Amdocs Says GenAI Is Ready for Deployment, Outperforms Human Agents

The deployment of Generative AI (GenAI) in the telecom sector is in its early stages. Companies like Amdocs are making significant strides. It has been making waves lately and after the recent Digital Transformation World (DTW) event in Copenhagen. Its CMO, Gil Rosen, revealed that its amAIz platform is ready to serve customers directly. He claimed that it can outperform human customer service agents.

Amdocs is a company renowned for its Business/Operational Support Systems (B/OSS) software. It leverages telecommunications-specific data to adapt large language models (LLMs). It simply trains LLMs using proprietary data and guarantees both accuracy and efficiency.

Partnerships with giants like Nvidia, Microsoft, AWS and Google have positioned the company as a key player in the telecommunications sector. It has become a reliable platform for connecting customer information systems to LLMs.

GenAI, trained on telecom-specific information, can also troubleshoot network issues, reducing the time it takes to identify and resolve problems. By identifying affected customers, especially high-paying ones, GenAI offers a significant advantage. Amdocs is moving from proof-of-concept (PoC) to production environments, with PoCs typically taking two to three months to move to production.

amAIz functionality is pre-integrated in Amdocs’ latest product suite. It can also be added to existing products. GenAI is known to improve the performance and profitability of telecom operators by 30-80%. It also reduces customer call handling time.

GenAI faces a significant challenge: the high computing power required. But it also has environmental benefits. Operators can optimize tasks by analyzing them in manageable chunks. Rosen acknowledged the problem of LLMs “hallucinating” or making up information during the event. He said the accuracy of the initial training was about 80%. Amdocs has significantly minimized these hallucinations. The situation is gradually improving.