{"id":10100,"date":"2025-02-04T10:50:12","date_gmt":"2025-02-04T10:50:12","guid":{"rendered":"https:\/\/www.gsma.com\/get-involved\/gsma-foundry\/?post_type=gsma_theme_resources&p=10100"},"modified":"2025-09-19T11:32:42","modified_gmt":"2025-09-19T11:32:42","slug":"kddi-utilising-generative-ai-for-chatbot-inquiries","status":"publish","type":"gsma_theme_resources","link":"https:\/\/www.gsma.com\/get-involved\/gsma-foundry\/gsma_resources\/kddi-utilising-generative-ai-for-chatbot-inquiries\/","title":{"rendered":"KDDI: Utilising generative AI for chatbot inquiries"},"content":{"rendered":"\n

Challenge<\/strong>: At the KDDI Customer Centre, KDDI provides a chatbot inquiry service for customers to make inquiries anytime and anywhere. The chatbot recognises the intent behind the customers\u2019 inputs and selects an appropriate response from some options prepared. However, with existing chatbots, there are instances where the customer’s intent may not be accurately recognised if the input is lengthy, or the information is not sufficient. Currently, about 30% of inquiries fall under this category.<\/p>\n\n\n\n

If the chatbot cannot resolve the issue, a chat advisor takes over the interaction. However, this requires the chat advisor to read and comprehend the previous exchanges between the customer and the chatbot, which sometimes takes time.<\/p>\n\n\n\n

Solution<\/strong>: The followings are developed for customers using the LINE (https:\/\/linecorp.com\/en\/<\/a>) “au Support” account:<\/p>\n\n\n\n

    \n
  1. Utilisation of the chatbot: <\/strong>For lengthy inquiries, generative AI summaries the customers’ inputs, organising them into a format that makes them easier for the chatbot to accurately recognise the intent of the inquiries. If any necessary information is missing, generative AI asks follow-up questions to the customer, guiding the chatbot to identify an appropriate response.<\/li>\n\n\n\n
  2. Utilisation in advisor response support<\/strong>: If the chatbot’s response does not resolve the inquiry, a chat advisor will take over it to assist the customer. Before the chat advisor begins interacting with the customer, generative AI summaries the prior conversation between the customer and the chatbot. This allows the advisor to quickly grasp the essence of the issue and provide an appropriate response. Generative AI is also used in the training of chat advisors. By simulating customer interactions with generative AI, the training becomes closer to actual scenarios, leading to improved skills for advisors.<\/li>\n<\/ol>\n\n\n\n

    Impact:<\/strong> KDDI aims to enhance the resolution rate of the chatbot, reduce time to resolve customer inquiries, and improve customer satisfaction. Preliminary testing has shown instances where KDDI could shorter response times by approximately 5 minutes (about 20% of the typical inquiry time) compared to responses handled by existing chatbots. Additionally, KDDI aims to improve the skills of chat advisors by utilising this technology in their training.<\/p>\n\n\n\n

    KDDI intends to leverage the characteristics of generative AI, which includes human-like conversational abilities, along with high accuracy cultivated through its customer centre’s chatbots, to provide more convenient and comfortable services to a larger number of customers. KDDI will continue to add features and enhance functionality in the future.<\/p>\n\n\n\n

    Topic:<\/strong> Customer experience, and language and content AI (NLP\/LLM\/GenAI)<\/p>\n\n\n\n

    Cost:<\/strong> High to low<\/p>\n\n\n\n

    Location:<\/strong> Japan<\/p>\n\n\n\n

    MNOs:<\/strong> KDDI<\/p>\n\n\n\n

    <\/div>\n \n