future it might be much more um complementary.

SPEAKER_02: 48:02

Yeah. Johannes, any closing thoughts on what you think the future holds? Uh either bouncing off what Dave just said or or something else.

SPEAKER_01: 48:08

Yeah, absolutely. I think Dave kind of touched on here how how would the experience the customer experience change and what value can service providers add? You know, I'd say on the on the flip side, I think the way service providers manage their network, the way they interact with their network, right? Um, scrolling through a bunch of data and all the charts to see what's you know how the network is performing is gonna be a little bit different, right? So it's gonna be a cleaner approach. You you can you can ask and get the information using human language. Um, I think to be able to provision services quicker, right? Then you don't, you know, you don't have to do detailed configurations and you know, buy new boxes. How do you do that? That's gonna change a little bit. Troubleshooting is gonna change, being able to pinpoint in this complex network what's going wrong and fix it quicker so that customers are not impacted. I think that that way the network, we look at the network and manage it is going to be a little bit different as well.

SPEAKER_02: 49:04

Yeah, yeah. Well, great stuff. I mean, it's certainly an important conversation that um, you know, not just service providers, but you know, clients of all uh sorts and of across all industries need to be considering. Dave, Johannes, thanks so much for taking the time out of your uh calendars uh here today, and and we'll have you on soon to have another conversation about AI in the service provider arena.

SPEAKER_01: 49:25

Thank you. Thank you, Ryan.

SPEAKER_02: 49:28

Okay, thanks to Dave and Johannes for the great conversation. After listening, a few key lessons I think you should be taking from this episode. First, AI's value isn't found in grand promises, it's found in disciplined, deliberate progress. The service providers succeeding with AI are the ones starting small, cleaning their data, and choosing high-impact use cases they can measure and scale. Second, trust is non-negotiable. In environments where reliability is everything, AI has to earn its place. That means human oversight, governance, and an uncompromising approach to accuracy before automation. And third, innovation only works when people are aligned. From engineers to executives, the most successful organizations bring everyone to the table early, defining what success looks like before the first model is even trained. The bottom line: AI can't just make networks smarter, it has to make them faster, safer, and more human in how they operate. The companies that get this balance right won't just keep the world connected, they'll redefine what reliable innovation looks like in the age of AI. If you like this episode of the AI Proving Ground podcast, please consider giving us a rating or a review. And if you're not already, don't forget to subscribe on your favorite podcast platform. And you can always catch additional episodes or related content to this episode on WWT.com. This episode was co produced by Nans Baker, Maggie Ryan, and Kara Kuhn. Our audio and video engineers, John Nomblock. My name is Brian Phillip. See you next time.