Gartner's 2023 Emerging Technology Maturity Curve shows that generative AI is currently in a period of inflation of expectations, which is expected to yield significant benefits in two to five years, leading to new opportunities for innovation. Generative AI enables operators to efficiently process the rich data generated by their networks every day, identify patterns, and generate innovative solutions that improve operational efficiency and expand revenue streams. As a result, operators see generative AI as a means to create incremental value, and have been tested in multiple businesses and domains, exploring the potential of generative AI in the testing process. In the face of this important trend, what kind of generative AI development strategies have international operators adopted, what applications have they practiced, and what challenges do they face?
01
Operators' generative AI development strategies
Generative AI is currently in its early stages of development, and the biggest difference between operators' layout in this field is whether they choose to develop a basic model from scratch (radical) or use an off-the-shelf model (conservative). The majority of these operators use off-the-shelf models, which use proprietary internal data to train models that are more relevant to their customer service, operations, and network needs.
Radical develops large language models customized for operators
Operators such as South Korea's SKT and Deutsche Telekom are more aggressive in their generative AI layout, using alliances to develop large language models (LLMs) specific to the telecom industry from scratch.
At the MWC2024, SKT, Deutsche Telekom, e&Group, Singtel and SoftBank announced the formation of the Global Telecommunications Artificial Intelligence Alliance (GTAA) to deploy work on innovative AI applications and develop large language models specifically for the needs of telcos. Generative AI relies on a large language model that needs to be trained in a language-specific data corpus. On the one hand, due to language barriers, the application of generative AI in the Asia-Pacific and Middle East regions is lagging behind. Most of the leading large language models have limited ability to process non-English languages, especially Asian languages. For example, ChatGPT is not as effective in Thai as it is in English, and operators in some countries or regions may need to train some language models on their own to achieve the same effect. On the other hand, the training data of general large models rarely involves information such as packages, contracts, and special hardware (such as routers) that are unique to the communications industry, so operators need to fine-tune the model with their own data for specific problems in the telecom industry and optimize it for specific languages, which reduces the deployment efficiency of generative AI. Tailoring large language models for the telecom industry will enable operators to deploy high-quality generative AI more quickly and efficiently, accelerating the AI transformation of telecom services and services.
Conservatives use off-the-shelf models to optimize operations
Compared with radicals such as SKT, most operators are relatively conservative in their approach to generative AI, focusing on using off-the-shelf models to accelerate the adoption of generative AI, optimize their networks and operations, improve user experience, and empower employees. A small number of operators with strong data processing capabilities are exploring the use of generative AI to create new revenue streams.
Enhance services
Generative AI-based user service chatbots are the most important and mature application. The app's ability to understand user intent and quickly understand, summarize, and respond to specific user questions reduces call volume while providing personalized service to users, freeing up agents to work on more complex issues. IBM research shows that companies save $1 million annually by reducing the average call handling time in a call center. Chatbots can reduce the average query processing cost from $5~$12 to $1. In the long run, generative AI can provide users with services such as plan changes, upgrades, or contract renewals based on the information provided, without human intervention.
For example, Veronika, a virtual assistant launched by Telkomsel, an Indonesian telecommunications company, uses a range of programming languages rooted in natural language processing (NLP) and machine learning, which allows it to provide more natural and intuitive interactions when serving users. The benefits of machine learning technology allow Veronika to gain a deeper understanding of the user's service needs, allowing it to provide solutions tailored to each user's requirements and usage in a more specific way. For example, we can recommend the right products and related information based on the user's needs to provide a more personalized service.
Empower your employees
Similar to generative AI for user services, generative AI that empowers employees also takes the form of "assistants" to help employees complete tasks more efficiently and better.
Vodafone uses generative AI technology to write code, and in a trial with about 250 developers, productivity has increased by 30%~45%. AT&T has launched "Ask AT&T", a ChatGPT-based generative AI platform that allows employees to focus on more complex, higher-value tasks. Ask AT&T is able to use generative AI to convert legacy code into modern code, or empower employees to complete common HR tasks, such as changing withholding taxes, adding dependents to an insurance plan, or requesting a computer for a new hire. Employees can ask questions or give commands to Ask AT&T, and the platform will route the questions or commands to the appropriate people on their behalf. Perhaps in the near future, operators will also be able to leverage data from user interactions to provide coaching and performance improvement coaching to agents, or use AI tools for real-time sales coaching to continue unleashing the power of generative AI.
Network optimization
The network is a core asset and a major cost for operators. By applying generative AI technology, operators can proactively track and study their network and other data parameters to optimize network planning and management, reducing costs and increasing efficiency while providing users with more efficient and reliable services. However, the sensitivity of network data is one of the limiting factors for the development of applications in this area.
Based on data collected from aspects such as network performance, traffic, and user behavior, Three UK leverages Azure Operator Insights to optimize network configurations, policies, and parameters. With the further deployment of these technologies, generative AI can learn from existing network settings and generate new ones, thereby improving network efficiency, reliability, and security, helping operators design and deploy optimal network slicing solutions for different applications and users, or adjust network parameters to meet changing needs. Generative AI can also generate and perform appropriate actions based on network status, helping to automate network management tasks such as fault detection, diagnosis, and troubleshooting. In addition, generative AI can also use natural language processing to enable more human-friendly interactions between operators and network systems, such as using voice to control network functions and receive natural language explanations of network status and its problematic nature.
Of course, generative AI can also play a key role in all phases of the network lifecycle. For example, engineers often rely on manuals and records when installing network components, and generative AI can learn from this data and provide interactive guidance and prompts to simplify installation tasks and speed up installation. Generative AI can also train the underlying model based on network topology and configuration data to recommend the configuration of network elements. In addition, generative AI can also recommend troubleshooting actions and procedures to engineers when the network fails.
Create new revenue streams
A small number of operators with strong data processing capabilities will leverage generative AI to create new revenue streams. For example, Orange announced that it will use generative AI to create new products in the sports market, while trying to explore ways to monetize generative AI in the media environment. SKT has developed a consumer-centric app that allows users to interact with it for things like listening to music, sending text messages, making payments, and managing schedules.
In addition to the above uses, operators can also try to use AI-based tools to understand and analyze subscriber data, and use generative AI to improve performance through personalized messaging and precision marketing. For example, use generative AI to create titles, snippets, keywords, and content for online content based on the interests and behaviors of your target audience to make your content more appealing to users. Generative AI can also use data from a variety of sources, such as web analytics, user relationship management platforms, and social media, to segment users based on their attributes, desires, preferences, and behaviors, and then tailor personalized offers accordingly, or recommend appropriate personalized products and services to users based on data such as their previous purchase history, web history, and feedback. For example, operators can leverage generative AI to provide each user with the best package configuration or service recommendations based on their budget, needs, and usage habits. Generative AI can also use data from a variety of sources, such as third-party databases, social media, and more, to identify and reach out to potential users who match their ideal user profile, helping sales teams create opportunities. For example, using generative AI to identify and reach out to potential users who are searching for similar services, have obvious intent, and meet decision-maker criteria to create sales opportunities for operators' business solutions.
02
Problems and challenges
As operators deepen the deployment of generative AI, the problems and challenges they face are becoming apparent.
Data governance issues
Generative AI is only as good as the quality of the data used to train it. Carriers have abundant data, but in most cases, the following problems are present with this data. The first is fragmentation, where data is collected and stored by disparate systems without a unified database accessible to AI systems, the second is unstructured, where data is stored as large amounts of unclassified data without any context or interpretation of its relevance, which is of low value to AI algorithms, and finally, it is incomplete, where operators often lose data, and incomplete data can lead to inconsistent learning or errors in AI systems. For example, Verizon has a lot of data, but it exists in 29,000 different data sources, which is fragmented in many ways, and there is no unified way to classify it. As a result, Verizon is consolidating all of its data into a common platform and translating it into a common governance and classification structure.
As a result, operators' data processing capabilities will have a significant impact on their adoption of generative AI. Operators with strong data processing capabilities that are leaders in generative AI applications share the following organizational characteristics: dedicated AI centers of excellence, universal use of advanced data analytics (using advanced analytics across the organization, not just in certain business areas), and modern data infrastructure (e.g., cloud computing facilities).
Security compliance issues
Leveraging generative AI requires large amounts of proprietary data, and operator data involves sensitive user information and strict data protection regulations, ensuring data security compliance, including intellectual property, is a major challenge.
Lack of implementation and management of generative AI
In-house expertise
Hiring and retaining talent is challenging for operators to recruit and retain talent, often with limited in-house AI talent, and it can be difficult to find professionals who are proficient in both data science, machine learning, and AI algorithms. While AI training for existing employees can be time-consuming and costly, setting up an in-house AI team is time-consuming and potentially ineffective. Before diving into and developing critical applications, you need to make sure your team has the right skills and is familiar with the tools they need to use. This is extremely important for operators to use generative AI, as the tools and skills required to use them are changing rapidly.
Legacy system integration
Operators still face significant hurdles when dealing with outdated technology systems, i.e., legacy systems. The burden of technical debt (the hidden cost of not addressing future-impacting problems) and complex integrations have led to a slower pace of modernization for communications operators than other industries. Gartner predicts that technology debt will consume more than 40% of operators' current IT budgets by 2025. Difficulty integrating new generative AI tools into legacy systems is one of the most common reasons for AI integration failures. At the same time, integrating AI solutions with legacy systems, applications, and networks can lead to compatibility issues. Careful planning of the scenario and ensuring compatibility with legacy systems is critical to the successful integration of AI.
Communication and coordination issues within the organization
In organizations, operators need to engage multiple stakeholders when adopting generative AI, with business and IT stakeholders being particularly critical to evaluating and implementing generative AI. On the one hand, leaders of business units (e.g., customer services, marketing, and networking, etc.) need to be involved in the generative AI ideation process and work with other stakeholders to conduct proof-of-concept. On the other hand, the IT department, while playing a smaller role in the assessment, plays a key role in generative AI applications. These two key stakeholders work together to decide if and how generative AI will be adopted in the organization, so communicating and coordinating the collaboration between the two parties is a challenge.
03
Future outlook
Generative AI enables operators to process large amounts of data, identify patterns, and generate novel solutions that have the potential to change traditional practices and drive industry-wide innovation. A study by AWS predicts that operators' investment in generative AI will continue to increase. Operators who have already adopted generative AI are spending less than 1% of their total technology spending on these functions, but 45% of those surveyed said that spending on generative AI is expected to surge to 2%~6% of total technology spending in the next two years. This suggests that spending on generative AI could grow as much as six times in the near future as operators scale up their applications and explore new application areas. In order for the industry to thrive, operators must face many challenges head-on and take a long-term view, integrate generative AI into their operational structures, and unleash the true power of generative AI.
*This article was published in Communication World Issue 940 on March 25, 2024 Issue 6 Original title: "Generative AI Transforms the Communications Industry, How Do International Operators Layout?"
END
Author: Hu Mengshan, China Unicom Research Institute
Editor-in-charge/layout: Wang Yurong
Reviewer: Wang Tao, Mei Yaxin
Producer: Liu Qicheng
【Communication World New Media Matrix】
央视频 | 微软MSN | 视频号 | 微博 | 今日头条 | 百家号 | 网易号 | 搜狐 | 腾讯新闻
新浪看点 | 雪球号 | 抖音 | 快手 | 爱奇艺 | 知乎 | 哔哩哔哩 | 咪咕视频 | CSDN | 36氪
【New Media Team】Producer|Liu Qicheng Proofreader|Wang Tao, Mei Yaxin Editor|Wang Yurong, Gai Beibei Video Editor|Huang Yangyang, Lu Ruixu Operation|Lin Song
Dots to share
Point Collection