Enabling Flexible Conversational AI by Integrating Large Language Models

Customer service in the banking, insurance, and telecom sectors is increasingly provided through conversational AI. While conversational AI is helpful for many customers, the technology can also lead to frustration if perceived as too rigid or generic. This is a major hurdle for companies like us, who are in the business of making conversational AI interactions as smooth and helpful as possible. Enter the world of large language models (LLM) like GPT-4! These advanced models are known for their incredible ability to provide flexible and personal responses on various topics. In this project, we are looking to harness the power of LLMs to revolutionize conversational AI for customer service. The goal? By integrating LLMs into our platform, we aim to make conversational AI not just smarter but more flexible, trustworthy, and user-friendly. This project isn't just about making conversational AI better. It's about transforming customer service as we know it. This is no small feat, and several challenges are ahead, such as identifying the best ways to use these LLMs, ensuring that they can be trusted, and keeping an eye on their impact on business. To tackle these challenges, we have teamed up with big names like Telenor, SR-Bank, Tryg, and SINTEF. We have a lot of work to do. In 2024, we will pour our resources into identifying key use cases and ways to apply LLMs to support these. This means hard work on user research, concept work, and testing. Building brick by brick while keeping a customer-centered focus. By the end of the project, we will have developed the means needed to scope, build, improve, and manage LLM integrations to improve conversational AI for customer service.

Service providers in banking, insurance, and telecom have largely adopted conversational AI for customer service. However, approximately 40% of end-users find conversational AI frustrating due to inflexibility and generic responses. This poses a challenge for conversational AI providers like boost.ai, as dissatisfaction hinders user experience, erodes trust, and hampers automation. The challenge has profound impact on the value created within the industry. In 2024 alone, the unexplored value in automation potential is projected at 300MNOK. If left unaddressed, untapped value will exponentially grow, reaching 1700MNOK by 2027 and 4300MNOK by 2029. To capture this value, boost.ai sees promise in integrating foundational large language models (LLMs), like GPT-3 and -4, into their platform as these demonstrate impressive flexibility and contextual adaptability. However, realization of the innovation hinges on unresolved research challenges related to identifying LLM use cases, designing trustworthy LLM integrations, assessing LLM model performance, and monitoring LLM business performance. The project aims to solve these challenges and enable human-centered, trustworthy deployment of LLM integration in the boost.ai platform, thereby enhancing conversational AI for customer service and strengthen customer experience. The project will yield two innovations in the form of new products and services: (1) the development of domain-adapted, dependable, and trustworthy LLM integrations with monitoring features, and (2) comprehensive framework and methodologies for scoping use cases, human-centered conversational designs, and assessments of LLMs The project team to solve these challenges includes Boost.ai in collaboration with Telenor, Sparebank 1 SR-Bank, Tryg, and SINTEF. The innovation will establish boost.ai as an internationally leading provider of conversational AI and enable service providers to offer LLM-powered conversational AI for superior customer service.