AI enables a huge step in Telefónica’s Autonomous Network Journey

Telefónica is one of the most advanced telcos when it comes to network automation. The operator launched a comprehensive Autonomous Network Journey (ANJ) program in 2021 for all its operating companies in Europe and Latin America with the aim of increasing autonomy and transforming its architecture, data management and organizational structure to support AN.

The operator aims to reach AN Level 4 for multiple processes in multiple network domains, including radio access transport, IP, core, IT cloud and fixed access networks. Strong emphasis is placed on operations and management, RAN optimization, IP deployment and transport planning.

TM Forum Inform caught up with Edi Fernando Vieira Filho, Head of Autonomous Network and Operations at Telefónica, to discuss the next steps for scaling the program in the next few years and how it is incorporating artificial intelligence (AI) and generative AI (Gen AI) in its AN strategy.

MD: What is Telefónica’s motivation for starting the ANJ program?

EFVF: Technology is evolving, and we cannot manage the new technologies with the old techniques and processes. The quantity of data that we are generating is just impossible to manage or take advantage of. And we need autonomous processes to do new services, like network as a service or slicing. The idea is not to have another project on top of what we are doing, but to do everything with new capabilities and put the customer at the center. The ANJ program has nine work streams: technology, data management, operational support systems (OSS), full life cycle automation, artificial intelligence (AI), customer experience, sustainability, security and new ways of working. It is a comprehensive program, and we have leaders in our geographies and at corporate headquarters to drive and embed these new mindsets in the company’s strategic plan.

MD: How is Telefónica implementing the ANJ?

EFVF: We have started by building use case by use case. There is no other way. We are doing a lot of things in parallel, exploring new techniques and building copilots. For every use case that we build we also try to get the right data management for the new model.

MD: What do you expect the ANJ program to achieve in the next few years?

EFVF: Within between 18 months and two-and-a-half years we expect to have a common base with all the data organized and all the new systems for the automation platforms. We are building platforms on top of the technology to provide capabilities for the processes. By building block by block, we expect to have a common set [of platforms] to scale the amount of use cases and transform processes.

MD: Has the ANJ program required a lot more investment?

EFVF: We are not doing additional investment on top of what we did in the past. We have a framework, and we know the kind of tools we need and the kind of process that we expect to have. We invest the budget we have to fill any gaps.

MD: Which domains and processes are your prioritizing?

EFVF: We are addressing all network domains and processes. But we are putting a lot of effort into the observability layer and everything related to assurance, performance and fault management processes. We open thousands of tickets daily. This is our life; we have problems in the network and we must be ready to react. This is something that we manage all day. So, there are a lot of use cases for faster and more accurate root cause analysis to dispatch the right person to the right place. Other processes where we are focusing on, and identifying good results, are in planning, development, and network optimization. From the point of view of new techniques, we are investigating and evaluating results of applying digital twins or copilots.

MD: When do you expect to achieve Level 4 automation and where are you now?

EFVF: We are following the TM Forum’s autonomous networks framework and have implemented it in all Telefónica geographies.

Our objective is to have some processes or some tasks in the Level 4 in 2025 and much more in 2026. But even if you have processes for transport or assurance processes for RAN in the higher levels, the compound KPI for the whole company will be lower.

Our thinking on measuring levels has evolved a lot and it is something that we are talking about with TM Forum and with the other telcos. It is important to align and to understand what it means to be at the Level 4 [and] to have objectives. But from a business perspective it's important to understand which processes must be at Level 4 or higher and focus on that.

TM Forum is helping a lot on this question. If you talk with other telcos, they agree that Level 4 is something that we aim to reach for specific processes. We are measuring the whole processes, and have objectives to evolve in general, but in the end, choices must be business oriented.

MD: How has AI changed your approach to AN?

EFVF: Automation is something that Telefónica has always done. In the last 20 years, we developed a lot of robotic process automation (RPA) bots and rules-based automation. Before AI, for example, almost all the tickets in our network operations centers were opened automatically. With AI, we can take a huge step and do things that were not possible 10 or 15 years ago. We can improve the quality of the automation. This is the difference between automation and autonomy. Autonomy means that we are applying AI, and not just generative AI (Gen AI) but traditional machine learning and other techniques, which makes it possible to automate flows. With AI, we’re starting to look for the famous closed loops and to have flows automated. This improves the time it takes for physical tasks and the quality because we have less interference.

MD: Where is Telefónica applying AI?

EFVF: One of the areas where we are applying AI is full lifecycle automation. We have a lot of focus on operations. It’s where we started with automation in the past and it’s where we have the most manual activities and people. We expect to collect more benefits in applying AI here. We are exploring and fostering AI in our lifecycle from planning to deployment to optimization, and not just in fault management.

MD: How does Gen AI fit into your AN vision?

EFVF: We try to use the best techniques [for specific tasks]. We have a lot of machine learning and rules-based AI and other techniques. We think Gen AI can be used a lot, but perhaps not at scale in some of the processes. For example, it doesn't make sense to apply Gen AI to open tickets (and we’ve done some trials), because it’s not efficient to train Gen AI to do something with an extensive and finite list of events. And in some cases, you don’t need AI because the problem is straightforward, like a cut fiber. But when the problem is identified, then Gen AI can be helpful to take insights from four or five data sources to get a root cause analysis.

We are researching in some proof of concepts how to apply Gen AI to help us in the root cause analysis of complex problems where you have a lot of information and the trial results are good. Gen AI can be really useful to aggregate information that we have in product manuals or other sources and then translate this to help you with a specific question.

MD: Are Large Language Models (LLMs) adequate or do telcos need domain-specific LLMs?

EFVF: In my personal opinion, the domain-specific Language Models (DSLMs) will have more opportunities than the LLMs for specific industries. Every time you get an LLM, you have a huge effort to train and apply this general model to your reality. We are doing trials with copilots from different providers and hyperscalers, because some of these copilots must be domain specific. Another option that we cannot forget is the idea of not using LLMs but Small Language Model (SLMs). That is something that is gaining weight for facing some types of issues.

MD: Is Telefónica working on its own domain-specific LLM?

EFVF: We are not building our own LLM. We try to use the tools that are available in the industry. I don’t think it makes sense from the business perspective. It’s like we developed OSSes in the past and that’s something we’re not doing nowadays.

MD: What are some of the challenges and lessons learned so far in this journey?

EFVF: We have a lot of challenges as you can imagine. The first thing you learn is that we must wisely apply the budget we have. The most complex one is the human part. One of the ANJ work streams is new ways of working. The most difficult part is changing the mindset and taking risks, because our [current] processes are very well structured, and things are working. We must explain to people that the things that we are doing put us in the position that we are today, but perhaps they are not the right ones to give the next step and provide better service to our customers in future. Sometimes it's difficult to justify the investment in time and in budget to do something that starts with the same results that you had historically. Those are the main challenges: Be ready to change.