SK Telecom embraces ODA to support Global Telco AI Alliance

The Global Telco AI Alliance has made strides in developing Large Language Models (LLMs) for telecom operators since it was formed in July 2023. As the group works to fine-tune and commercialize AI models, TM Forum’s Open Digital Architecture (ODA) is set to play an integral role in helping the initiative to scale.

Founded by Deutsche Telekom, e&, Singtel and SK Telecom, the Alliance members formalized their commitment in February with plans to create a joint venture company and added Softbank to the group.

In an interview with TM Forum's Inform, Will Cho Sang-hyuk, Vice President of Strategic Partnerships at SK Telecom, explained why ODA is important for telco implementations of Generative AI (GenAI) and shared next steps for the organization.

“[Operators] are excited about GenAI and thought using big tech companies’ APIs is sufficient to apply the technology to respective telco use cases, but this is not true … Without seamless integration with the back-end system, you cannot achieve the outcome you want to deliver for customer satisfaction,” he said.

The Alliance is initially focused on creating LLMs to support digital assistants and chatbots to improve customer interactions by making them more conversational. Examples of the GenAI use cases include providing real-time assistance to call center agents and generating post-call summaries.

Customer service use cases require access to Business Support Systems and Operational Support Systems (BSS/OSS), and the Alliance needed a standard way for LLMs to integrate with telco systems without having to handle “scores of APIs,” he explained. This is where ODA comes in.

“The ODA framework is an efficient tool to solve this issue at scale. If we can aggregate collective requirements of telcos wanting to apply GenAI to call center and other business cases, we have a more cohesive platform to collaborate on. TM Forum is a really good platform for that,” he said.

SK Telecom joined TM Forum in February to collaborate with peers and share knowledge around GenAI and telco AI models. The operator’s corporate strategy is to become an AI company and it plans to leverage ODA and Open APIs to support its transformation as well as help to extend ODA to be AI-ready.

The telco also hopes to expand the Global Telco AI Alliance through its partnership with TM Forum. “There should be various ways to encourage the TM Forum members to get involved with this Alliance and open standards is one of them. We also seek tech partners for the Alliance, including BSS and OSS companies,” said Chung Suk-geun, Chief AI Global Officer at SK Telecom in an interview with Nik Willetts, CEO of TM Forum.

Alliance formalizes AI ambition

The Alliance’s five members have committed to creating a JV, with each contributing an equal, undisclosed investment. This will establish a legal entity for joint development work on LLMs and for engaging with LLM providers, communications service providers and other tech partners.

The new company will be able to procure basic LLM components and aggregate anonymized data from telco members as well as their “token consumption”. By combining LLM demand from five telcos, with total customer base of around 1.3 billion, the Alliance JV will be able to negotiate better terms and receive strategic and technical support from LLM developers, such as Anthropic or OpenAI, compared to one operator.

“Many companies sign MOUs and announces partnerships … but without formalised commitment … it doesn't have any follow through. That's why we wanted to create this vehicle to deliver tangible outcomes,” said Cho.

Together, the founding operators will create telco-specific LLMs and LLMOps to be used across their own operations. Initial LLMs will be optimized for Arabic, English, German, Japanese and Korean and there are plans for supporting additional languages.

The Alliance aims to make the telco LLMs easy to use on a platform-as-a-service model.

Learning to speak telco

The LLMs will also be trained to learn the language of the telco industry. That is, fine-tuning the models for telco-specific questions related to customer equipment, contracts or service troubleshooting. With an understanding of telco operations, LLMs can make chatbots more sophisticated and able to handle a wider range of complex customer interactions and provide more helpful responses.

The Alliance has said that LLM training is underway. The first joint development agreement was announced by Deutsche Telekom and SK Telecom in October 2023.

For example, if a customer contacts their service provider to say they are thinking of leaving because their bill is too high, a chatbot powered by an off-the-self LLM might show the customer the easiest way to stop the service. But a telco LLM would be trained to help the customer find a better service plan to meet their needs.

SK Telecom’s Cho said telco LLMs can make call center agents’ jobs easier while improving customer satisfaction. Using telco-specific LLMs, agents can have faster access to information from BSS and OSS systems to support customer queries, such as contract details, rather than manually searching for details.

“If it is telco LLM, tailored for telco use cases, we fine tune the model using telcos’ proprietary data and connect it the BSS/OSS system … We can collapse those lead times … and reduce the cognitive burden on the call center agent’s shoulder… So it’s a win-win for customers and agents,” said Cho.

This can drive cost savings for operators by increasing call center agent retention and streamlining the training of new agents. “It is beneficial for the operator because retaining and training good call center agents takes time and money,” he said.

Generative AI and the ODA will be hot topics at DTW24 – Ignite, which runs from 18-20 June in Copenhagen. View the agenda.