Transforming telecom with GenAI: The journey from chatbots to network agents

As 5G and cloud-native architectures expand, network complexity increases. Generative Artificial Intelligence (GenAI) has the potential to transform the telecommunications industry by processing and analyzing large volumes of data to generate valuable insights, offering opportunities for innovation and increased efficiency. This technology opens a wide range of use cases for telecom service providers. Initially, these early adopters  focused on using GenAI in marketing and call center applications. However, they are now rapidly expanding its use to network operations.

GenAI-powered chatbots capable of delivering instant, personalized responses to customer inquiries have become a reality in the telecom sector. While most of the service providers we spoke with have already deployed AI/ML in the telecom domain, they are still experimenting with network-near GenAI-related use cases with only a few public announcements so far . 

In this blog post, we explore how the telecom industry can leverage GenAI technologies to transition from simple chatbots to autonomous intelligent agents that enhance customer experiences and streamline operations. We introduce a framework for network domain GenAI applications for telecom service providers and categorize them into three groups:

1. Knowledge assistants
2. Product assistants
3. Network-near use cases

Knowledge assistants expedite and simplify access to information for service providers. For example, vendor product documentation usually provided to customers as a web-based database can instead be queried through a GenAI-enabled assistant (utilizing LLMs and retrieval-augmented generation¹ or RAG) to quickly provide instructions. The objective is to increase employee productivity by saving time for a task already performed today. The integration complexity, or the effort to integrate the capability into existing systems / workflows, is low, as the assisted knowledge is detached from a product /workflow. 

Product assistants are similar to knowledge assistants but they are introduced as an adjacent/embedded part of a software. The tool enhances current workflows without disrupting current ways of working. It improves efficiency and enhance product performance by offering features such as, summarizing observations from network performance data from analytics software. A few examples that Ericsson is developing in this area are:

• Ericsson Expert Analytics (EEA) chatbot, which allows users to query network data using natural language, reducing the threshold of understanding and extracting insights from complex network data.
• Product configuration assistant; it automates the conversion of business requirements to product offerings and implements correct configuration in revenue management and product catalog systems.

Network-near use cases are the most complex category and key to the journey towards autonomous networks. Service providers see GenAI as a critical enabler to advance telecom automation and enable future autonomous networks, with 48 percent of communication service providers (CSPs) expecting to reach Level 4 automation² by 2028³ .

Unlike knowledge and product assistants, network-near use cases would, in most cases, require models that are contextualized with telecom-specific data. This data includes methods of procedures, product configurations, network logs, and industry standardization documents. By using RAG or a fine-tuned model, these use cases transform how certain telecom workflows are performed, such as  in fault isolation, capacity planning, or radio resource management. These altered workflows often combine GenAI use cases and traditional machine learning models to deliver the desired outcome. Although experiments are currently underway, we believe service providers will deploy their first live network-near use cases by 2025 (as indicated by customer and partner interviews, as well as internal development). 

Ericsson is developing network-near-related use cases such as:

• Network Operation Center (NOC) digital engineer; it is designed with an agentic workflow to automate issue detection, fault correlation & resolution. 
• SMO conflict management; it avoids executing conflicting rApp inputs, like policies and configuration data, towards RAN that may negatively impact network performance and compromise security.

GenAI is an area of immense innovation and constant evolution. Network agents, built on agent architecture, integrate with knowledge assistants, product assistants, and network-near use cases and autonomously analyze, reason, and act to solve specific problems. They will be a vital part of achieving the industry vision for an autonomous network.

GenAI is still in its early stages and while it can be utilized to improve efficiency and quality with the first two categories, several challengesmust be addressed before it is introduced in the network domain through network agents. 

• Accuracy: Many telecom service providers we spoke with require an accuracy of +95% for network-near use cases. Hallucinations need to be minimized, with output explainability from a security and compliance perspective. 
• Compute cost: Training, fine-tuning and maintaining GenAI models requires significant compute resources. This can lead to a challenging return on investment for CSPs, especially smaller regional players. 
• Data privacy and compliance: GenAI requires huge amounts of data. Service provider handle sensitive customer data, which raises concerns around privacy and regulatory compliance. 
• Energy consumption and sustainability: Training and running GenAI models requires substantial energy resources, raising concerns about sustainability and future energy costs as service providers aim to achieve their financial and sustainability goals.

Our proposed framework - knowledge assistants, product assistants, and network-near use cases – is intended to help service providers navigate the telecom GenAI landscape and the benefits from the different categories. Insights from implementing knowledge and product assistants will support service providers as they adopt network-near use cases, transforming network operations and lowering the overall total cost of ownership (TCO). 

GenAI can also open new revenue streams by powering AI-driven consumer and B2B services, which could potentially drive network traffic and set new requirements on the network. A recent Technology Report from Ericsson⁴ examines the impact of GenAI capabilities such as AI-agent and semantic compression on cellular networks. As a leader in telecom networks, Ericsson is at the forefront of developing GenAI to support service providers in their journey towards autonomous networks.

Read more:

• Understanding the impact of GenAI
• Four ways generative AI is set to transform the telecom industry
• Telecom AI

1. RAG combines a generative model with a retrieval system to enhance responses by incorporating relevant, external information in real-time 
2. TMForum
3. Omdia, Telco Network Automation Survey Report 2024
4. Ericsson, Technology Report, Impact of GenAI on Network Traffic (2024)