Chief Executive Officer (CEO), Deutsche Telekom AG, Deutsche Telekom AG
Tim Hoettges, CEO Deutsche Telecom speaking 26 Feb 2024 MWC
DEUTSCHE TELECOM WORKS WITH OTHERS TO BUILD THEIR OWN INDUSTRY-SPECIFIC LLM. Is AI the customer of ...
About Timotheus Hoettges
In a late February 2024 speech at Mobile World Congress, Timotheus Hoettges, CEO of Deutsche Telekom, discussed the company's approach to artificial intelligence. He stated that "nothing has affected every business model in the world" like AI and described it as "the biggest elephant" for the telecoms industry. Hoettges detailed several internal AI use cases, including a customer service tool that improved first-call resolution by over 50%, an employee service tool, and a system for planning fiber-optic network rollouts. He also noted that the company is training 20,000 employees on AI and advocated for a multi-LLM strategy, saying "don't rely on one llm own go into a multi llm strategy for your company." Hoettges expressed caution about the reliability of AI models, citing hallucination rates of "something around 3%" for OpenAI's models, up to 5% for others, and "sometimes even 27%" for models from other providers. He stated, "don't trust AI," but argued that the industry must "Embrace" it despite these trust issues. Additionally, he described a business opportunity for Deutsche Telekom, noting that the company makes "10 billion of revenues with B2B customers in Europe alone" and that small and medium enterprises are "looking desperately to get a solution for artificial intelligence" where his company can act as a "facilitator."
Transcript (16 segments)
T
Timotheus Hoettges0:01
Good morning everybody. I hope the energy is in the room. I'm not talking about politics today. I'm talking about the future of our industry. And what is the biggest elephant for telcos this day? It's artificial intelligence. Definitely the elephant, the biggest thing which we have to embrace. Nothing ever has affected everybody on this planet, nothing has affected every business model in the world, nothing is affecting the way how we operate and process in the future than AI. So the question is how can we practically tackle that? And I want to give you a brief overview about how Deutsche Telekom is working on AI at that point in time, in this early stage.
Let me start with a picture which you know was put into the AI of Aleph Alpha, which is a great German AI company. And the question was raised: if this kid would be elected president, what would be its first official act? You see the cat. Now this model was put into an LLM model with 64 GPUs for 37 days, and the answer at the end of the day to this picture was: the cat will make sure that the economy is going to be strong. Okay. Now you extend your LLM model and make it even bigger, and you ask the same question again. And when you put 256 GPUs for 22 days on the system and ask the question, the answer is: it will declare war to the dark. So what is that telling us, guys? It's telling us don't trust AI, don't trust your foundations. But how can we embrace something which we can't trust? And by the way, there's a study made that at OpenAI you have a hallucination rate of something around 3%, on other AI models up to 5%, PaLM and Google sometimes even 27% hallucinations in the systems. So what are you doing? You're not just putting AI into your system and suddenly find out that your business is going in the wrong direction.
Now what you see is there's no way that we can ignore it. And you see the productivity gains through AI based on a study which was made by Accenture by Julie Sweet's team. And what they say is with the same input we can maximize the output. So it's a maximization of up to 37% of our labor productivity which we can gain by embracing AI into our businesses at every level of the organization. And what you can see here is nothing else than the simple organization of a telco: from the network, from the network layers, from the use case and applications. This is the value chain of a telco. And these are the cases where we embrace AI at Deutsche Telekom. At that point in time we stopped counting after 400 cases within our organization where we are already using AI. So I cannot go into every element of it, but what we are striving for is not about cost savings alone. Yes, productivity is one of the elements, but I can tell you: being more energy efficient, ensuring higher quality, increase network stability, build autonomous, self-healing, automated networks, predictive maintenance, anomaly detection, increase customer experience, individualized offerings to the customers. Endless gains in the way how a telco can better serve its customers via AI.
Let me go and share the way how you can embrace it. And by the way, I would split the world: if you're sitting in front of that issue and saying how can I bring it into my company, you can be a taker, you can be a shaper, you can be a maker, and for some telcos a facilitator, especially when it comes to B2B. A taker is somebody just using standard solutions from the open AI world. Microsoft Copilot is the taker model. Your value as a company is very small, you just take it as a tool. Being a shaper, you already fine-tune the model, you are trying to influence a better reasoning of the LLM. You use the foundation of the big companies but you train it yourself. This morning we signed a deal at Deutsche Telekom with five other operators to build our own LLM model for telco-specific services. You can be a maker. I don't see any telco who's really able to build its own foundation, but maybe this is coming over time. And then there is for telcos a big, big space. Telcos are making in Europe alone 10 billion of revenues with B2B customers. We are serving the middle segment, the small and medium enterprises. And these companies are looking desperately to get a solution for artificial intelligence, solutions for their specific business. And here we can become facilitator with the skills we have both on the AI but as well in cloud and the connectivity element, which will make a business for us.
We have built an ecosystem which we call the AI Competence Center at Deutsche Telekom. So we have built a team of around 500 experts of AI experts which are sitting in a kind of virtual team, a competence center developing AI products, cooperating, leading for security and support. So what are we doing? I can tell you, if you go into your legal department, they have no clue about AI, no clue about it. I'm glad that these guys are finally using a computer. So therefore if you say use an AI tool to make your work much better, they would say I have no clue. So we have an AI Competence Center. We send the expert out, they select the partners. Then you have everybody in the organization. By the way, everybody wants to do AI, everybody wants to have his own license, everybody is trying to book something. I can tell you, highly inefficient. I want to prompt my data, so I put all my customer data into that. Whoa, holy, this should not happen. You should protect your data, you cannot share that in every foundation model of the world. So who is defining the rules? Who is defining that this is all managed from one central piece which is the AI Competence Center? Buying licenses, organizing own developments, and giving it experts into the respective teams throughout the entire organization.
Let me give you some examples. Look, this is how AI is working normally. Germans are tending to go the left way, but if this doesn't make sense, this is the way we take. And this is what AI does: AI is finding a smarter way of approaching a situation much easier. In customer service we have two elements. On the one side the classical IVR which we will substitute very soon for all our clients, and we did it already for the chatbots which are doing a lot of conversation with our clients. In the past we had no LLM model, we had a designed dialogue, a kind of static conversation. In the new world everything is prompted in LLM. We have a personalized dialogue, a fluid conversation is going on. The answers are with much better quality and the first call resolution has increased by more than 50% in the way how we are interacting with our customers.
Second example: we have a lot of requests from our people in the organization when it comes to my vacations, when it comes to my travel expense, when it comes about understanding my pension scheme. So all of this is something where I need a personal interaction. This is now built in a kind of AI-based employee service. So one tool which is a compass to everything employees need in the organization.
The next example: AI for the FTTH, for the fiber build-out and planning. And what you can see here in the back is that we have built a car and this car is full of learning tools. The car is driving throughout the build-out area and the car is recognizing what kind of street it is, what kind of houses are close to the street, and by the way it identifies how deep the roots of the trees are going under the trees, are they shallow roots or are they deep roots. And from the system automatically designs where we build our ducts and where we build the fiber in the respective roll-out. A big issue: we are improving the productivity of the FTTH build-out by 75% via this tool.
Another tool: AI which we use for FTTH build-out as well. We have a lot of external technicians building the basis for the construction and the build-out in our German and European entities. These guys need a chatbot and they have an online chatbot which is answering all the questions which are related to the connectivity, connected lines, call it the switch houses, whatever kind of functions, even the malfunctions are identified. 1.5 million cost savings just in the first 6 months since we have deployed this tool in the build-out for our technicians on the site.
Another one: the run management, a big topic. And by the way, most of the day our network is not utilized. The network is on energy saving is a big topic for us, especially in Europe after we have faced this explosion of energy cost throughout the last two years. With this tool we are able to switch off sites, tiers, elements of our antennas during the day or the evening when there's no capacity needed. We had already energy savings of 16 gigawatt hours in Germany, which is equivalent to 10,000 households covered by electricity throughout a year. 50 million savings just in the first year of an AI automated steering of our mobile sites in the German footprint.
Another one is coding. Here we have a lot of errors. We use a lot of prompting here. We have GitHub Copilot and other things. And for us it's very important to transfer old legacy programming languages into modern ones. There's a language you call it Fame, maybe some of you know it. It's very hard to find programmers for Fame, but a lot of customers are still using it. We transform the code into Python and therefore we have more developers which can then work on improvements or changes of AI programs in the future.
There's another one: AI at Deutsche Telekom which is on security. Malware, and 15,000 customers were prevented by automated code leakage prevention. And there is another field which we are showing here on this slide, which is the end of the app time. This is our phone, this is our T-Phone which we are developing in the US and in Europe. And by the way, together with an outstanding company Brain AI from the US, we will make this phone entirely app-free. Who the hell needs apps? I don't want to have an app ID management. The app is doing something in the back of my phone with my data. It is always password required. I don't need the functions very often. Why can't I talk to my phone and say by the way I want to buy something for my daughter, I want to do something on vacation, and automatically the AI is looking for the service via my apps and giving me the result immediately. No intermediate anymore. I can tell you in 5, 10 years from now, nobody from us will use apps anymore. We will use just the interface of speech or maybe an easy way of asking the system and automatically the connection to the functionalities of the apps will take place.
I talked about the global alliance where we built our own LLM. And this is taking place with it is Alot together with SoftBank, together with SK who is the main driver of it, Singtel, and Deutsche Telekom. Because we do not want to become dependent on the hyperscalers alone. We want to build our own system. We want to refine it in the way how we want to have our reasoning, then a reasoning coming from the outside. I think we understand our world better and therefore we have to train the model ourselves. Maybe at a later stage, happy to invite other telcos to this alliance as well.
So let me finalize on this slide about what can you do if you are a telco, what can you do if you are a corporate with a big organization, what can you do with all the legacy and this new stuff coming to manage it. How can you embrace AI? First thing: it's a CEO topic. This company, your company will not drive it from the base. It will not be managed by the IT people or some experts in this organization. You have to be at the helm of it. You have to define the strategy. You have to train. You have to show this organization how important it is. And you have to take away the anxiety in the organization about losing their jobs. In others, you have to decide what are you doing. The more you want to differentiate with AI, the more you have to become a maker and invest into it. The more you want to do what everybody does, just embrace the technology for productivity gain, then you can be a taker. But you have to define for each of the areas of your value chain how you embrace AI and what kind of tools you're taking.
Make clear to the organization the ethical standards on how you prioritize your data and how you are managing data. Make clear how you're protecting privacy in the system because people are concerned about what's happening with all this data. Empower talents. We have trained already more than 75,000 people in the organization on AI. And by the way, not only IT people. Everybody should have a glance, a feeling about AI. Otherwise they will be against it. It will be like an antivirus against AI. So teach the people about the benefits of AI throughout the entire organization and empower new talents to your organization, especially in the IT landscape. Data sovereignty: a big thing. Just one sentence: don't rely on one LLM. Go into a multi-LLM strategy for your company. By the way, even synthesizing different LLM models for better results and small LLM models is a big topic on the booths on the fairs here already. And last but not least, measure and govern all these projects on a weekly basis. People have to come to me. We call it drum beats. But we talk now about what is the ambition you have in mind, what do you want to achieve, what is the benefit for the organization, and what do you need from the AI Competence Center that you really can embrace AI and make it a success.
I think AI is here and it came to stay. And I can tell you in the next years we will see hundreds, thousands of new use cases which will make our industry better. And I'm convinced we can serve our customers with AI much better than we can do that today. Thank you very much for your listening. Thanks.