To embed our video on your website copy and paste the code below:
<iframe src="https://www.youtube.com/embed/0QYTLiYIZtk?modestbranding=1&rel=0" width="970" height="546" frameborder="0" scrolling="auto" allowfullscreen></iframe>
This session is all about data foundations for AI native networks. As we know, telcos possess extensive network customer and operational data yet as we have learned at TelecomTV on our reports, two thirds site data integration and management as their top AI implementation challenge. They say data is fragmented across divisions, domains, and vendors with inconsistent formats and governance and that this silo data is preventing AI value. It's blocking real-time orchestration and it does make unified customer experiences difficult if not impossible. So with that said, let's meet our panelists and just like Ray did in the first two sessions, I'm going to ask our panelists to briefly introduce themselves starting on my far left with Vish.
Vishal Mathur, Telecom Infra Project (01:18):
Hi, good morning everybody. My name is Vishal Mathur. I'm the chief commercial officer at the Telecom Infra Project. Thank you.
Régis Lerbour, Infovista (01:25):
Good morning everybody. I'm Régis Lerbour, VP of Product at Infovista. Been the industry for about 20 years.
Itsuma Tanaka, NTT DOCOMO (01:32):
Hello everyone. I'm Itsuma Tanaka from NTT DOCOMO. Nice to meet you.
Diego Lopez, Telefónica (01:38):
Diego Lopez from Telefónica since two months ago I am now responsible for data and enablers, which is something that is
Guy Daniels, TelecomTV (01:47):
Highly appropriate and thank goodness for that job move. That's great, Diego. Welcome all of you. Diego is our DSP co-host for this session. So before we begin our actual panel discussion, I'm going to invite Diego to take to the podium to deliver the DSP Leaders address.
Diego Lopez, Telefónica (02:11):
I'm not sure if it is totally true, but there was an advisor to a US president that once talking about what was important for elections, et cetera, made a very short phrase about saying, "It's the economy, stupid." I have used that in the past in a couple of opportunities, among other things, because it's extremely useful for calling stupid your senior management, for example. And this is something that is always something that you can get some advantage from, but it's true and if you think about it when talking about AI, AI nativeness or whatever, it's about you can summarize it or I both summarize it as it's about data, stupid. It's a core thing. It's a basic substrate that you need for making anything happen with AI in networking or in any other reasonably complex enterprise. And for that, what we are trying to have is thinking about how to build data architecture, data solutions, data infrastructure that in my view has to have three main properties.
(03:19):
One has to be open. It has to allow you to plug and unplug different elements in a way that allows you to freely play with different models, different sources of data, different consumers of data, not to forget that AI models, inferences are at the same time consumers and probably sources of data for other elements. For these, for example, we are working in ETSI in building these and on standardizing this and building the necessary standards and ontologies, semantic models, et cetera. Second, it has to be distributed. The network is distributed by itself. We need to be prepared for not having centralized data mechanisms of whatever the nature. I'm not talking only about having a single data lake or data repositories about having distributed and different models and interfaces, et cetera. And finally, most of all, it has to be trustworthy. You need to trust the data.
(04:22):
You need to trust the data you're consuming and you need to trust the data you are producing. And for this, basic is identities, identity everywhere. We need to identify every single piece of the machinery. People as we have done in the past, data sources, data themselves and machine and whatever, call it whatever, the module, the model, the LLM, whatever. And this is something that we are trying to experiment with on how we build these reliable trustworthy AI identity based system in a couple of research projects. We are currently involved. For example, one that is very much focused on this, applying this to security and securing the use of AI like iTRUST CG. With these three properties at hand, I believe that we will be able to build the data substrate to enable in the, well, whatever AI native telco is, which is something that we have to sort out.
(05:27):
Thank you. Thank you.
Guy Daniels, TelecomTV (05:32):
Great. Thanks very much for setting this in there for us, Diego. And let's open up the discussion to our panelists. I mean, you've very succinctly laid out the open element, the distributed element, the trust element here that are so important. But I'd like to start by just trying to understand just how fragmented telco data is at the moment, because we hear about it a lot, but just to what extent it is and the problems that might cause it, is data that fragmented that it causes real world problems?
Itsuma Tanaka, NTT DOCOMO (06:07):
Yep. Well, let me explain, first of all, how DOCOMO deals with this. We are trying to build one big common data foundation like 14 petabytes environment and network customer domain are generating more than 200 terabytes per day and they are also coming from like more than 70,000 base stations, everything. We try to put everything in one foundation. So in a way, we're trying to remove all the silos fragmentation. And what our recent learning is that what is interesting is the data normalization is not really the problem. I mean, now we can have AI tools to align, normalize those data. What is the biggest challenge? The biggest challenge is actually the operational understanding, how people access the data, how they do it, and the meaning, the semantics behind data, which are traditionally in our human brains. And so what we see is biggest fragmentation is the data itself and what is in our brain.
(07:28):
That's our recent learning.
Guy Daniels, TelecomTV (07:30):
Diego, I mean, that resonates with you as well. Yeah.
Diego Lopez, Telefónica (07:33):
No, but it's my breathing. No, that's true. And I would say to summarize that is the ... We have the data architecture fragmented because we have the net architecture fragmented as well, because it's the difference in technology, organization, the historical background, how networks have evolved and ground is very difficult right now. We have the people from radio. The semantic field that people are playing with radio have is totally different from the semantic fields of the people that are dealing with optical. And then you go to the BSS. That's totally another story. Breaking those silos and finding the way of making some sense out of it is I believe is a big challenge. We need the models. We need the common semantics that would allow us to build that and relate to one set of data to another set of data, but it's important as well that needs an organizational push that is extremely important as well.
Guy Daniels, TelecomTV (08:45):
Great. Organizational push. Well, let's look at industry-wide push. Vish, if I can just come to you next and then Regis, I'll come to you next. TIP is doing interesting work here, but as you understand it, you've already identified that linkage between the fragmentation of data and an OPEX and actual real value.
Vishal Mathur, Telecom Infra Project (09:05):
Look, I mean, from our perspective in TIP, and I'm not trying to ... I mean, let me build on what's been said previously in the other sessions. Amol in TechM mentioned global federation platforms as a enabler and mechanism. And Georgee in the previous session talked about intent-based and then closed-loop automation techniques. I think generally speaking, if you look at the telco industry, there is a good 60, 70% of OpEx burden from IT and network systems because of the fragmentation. So we recognize that and there are mechanisms now to try and channel data and harmonize data, whether it's using off-the-shelf tools from Snowflake or Redshift or whatever. But actually what we feel is really important is to have very specific use case driven approaches that actually then leverage a global federated set of language models and use the data in the right way. So if you end up doing that and then you can be very specific about ... And maybe TM Forum has got a very good landscape of fault resolution, energy saving, these types of use cases.
(10:21):
These are common across the industry. So effectively, let's create a standard approach to trying to work out how do you use these platforms to really drive forward a level of incremental benefit. And the benefit has to be business related. There has to be OPEX burden release because you're being much more specific about which domains, which scenarios, which use cases you're trying to drive forward. And then from there, it releases a little bit of operational pressure for the operator to then drive forward new outcomes basically.
Guy Daniels, TelecomTV (10:56):
Thanks, Vish. And Régis, what are you seeing from conversation with your customers when you're developing your products and solutions? What's your view of the fragmentation at the moment?
Régis Lerbour, Infovista (11:05):
Well, whether we think that the data architecture itself is fragmented or not, I agree with Itsuma, we have big data lakes, all the data goes in there. So in a sense, it's not fragmented. What's fragmented in my opinion is the awareness of the data and the understanding of the data. We're lacking expertise and that's all back to the people. And we've been talking about the people this morning already because we have teams that are extremely siloed. They understand very well. They actually own their data, they own their processes, but they don't share. They don't expose this data to other teams. Other teams are mostly unaware and that's something that we see on a daily basis with our customers, whether we work with the planning teams, the validation teams, the operation teams, the customer experience teams, they know the data that they have to use to do their daily job, they ignore the rest of the world and that's what has to change.
(11:53):
So the big question is that's kind of a chicken and egg thing. Are the teams siloed this way because the systems and the processes are built this way or the systems and processes a consequence of the organization? And either way, we need to break these silos. And our approach, and I agree with Vish on that, is we need to break those silos step by step. Let's start with the team, the processes, the ecosystems, implement use cases. So step by step, work on processes that bring value to create momentum and incentivize all the teams to go after the next use cases.
Guy Daniels, TelecomTV (12:25):
Does it matter where we start on this? Because you just said there, let's start with the teams. Let's start with the people. Is it important to start there or is it up to individual operators in their own situations? Diego?
Diego Lopez, Telefónica (12:39):
Well, I would say that in general, more than with the people is about the incentives that people have.
Guy Daniels, TelecomTV (12:45):
Incentives, right.
Diego Lopez, Telefónica (12:46):
Because the point is that right now, and this is something that is not specific of using AI or it's not specific of our network providers. In general, teams are assigned goals and their performance is measured according to certain objectives that very often are conflicting with the objectives of the other team. And in that situation, it's very difficult. Whether you make it automated, whether you make it manual, the only difference is with the automation, the loop is much tighter and then you notice it much clearer way that's limitation. It's difficult. I believe that it's possible and I put something that if you would love to hear about any examples of how you can break this by example. And this is something that we are trying to do and it's about showing people that by collaborating they can improve and they can achieve their goals or they can show us their management that their goals can be better achieved that way.
(13:53):
Fighting by example is currently because the other approach is the typical top down that the management says that you have to do this and you have to collaborate and then well, then people start to, well, reluctantly start to share something, but showing an example in which you can show that even in the case that is something that is in a limited environment, the thing is something that's very valuable and if you have any story about this to share will be really, really interesting, I would say.
Itsuma Tanaka, NTT DOCOMO (14:25):
Can I add? Oh, go ahead. Thanks. On top of that, building on top of that, I think the first thing to start is visualizing, verbalising every single process and steps in our value chain and how we operate. I mean, telecom network business, it's a huge business involving different departments and different steps and different departments, different operation teams. And sometimes they have different KPIs and they kind of clash. So that comes to the discussion incentives, but well, a lot of us engineers know about data where we can find data, where we can find they're nicely documented most of the time, but how we do the work is not really documented. And clarifying those steps step by step, I think that's where we are working on and the first steps we need to do.
Guy Daniels, TelecomTV (15:25):
Okay, that's great. So Regis, you want to come in as well?
Régis Lerbour, Infovista (15:28):
Yeah. I just wanted to bounce off your question about examples on how we create incentives for the teams to work together actually to exchange data sources that they have access to. To me, there are three different types of incentives. One is providing insights on the network, whether it's network quality customer experience, even to people that are not deeply familiar with how the network works, like the CXOs now mostly interested in the dashboard kind of KPIs. Well, AI can do that and that saves a lot of time for them and it create trust to them on what AI can bring. Second aspect is on automation, obviously, and there are tons of use cases where automation brings time savings that everybody will understand. Think about root cause analysis. We work with teams in the RAN that spend an amazing amount of time investing in issues that do not occur in the RAN.
(16:22):
The root cause is in the core. So AI could tell them very quickly, "Don't even investigate this one because it's in the core. Pass it on to the core team."That's a problem solving two minutes. And the third one is monetization. There are also incentives about how do we create new revenue sources through monetization and there are lots of opportunities between the OSS and the BSS. Can we create some kind of upsell potential promotions in the BSS based on data that sits in the OSS right now? It's impossible without AI connecting the dots and without having the right data formation for that. So lots of opportunities to create those incentives in my opinion.
Guy Daniels, TelecomTV (16:58):
Okay, great. Vish, do you want to come in?
Vishal Mathur, Telecom Infra Project (17:00):
I just want to build on this wonderful sequence of conversation. I think there is value in some level of standardization and commonality. So if we followed this pathway and we just talked about like Georgeia's example in the last session around China mobile really driving forward product, scale rollout and deployment of fault resolution. Now you could actually perceive that to be a similar sort of problem set challenge, whether it's RAN, transport, or across domain, across multiple operators. Now I think from my perspective, there is an ingredient set of bringing a level of collaboration on the techniques that can be employed at a minimum viable level. What are we talking about? What is anomaly detection fault resolution? What are the scenarios that need to be talked about? There can be some sort of requirement set that can be surfaced out and suppliers and app developers can develop stuff that are very specific.
(18:01):
You have the data itself required across those platforms that need to be surfaced up. Now every single operator will have their own language models. They'll need to keep their data sovereign, resident, but we need to federate that the intelligence behind it, the training weights so that actually what manifests above are the enablers, the actual MCP or A2A connections that are required to interface upwards. You can build digital twins in there and test out the applications, but in general, this is what we're doing in TIP. We've got AT&T, Telefónica, tellers coming together and we'd like more. So if you do that and you work on a very specific set of use cases, you can start to see the incremental benefit of really building out a platform approach in this way and start to think, go from pilot to scale in a much more efficient way, working with the vendor ecosystem and the application ecosystem in the right way.
Guy Daniels, TelecomTV (18:58):
Well, as you said, good pilot to scale. I mean, there's been a lot of linkage between the sessions so far this year's event. I mean, we started in the morning, I was just taking some notes here Amal spoke of the federated data model during the first session
Itsuma Tanaka, NTT DOCOMO (19:12):
And
Guy Daniels, TelecomTV (19:12):
Then Francesca was talking about the need to be data driven before you can be AI driven. So we're talking themes here, but can I move on to the unified data architecture and that approach, whether it's data lake, data mesh, data fabric, whatever it is you might be talking about, will it really enable real-time access across domains, across customer network domains?
Diego Lopez, Telefónica (19:37):
Well, as long as it provides two things that they believe are essential, one is a semantic layer rather than a data model. I mean, if we're trying to properly speak in technical terms, data model has more to do with the syntax of the data, how the data structure and how it is presented rather than what those data mean. Having a unified semantic model that allow you to make the appropriate transformation recently, just an example, we have been playing two different teams in our lab. We're playing with different digital twins of the network using different technologies. And the problem is that we found that the way of expressing the topologies were not the same. So the only way that we found that to make transformations that were dynamic was with a semantics topology model that would help us in translating from one environment to the other and back and forth as it was required because of the characteristics of the ... And that has provided us a very good experience on how to do this and notice that with this, it's real time.
(20:52):
I mean, it's not pure real time. We're not talking about milliseconds. If you're dealing with semantics right now, probably in the future it will be, but right now we're talking about seconds, but it's about seconds. It's something that is very close to, not to network real time, but very close or close to or within the limits of human real time. This is one thing. The other thing that is essential is a proper governance and the proper governance that goes is not about you have access to this database or you have access to this set of records. It has to be a much deeper level of detail, a much finer granularity that allows you to decide what to do with the data. And to break this silo, something that is very important is that the people that are in the silo feel confident that the data is not going to be misused, applied when they would not like it to be and facilitate this intercollaboration with other operators, with vendors, with regulators, with whoever.
(21:55):
And that requires this proper governance, it's very detailed governance.
Guy Daniels, TelecomTV (22:00):
But this is operator specific at the moment, isn't it? What's the industry doing collectively to help operators here? I mean, Vish, speaking a week ago about this and the fact that there seems to be no industry standards, it makes things complex.
Vishal Mathur, Telecom Infra Project (22:19):
Yeah. But I mean, hence it's too big a problem to solve in its entirety, hence why we're taking a very specific use case driven approach. What comes out as a result is what level of data granularity density is required and what level of standardization or harmonization is required. There's enough tooling out there to be able to then make sure that that data is purposeful and relevant. Because if the end goal, if the business outcome is to try and show there's cross-functional working, there are business outcomes that are being measurable, that is achievable, then as an operation, you've got to work on the specific functions that need to be transformed and therefore what's the data required. All around that is privacy, governance of management of use of that data. These are the ancillary elements. Why is there nothing being done so far? Because everybody looking at it is a big problem.
(23:22):
I think you need to just focus in and have very specific functional use case driven approaches and build from there. Make sure it's not just POC, go for scale. Create the model to try and drive that. Work with your vendor ecosystem and drive that through.
Guy Daniels, TelecomTV (23:38):
Régis, do you have a similar thought that maybe we are looking at too big a problem here and it needs to be broken down?
Régis Lerbour, Infovista (23:46):
Oh, absolutely. I have to say, I don't know if the approach you want to pull in place I seconded it would be great if there was an industry-wide initiative to make that happen. Whether that happens or not, I fully agree with you. There has to be a very dedicated approach that is use case based step-by-step. Identify use case that brings value to you. Don't care about what the others are doing. Just identify a use case that brings value to you, whether that's savings, whether that's revenue generation, et cetera, and then go full speed, step by step, implement it step by step, demonstrate the value, incentivize the next steps and keep the human in the loop initially. We'll go for full network automation down the road. Don't shoot for that. Just keep the human in the loop to gain trust from the teams.
Vishal Mathur, Telecom Infra Project (24:30):
Can I mention one more thing? Don't forget there's compute investment in all of this. I mean, I'm sure NVIDIA would love to be selling more GPUs and Intel is here in the wings as well. But the whole point is, I think as an industry, we are investing a lot in compute infrastructure and that is to drive forward big, large language models that can really surface up and develop out an understanding of data. Surely there is a much more incremental approach, which is much more purposeful and you use that compute infrastructure to really drive forward outcome based sort of stuff, basically. That's where I'm heading in my head.
Guy Daniels, TelecomTV (25:09):
Okay. Yeah. Diego,
Diego Lopez, Telefónica (25:10):
Were you- When it comes to these big, large language models, something that is important as well is that you can incumbents of that, and I know that this probably is not the most popular opinion in these days is that, well, the heavily centralized language models that requires a huge amount of computing power and power itself. Well, they will be good for certain kind of applications. I doubt very much that they will be able to deal with the kind of tight, really predictable and reliable behavior that we should expect for an infrastructure that is used to run on 59s. And we will need to find other mechanisms for, this is different, for automating the network, not for example, for customer relationship. Yesterday I had to call for a medical appointment and then I go to voice saying, "My name is Sophia." It was clearly a synthetic stuff.
(26:20):
First time didn't work because probably because I was not clear in my voice, it was very early in the morning, my voice was not clear, but then I can try and that's no problem. I don't feel like ... But in the case of something that is five nines and critical infrastructure is totally different. We should go for different approaches, probably smaller autonomous models that are very much focused on a particular matter and that can be controlled by these language models. And it's about building some kind of, well, I don't know if it will be a hierarchy or at least a layer structure in which, well, the current LLMs will play a role, but the ones that are going to play the real role probably is that those agents that are to come and that are not still there because right now an agent and LLM, well, that's another
Guy Daniels, TelecomTV (27:13):
Story. That's another session. We've got a few more those services coming up, but it's all interlinked. It's all interlinked. All this requires compute as well, as you just quite rightly said.
Diego Lopez, Telefónica (27:21):
No, no, definitely.
Guy Daniels, TelecomTV (27:21):
Yeah. So if I can come back to this unified data architecture, will it unlock AI at scale and what business value will it really drive for the networks and also for customers? You saying it's two separate things here, but what business value is this really driving having a unified data architecture?
Itsuma Tanaka, NTT DOCOMO (27:45):
Let me share one example. It's from our sales domain example because we try to build a big unified cross-functional data foundation and how we build it is that our salespeople can access and can build apps around it. So automate and for example, let's take enterprise solution proposal and the sales guy can access like how is the network situation around their building the customer insights that we collect for the customer and their customers as well. So everything together and they make the package, but before this data foundation, people could do this but simply asking. So we simply took weeks or months to consolidate everything and make it a nice package for your sales. But now with this, productivity become like 10 times before. So that's definitely their value, right? More productivity. The sales results is definitely contributing and the more salespeople are happy, the people around them will be more happier to provide more data into this.
(29:02):
The collaboration, that's what is happening.
Guy Daniels, TelecomTV (29:10):
Good example. Yeah, thanks for use case.
Diego Lopez, Telefónica (29:13):
And I would say that there are a couple of other things. One is that the customers can benefit themselves of accessing certain amount of data.Think about the typical situation in which you are feeling that you are under some kind of attack and you can derive or identify which kind of attack and what could be the measures that you can apply locally apart from the operator helping you, et cetera, derive from data that comes from the network and then from the traffic, et cetera, with all the proper caveats about being not disclosing relevant data, et cetera. But this is something that can be extremely helpful. This is one thing. And second is something that is one of my personal endeavors right now is trying to convince my company, among others, that there is a value In doing this, which is about eating your own dog food, et cetera, is about if we develop proper data architecture, proper data infrastructure, it's something that we could provide to our customers as well.
Itsuma Tanaka, NTT DOCOMO (30:11):
Okay.
Diego Lopez, Telefónica (30:13):
You want to share data, you want to share data with your customers, well, with your competitors? No. With your partners- There are limits. Or whatever, exactly.You want to share the data? Well, come. We can facilitate ways of doing it in an open, distributed and trustworthy way.
Guy Daniels, TelecomTV (30:34):
So it becomes a revenue opportunity, service,
Diego Lopez, Telefónica (30:36):
Digital service. Once we sort out, we're probably not sorting out the whole problem as Vish has said. Not the whole problem, but when we sort out some issues regarding particular aspects, why not? I mean, we could offer it as a service as well.
Guy Daniels, TelecomTV (30:51):
Interesting. Vish, tell us your thoughts about this, how the business value that might get unlocked here.
Vishal Mathur, Telecom Infra Project (30:59):
Three thoughts. One is I love what Ahmed said about tech excellence. So beyond sales and customer service, tech excellence drives forward internal efficiencies. So look at for those use cases that are really relevant, that are high value in terms of return like Georgee was expressing as well in TM Forum and work for those and look at a collaborative approach to do that. Second is I'd say sort of a provocation to the industry is that you have four like NGMN that are setting out white papered approaches and operating models, people, process tech. You have TM Forum setting out a roadmap to level four or level five. Step to level four and look at those use cases and work with TM Forum community and tip community to try it out and create the reference architectures and the use cases and showcase the business outcomes in that regard.
(31:56):
I would suggest that's my second thought because then from there, rather than working in a fragmented approach as an industry, there is real value in getting some level of movement through standardization commonality and then you can build on top. My third point is there are actual applications that are being used by end users, voice and video, RTC and voice applications taking 60, 70% of the network traffic. Now we can look at application data and network data and look at the combination of those to drive forward better qualities of experience for our end customers. That's another use case that no one's talking about per se yet. We're still internally debating around how do we drive forward optimization automation and adaptive networks. Let's look at those use cases that our customers enjoy, our applications that they enjoy and bring that in as well. And maybe take another level, but those are my thoughts.
Guy Daniels, TelecomTV (32:58):
Great. This is really interesting. I'm going to make an offer to the people who are at the back of the room who we never really get questions from because they're so far in the back of the room and it's so dark we can't spot them. So over the next 30 seconds or so, if anyone towards the back of the room has a pressing question, I think Tony is ... Yeah, I can just see Tony there. Tony is wandering in the back of the room. Any sign yet, Tony?
Tony Poulos, TelecomTV (33:19):
No.
Guy Daniels, TelecomTV (33:20):
No. Okay.
Tony Poulos, TelecomTV (33:21):
Wake a few people up and we'll get them going.
Guy Daniels, TelecomTV (33:23):
Okay. Well, I tell you what, let's move on roof forward, shall we. Let's give this offer to those in the middle of the room. Anyone in the middle of the room got a question at this stage for any of our guests or anything? Any thoughts on the data subject? We're here. Keep it moving forward, Tony. We're moving forward. Look, we're giving everyone a chance here, so we're moving forward. Who have you got at the front? Offer to the front of the room now.
Tony Poulos, TelecomTV (33:45):
It's Dean.
Guy Daniels, TelecomTV (33:45):
It's Dean. Hi, Dean.
Dean Bubley, Disruptive Analysis (33:47):
Hi there. Dean Bubley. How do you see data being federated between telcos and other infrastructure companies? I'm thinking say towercos for mobile, energy companies which might be supplying grid power to the telco network, neutral hosts or in- building system. There's a lot of data there around that relates to connectivity and overall say service reliability. How do you combine all of that?
Guy Daniels, TelecomTV (34:20):
I think a question. Who's going to take it first? Volunteers?
Régis Lerbour, Infovista (34:25):
Well, I will not give a definite answer because that's a very complex question. Of course, that's an ecosystem problem. As we mentioned earlier, to me, that's almost the last stage of that rocket to get there. But I want to come back to some of the challenges of federating that data because they come from different data sources and we tend to talk about data as a generic thing. Data is data, data source is data source, but all data sources are different, very different. They're different in terms of quality, they're different in terms of accuracy, then different in terms of how they're collected, how they're processed, how they are interpreted, et cetera, et cetera. I mean, I'll give you a very basic RAN example. Think about throughput. The throughput you get from PM counters, which is at the cell level, has nothing to do with the throughput you get from a drive test, which is a very controlled environment, which has nothing to do with the throughput you get from a crowdsourcing data which is completely on control and you don't know what application is generating that throughput.
(35:23):
We need to teach AI to use these different data sources, these different KPIs together because there is value in putting them together, but understanding how to use them together without making obvious mistakes in interpreting the data. So to me, data integration is one thing, very important, it's a necessity, but the expertise on the data and what the data means, how it was collected, what are the pros and cons, what are the time constraints of the data? Is this data source applicable to real time AI scenario or these data source more applicable to long-term redesign scenario? All of that is very relevant and is completely forgotten if you don't put the expertise back to the people on top of the data.
Guy Daniels, TelecomTV (36:06):
Very interesting. Thanks so much for that. Any other comments to that question?
Itsuma Tanaka, NTT DOCOMO (36:10):
Maybe talk about Japan's case. There's a government led initiatives to basically bring all different infrastructure data together so that they're hoping somebody will nicely process the data and come up with some use case. It's instilling a trial phase, but there are some people trying to do this. But from my standpoint as a telecom operator, it really depends if we need to federate all different infrastructure data. I mean, probably we don't need everything. As the telecom operator, what we need from power company may be limited. I mean, as far as we are concerned, like how much electricity we're consuming, potential, I don't know, potential disruptions somewhere, energy price, et cetera. So it's limited. And on the other side, for example, like train companies or water companies see as telco as a part of their supply chain providing IoT solutions so they have their own context. So probably, yeah, I don't know if you call this federated, but depending on the use case and their business objectives, probably will have different data models and how people want to access them.
Diego Lopez, Telefónica (37:38):
That's true because the federated world is something which you can drop almost everything. The important thing here, I believe, is that in the past the recent years, there have been a development on common data spaces for particular environments. And there have been some that have been extremely successful like in the automotive industry and the telco industry have been miserably failed in building such a data space. But the idea of finding a way in which those data from different or these joint data spaces can be combined, I don't know if they are formally a federation, but they can be combined, integrated. And as someone was saying, it's about not combining everything. I mean, probably as a telco, you are not interested in every fluctuation of the power in the power grid, but you are interested in certain aspects as long as you as a telco have the possibility of accessing the data under certain conditions and mixing it would be extremely interesting and I believe it's one of the most interesting opportunities we have ahead, this integration of data spaces.
Guy Daniels, TelecomTV (38:56):
Great. Oh, thank
Diego Lopez, Telefónica (38:57):
You. That gentleman seems not to be very-
Guy Daniels, TelecomTV (38:59):
Well, I've got to ... Is that Andrew? Yeah. Let's have a quick question and we wrap up the session with the final one here.
Andrew Collinson, Connective Insight (39:08):
You may regret asking this. Hello, Andrew Collinson, Connective Insight. Why is this so difficult? I mean, so I'm going to be devil's advocate. You're the IT guys, you've got data. The meaning of the data is the key aspect of what this is for. Why aren't your departments talking to each other? Why isn't it easy to create that cross meaning understanding? Why is it so hard?
Guy Daniels, TelecomTV (39:31):
Okay. Why aren't departments
Diego Lopez, Telefónica (39:32):
Talking to each other? Why is so hard? Because the representation of the data in general and the way we are processing it then is full of precondition of implicit assumptions that are not ... I mean, it's not that they're not formally described. They're not even shared. It's like what happens, you remember this issue with some space ship that crashed in Mars because somebody has calculated thing in inches and then the telemetry was in centimeters or something like that. Sorry? You have it. And this is something why it was ... You look at it and they say, "Well, how could this happen?" Because if you are in the US, you think in inches continuously and you come here, well, you're still thinking miles. We
Guy Daniels, TelecomTV (40:30):
Take a hybrid approach. We like the best of both
Diego Lopez, Telefónica (40:32):
Worlds. No, but this is an example, but in general, there are these kind of differences that are implicit and not explicit and you have to deal with this implicit apart from the formats, apart from deals like things like the timing, the position, et cetera. Then this implicit stuff is something that is difficult, very, very difficult to deal, not only in telco, everywhere.
Itsuma Tanaka, NTT DOCOMO (40:57):
Yeah.
Guy Daniels, TelecomTV (40:57):
Great. Great. Thanks Andrew. Look, we've got about three minutes left or four minutes left. So I'd just like to ask each of you in turn, a final question here and we've heard a lot of angles, a lot of interesting thoughts, but what is the main imperative that is going to compel telcos to fix their data architectures and try and defragment their silos? What's the override, or is there one single overriding imperative that will persuade the telcos that they've got to get on with this job? Diego, have you got a thought of what's driving you most? The one
Diego Lopez, Telefónica (41:32):
And only imperative. No, I would say this morning someone was talking about, I don't remember, the OPEX, the OPEX target, reducing OPEX. OPEX is something that weighs a lot to current telco operations and that's something that OPEX reduction would be essential to start
Itsuma Tanaka, NTT DOCOMO (41:54):
With.
Guy Daniels, TelecomTV (41:54):
Okay. Thank you. Itsuma?
Itsuma Tanaka, NTT DOCOMO (41:57):
I think maybe I'm repeating myself, but understanding how we do our business is really the first steps. So implicit things, our mental models that need to come out, but it's difficult, but unless we do that, I mean, connecting data is easy, but how to make it really work for everybody in the organization. We need to really understand people, how organizations work.
Guy Daniels, TelecomTV (42:26):
Understanding. Thank you, Itsuma Tanaka.
Régis Lerbour, Infovista (42:28):
I think different stakeholders will have different compelling reasons to invest. I mean, if we come back to operators needing to monetize more within the enterprises to start with, because there are lots of opportunities that'll be driven by new revenue generation. If you talk about AI, improving customer experience, increasing quality, then it's all driven by churn and the side effect of reducing OPEX. And then if we talk about reducing OPEX, then it's all about AI use cases that they bring automation into the day-to-day processes. But depending on the teams, everybody will have a company reason to go after that.
Guy Daniels, TelecomTV (43:07):
Sure. Great. And finally, Vish?
Vishal Mathur, Telecom Infra Project (43:09):
Yeah, bottom line, it's got to be there's going to be a CXO agenda that's saying, "Ah, the opportunity is to work out how do I make sure I'm part of the AI value stack? And therefore if I'm part of it, therefore what is the implicit financial numbers that I need to be achieving and how do I start to use this tool set that's available in the industry?"
Guy Daniels, TelecomTV (43:31):
Fantastic. Well, we must end it there. Thank you all very much indeed. That's absolutely fantastic discussion. We have to start our next session now, so we're going to start next session straight after this. So we're going to bring our next guest to the stage and while we do that, big round of applause for our guests now. Thank you very much indeed. Thank you.
Please note that video transcripts are provided for reference only – content may vary from the published video or contain inaccuracies.
Panel Discussion
A panel of telecom industry experts from NTT Docomo, Telefónica, InfraVista and Telecom Infra Project examine the critical challenge of data fragmentation preventing AI implementation across the sector. With two-thirds of operators identifying data integration and management as their primary AI barrier, the discussion covers how siloed data across divisions, domains and vendors blocks real-time orchestration and unified customer experiences. The experts share insights on building open, distributed and trustworthy data architectures, addressing organisational silos, and creating incentives for cross-team collaboration to unlock AI value in telecommunications networks.
Broadcast live 19 May 2026
Participants
CO-HOST
Diego R Lopez
Senior Technology Expert, Telefónica and ETSI Fellow
Itsuma Tanaka
President & CEO, Docomo Communications Laboratories Europe GmbH
Régis Lerbour
VP Product, Infovista
Vishal Mathur
Chief Commercial Officer, Telecom Infra Project