AI's $700B Opportunity for CSPs
Podcast episode 55
The telecom industry can’t pivot to 5G without artificial intelligence. TMForum’s Aaron Boasman-Patel explains how it’s as opportunity – and a challenge. A successful AI deployment is about the metrics you measure it by and the outcomes, versus what you are doing today.
Below is a transcript of this podcast. Some parts have been edited for clarity.
Michael Hainsworth: For the telecommunication industry to transition to 5G, it must also transition to the world of artificial intelligence. Machine learning algorithms are the linchpin to holding the entire infrastructure together, and CSPs need to get up to speed fast. TM Forum's vice president of AI, Aaron Boasman-Patel understands the strengths and weaknesses within the walls of today's modern communications service provider. He points out that a mobile provider may not meet all of its enterprise customer's needs, but it will need to understand the power of machine learning and partner to fill in its knowledge gaps. And regardless of who's providing the artificial intelligence, without the real power of AI, there is no benefit to migrating to 5G.
Aaron Boasman-Patel: Now is definitely the time to deploy AI, or there is no benefit, as you said, Michael, to deploy 5G. Because if we think about what 5G is all about, it's about ultra-latency, it's about high reliability. It's about billions of devices being connected. That's where all these use cases are going to be for 5G. Particularly when you think about the enterprise opportunity that we're going to have. I mean, there's 700 billion worth of revenue ready for CSPs who want to really go into the 5G enterprise space. But what's critical is to realize those business cases and those use cases is you fundamentally have to automate all of your services. I mean, today, on a traditional telco network, there's about 400 instances per hour. That's going to go up to 10,000 when we start looking at 5G and some of those use cases. So without automation, that's not going to be happening. We are not going to get the reliability that we need and all of those benefits from realizing, the 700 billion of value for CSPs.
MH: Particularly within the enterprise community, it's one thing that Joe Blow wants to walk down the street playing Angry Birds and texting and things like that. But the remarkable advances that 5G provides the enterprise means that you have no choice but to use AI to develop, maintain, and update all of those services.
ABP: Absolutely. I mean, this is what people have to really understand with those use cases. So if it's everything from... If you think about farming, that's one of the great use cases that's been looked at. There's going to be drones going in the air. We're going to be measuring air quality; we're measuring air temperature. We're looking at soil density. We're going to be looking at the amounts of phosphate you need to put on the soil. So this is a different type of service that we're going to offer today. As you say, it's not about Joe Blow anymore, wanting to do gaming or walking down the street following a Google map somewhere. This is about real services in real-time, which are going to rely fundamentally on automation and understanding what role that's got to play.
And think about food safety, for example. If you're using visual recognition. We've seen many recalls of late, whether it's baby milk being infected, whether it's cheese, chocolate, whatever, because small bits of plastic going into foods. Well, actually, CSPs can use visual recognition to do that. But this is where automation is going to be key because actually all of the machines or the visual recognition with the communication connectivity is all going to have to be fully entwined and working together to deliver that value of service. So yeah, the enterprise opportunity is what's so exciting because it's going to unleash this whole new range of services and it's actually going to help the enterprise realize what it is that they want. So it's not about transformation only for the CSP. And I think this is what people sometimes get a little bit lost at, Michael. This is about understanding, yes, it's transformative technology for a CSP to have new business models, to offer new services, but actually for the enterprise, it can fundamentally change the way in which they do business as well.
MH: And it allows the CSP to provide what's referred to as a zero-touch operations center. Again, the idea that you don't have individual engineers tweaking knobs and turning dials here or there because the clients may very well have different needs at different times and you need to make these adjustments on the fly.
ABP: Absolutely. And this is why we talk about automation. All we talk about is networks, AI, it's about zero-weight, zero-touch and zero-partnering. That's that whole idea is that automation is going to give us these great benefits. It's in real-time that you can get the service which you need. So, for example, if you're in manufacturing, there's going to be certain times of the day where it's more economical. I don't know what it's like for you guys, but here in the UK, the news everywhere today is about gas prices, astronomical prices, and there's some industries that are going to have to close down, but actually AI and automation gives you new opportunities because you can switch the machinery on, not on peak hours. So actually, rather than having a lot of production during the day, you can switch it to production at night.
So that just gives you some idea of how automation can allow you to really start to adapt to some real, whether it's energy consumption or whether it's consumer needs. At different times it makes more sense to maybe produce or to offer a different service. As we always say, electricity, consumption is another one. When people watch, it's a bit different today with on-demand television, but everyone used to, between in the UK coronation street, a fairly famous soap opera, soon as the adverts come on, everyone's kettles went on and energy consumption goes up. And it's the same principle. And that's why I give you some of these other examples because you can apply it to any use case for what you want to do with that.
It's about on-demand, zero-weight, zero-touch. If you want to offer a new service, you want to configure something, you don't want to wait two weeks for your engineer to come in to fiddle, as you say, with the knobs, coming in, changing things. You need the right consumption. You need the right bandwidth. You need the right analytics tools, all of those things in real-time. And that's where AI and automation are going to have a really important role to play. And I'll give you one example. We've been working with one CSP, on rollout of 5G base stations. And through automation and AI, they've been able to save between 10 and 15% on power reduction. That's a huge amount of money. If you could think between 10 and 15% on power reduction on 5G base stations. Well, think of everything else we can start to roll out with automation.
Another CSP we're working with, we're able to reduce manual activities by 30% and that goes back to what you were saying, Michael, that stops every time you want to upgrade your service, have something real-time sending an engineer out. Actually, it could be zero-weight, zero-touch, zero-partner.
MH: While the customer experience is often cited as the area that CSPs are investing in with technology for AI, like chatbots. I'd really actually like to get your insight into areas that aren't just as critical. You mentioned the beginning of the whole process, nobody puts up a 5G tower today without leveraging machine learning first.
ABP: No, absolutely. And I think this is the types of services that we need to understand about our customer better. So with 5G, we'll say, "Can my customer actually have this service?" So, like you said, do we have the right infrastructure in the right place? What is the coverage like? Does it give you the latency that you need for these services? Because with 5G, this is going to be really important when we think about ultra-low latency and the variables from one millisecond to four milliseconds, are going to be really critical, depending on what that use case is going to be. Whether you are launching a smart port; whether you're going to do healthcare. I mean, healthcare, is going to really rely on those milliseconds of latency.
And that's why, it's about, across that whole journey of buy, sell, operate, that whole life cycle is where automation is going to have a real critical role to play. And I think people have to stand back and say, "And what else can it do for us as well as a customer experience?" Well, the green agenda is on top of everybody's mind. We've heard about the climate emergencies that we've had. A lot of enterprises, a lot of telcos, have green initiatives. And again, AI and automation are going to have a critical role to play in that we've seen in data centers, we consume huge amounts of power, as well.
And actually, I saw a really interesting presentation given by the CEO of Colt last week and she was citing some stats. And if the telecoms industry was a country, , it would be the fifth largest polluter in terms of emissions. So that's where AI and automation has a really important role to play. As I said earlier, reducing power reduction-in base stations, reducing energy consumption in a data center. So it isn't just about a customer experience in a traditional sense; it's actually about that whole right infrastructure, right place, for right service, but also enabling you to also operate services at the cost point that the market requires.
MH: 5G also expands the attack surface to the point where it is impossible to manage security without machine learning algorithms, but security isn't a box that you buy at a tech store,you stick on a shelf, you plug it in Do CSPs understand the cost of AI-based security, isn't simply a one-time purchase?
ABP: Yeah. And I think CSPs, have gone through a great evolution in terms of their security. It's more now about security by design, security is embedded within every single thing. And there's a lot of complex questions, though, Michael, that have to be solved around that issue. It is about security by design. It is about it being embedded. We start seeing AI controlling AI and that's adding a lot of complexity. And I think with security, we've got to really understand, we still have those questions with AI, public versus private cloud for certain algorithms. Where it's going to be based? What does security look like within those different operating environments? So, and CSPs have been going through virtualization for a long time.
So it's not only about AI, where security is important. It's this whole modern infrastructure, as we migrate to 5G. We've got much more virtualization in there, but yeah, I think CSPs do understand. We've started to see regulation on world fining for security violations. Customers are incredibly sensitive about their data today. So there's actually been a change in what the customer expects, which also makes it even more important. But yeah, it's not a one-time fix with AI and security. It's got to be embedded in every single thing that we do. And it's going to be that first question. I think that every CSP wakes up and does on a daily basis, is my AI secure? Can we govern it properly? Because the challenge with AI, it's non-deterministic.
It's even more fragile than traditional software because of the way it is going to work and because of business models and people aren't going to give insights into how the algorithm works. They're actually buying the algorithm. That's where the IP sits. So it's got to be about explainability of AI, understanding it in real-time, and understanding what are the security parameters around it. Many questions still to ask around security, but yes, CSPs, really, I think if anything, in AI, I think security is front and center more than anything else. And in my role at TM Forum, I constantly get asked about security and explainability.
MH: That explainability is critical. You can't have a black box, that's spitting out alerts and not knowing why.
ABP: No, absolutely not. And this is a big area where we're going to see regulation playing a big role. And at TM Forum, one of the big things we do is we have a whole collaboration group that looks on AI governance, which fundamentally is, how do you explain AI in real-time? How can you use things such as really understanding, using APIs, for example, to discover what data's being used? Why are the decisions being taken? We're looking at things like AI model datasheets. So we're saying what is a minimum amount of information we need to collect on that data? Now, these are big industry initiatives because these are things we have to solve as an industry. We have to understand and explain in real-time, that was the data that was used.
And therefore these are the decisions that were made as a result of that AI, because explainability comes into everything from bias. It comes back into the security questions that you're asking. I mean, because it's non-deterministic, AI is updating in real-time. We have to be able to explain those decisions because otherwise, we're going to have whole systems shut off. You can be locked out of. You're going to have systems going down and also you can get that creeping in of bias if we're not careful. So explainability is, I think, center of everything that we need to do.
We need to understand what the data is, what the explainability is. And also, with the API suite, we've done the contract management one, it's about, what do we expect the threshold to be of which the AI can act in? And what actions should be triggered if there's an unexpected event? So you have to think about your software engineering, a little bit different in terms of, what actions are triggered if things happen within certain thresholds. And that comes back to that explainability question, because if not, you're going to see multimillion-pound fines coming in across the industry and that we want to evolve because there's going to be no trust if we're not careful.
MH: You mentioned bias in data sets within artificial intelligence. And this is a critical issue within the artificial intelligence community. As we know, the most common example that's cited is the financial services sector, and that some people weren't getting loans because of the color of their skin, not because they were or were not qualified. And if you look at that historical data, it would say some very horrible things about certain segments of the population. If you just let it run wild without cleaning that data, what are some of those issues within the artificial intelligence community? What kind of bias do you to remove within telecom?
ABP: Yeah, I mean, there's a huge amount of bias that we need that we need to look at. And I'll give you another example because when we start to think about, so obviously, race is one that has been looked at in the financial sector because of bias there. And if you think about healthcare, that was another great example where there was three hospitals testing, looking for tumor growth. It's a very well known case study. Certain hospitals had red tags, some had blue tags, some had green tags; the AI machine algorithm was working. And it worked out that people with red tags were less likely to have cancer. That was actually because of their access to healthcare because again, there was a demographic issue underlying what that is.
So that happens across all data, across a telecom network. And that's what we have to understand. What do we mean by bias? And that could be anything from looking at detection of base stations. We're using visual drones, visual recognition technology to look at base stations. Well, not all base stations are configured in the same way, right? There's different vendors with different technology stacks, different color wires, for example, on the way that different systems are connected together. Bias can creep in absolutely everywhere and again, you can have fault diagnosis going wrong again, because of bias. And that's why the medical one's a good example because whether one vendor uses one color lead for one certain area of connection versus another, you're going to very easily get bias creeping in time and time again. So it will affect everything from same use case.
It can affect things from how this service provider offer financing for a handset or can they be postpaid or prepaid based on certain demographics? It's the same with, is my engineering going to suffer because actually we've only trained it against one certain vendor's equipment versus another one that we have? So bias can creep in all sorts of horrible ways. And that's why we constantly must, really reassess that data we have; as I say, it's not just about the customer, it's actually that bias can creep in, in terms of your service quality, in terms of the zero-weight, zero-touch, all of that stuff we're trying to do with AI. Bias can creep in there as well.
MH: You've said in the past that we talked too much about journeys when all the customer really wants is reliability. Explain that in more detail, have CSP has been investing in the wrong areas of customer experience?
ABP: Yeah, I would say absolutely great question. I'd say that's absolutely something we see time and time again. So call centers were very costly and you didn't always get the best experience. So the first applications of AI was saying, "Okay, well, let's get chat bots." Well, the experience of those weren't particularly great anyway, but this is the challenge, Michael, why are we using chat bots? Why does it not just work? When you've order something off Amazon, it's very rare something goes wrong. You order a book, the book comes through the post the next day. I order connectivity, I want connectivity. I want the service that I offer. So this is why with AI, it is about predictive maintenance, it is about ensuring quality of service. Wouldn't it be great, rather than me having to go on a chat box, let's say there's some maintenance in the area happening between 10 and 12 tomorrow morning.
I'd love to have "Hi, Mr. Boasman-Patel, really sorry, but tomorrow between 10 and 12, we're going to do maintenance. Don't worry. We are right on it." And you'll be back and give you any updates. Then I don't need to go on a chatbot and say, "Hi, my service isn't working." It's these types of proactive experiences we can get through automation and AI. And again, if there's five different routers, in my area, they can reroute me to a different box to make I've still got a good quality of service, particularly as we're all working from home now.
MH: Oh, you bring up chat bots. They're all the rage, but they can be rage-inducing. How do we execute a chat bot correctly?
ABP: Yeah, I think that's a great question because I've certainly, have been rage induced several times trying to look at a chat bot. So there's several things that you can do. And again, this is where diagnosis becomes really important. So with my service provider, I'll give you a very, very good use case. And this is where the important things like APIs become important. And then also doing things like looking at automating your dashboards and doing things for service revolution and automated workflow. So time for me, lovely, got a text message. "Hi, Aaron time to update your phone," have a new service I went through, tried to do all on my phone app and then I kept trying to pay. It wouldn't let me pay, nothing on my credit card. It wouldn't even let me get through to the payment section.
So there's a problem with my certified payment API. It should know that already, and it should redirect me through. The same principle when we're trying to talk on chatbots. If a computer says, "No, if you don't understand, you then need to direct me straight through to an agent." So it's about understanding what those thresholds are for customer pain and customer frustration. Chatbots can be great things to upgrade. And that's what mine was through. It was actually through a chat bot to upgrade. But when it didn't work, there was no help. It just didn't recognize it.
But also surely on the backend on the operation system, it should have come on the BSS saying, "Oh, there's a problem here with billing." And it should have flagged through to the operation center and it should have been a quick fix. Instead, I think it took about a week to get done because it took be a week to upgrade. So that is how chatbots really need to be integrated with your backend systems. And I think today, chatbots have been seen very much a frontend customer experience as a way to get in. We haven't seen full integration end-to-end with the operation system and that's where we've got to get to.
MH: What skillsets are the vendor partners bringing to the 5G table to fill in the missing blanks of expertise at the CSP level?
ABP: I often sit here and talk a lot about CSPs and the changes that they're doing and their skillsets, but I've got to say the vendors have gone through an amazing revolution change themselves. And what we're seeing is different sizes of service providers looking for different skillsets. So some service providers, Vodafone is a good example. It's public, they've just recently recruited 5,000 new software engineers to come bring on board, look at AI, but they've also recognized we can't do it all ourselves. So they're looking at a hybrid of, we'll do some things in-house, we'll partner on other things, and looking for the vendors to find those skills, which they don't have. There's other operators that I work with that say, "Hey, we don't want to do this ourselves. We're too small. We don't have the finances. So we need our vendor partners to look at doing that."
So what I've seen is really a lot of vendors starting to step up and say, "Actually, we can provide this service in these areas and offer this skillset." So I remember when I started telecoms about 15 years ago, it was all about outsourcing. A lot of managed services, operators brought it in-house and now with AI, we're starting to see a mixture. And again, that mixture starts to mean we've got to have a mixture of skills, so big areas in there. Data science becomes really important, that is a skillset that we really see a big need, coding is another area. Not everybody's going to be a great coder, some people...
And then, you've seen vendors bringing some great innovations around low code. So actually, they're giving people who aren't maybe the best coders, the power to do some coding, because they're actually brought in some low-code areas in that space. So I would say, vendors are stepping up, but I think the biggest thing is every service provider's going to have different needs and at different times and on different applications. So I don't think anyone can say, "Right, we're going to take this approach to AI." I think it's going to be so use-case-dependent. And I think some of the big guys have got a really unique role to fill because you not only work in telecom, you've got many other verticals. And I think the integration part, understanding what those other enterprises need, is going to be a critical, unique selling point.
MH: Let's come full circle on everything we've discussed here. To you, what's the secret to a successful AI deployment in a large-scale enterprise?
ABP: It's about scalability. And I think this is what success looks like. It's about understanding what you want to deploy. It's about understanding the needs. It's about applying that to your organization and getting everybody on board, because what we've seen, we see very fragmented implementations of AI and automation to date. It's been very siloed. The rest of your organization don't understand what the benefits are. We often understood what the benefits should be. What does good look like? So to me, Michael, it's going back to basics. And that's what a good AI deployment looks like. Is about saying, "this is what I want to do with my AI. These are the metrics which I want to measure it by, and here's the outcome and we are doing that today." Believe it, or not, we haven't seen many deployments that look at that.
And chat bots, I think was a great, easy example. Where you can say, "We've done a terrible job at doing it." We weren't measuring the right outcomes; we weren't measuring the right experiences. The whole organization weren't brought in because the CM team was doing it. It was the call centers producing costs. It didn't integrate with the OSS, with the BSS. And we had challenges.
So it's about, for whatever that deployment is, seeing it through end-to-end, and I have to say time and time again, you have to work with the right partners. You have to really understand what is it you are buying? Are you buying expertise? Are you buying software? Are you buying a component? And how is that going to work and be managed? Because with AI, it's not good enough to say I'm buying that algorithm from you. And we're going to deploy on our network or the AI module, make, it smart, AI's too complex to work like that. For all the reasons we've talked about in this podcast, from everything from security to integration, to understanding what's going on. So it's got to be back to basics.
MH: Aaron, this has been fascinating before we let you go though, tell me what excites you the most about the deployment of AI in a 5G world?
ABP: What excites me, the absolute most of deploying this is the fact that CSPs can really now take advantage of 700 billion of brand new revenues, right? We've talked about what is the role of CSP for years? We've talked about digital transformation. I spend all the time in doing that. Now is not the time to talk about digital transformations. Now is the time to do it. Now, there is no more excuses to wait. 5G is going to unlock so many new services. 700 billion in the enterprise space is desperate for transformation. We've seen through the pandemic, enterprises and businesses that have changed have grown their revenue seven times larger than those who haven't. So what we want to do is say, "Now is the time for 5G." You've got 700 billion, take your slice of the pie and deploy AI and automation to do it because without AI and automation, those 5G opportunities are not within your reach.