Easily Implement Complex Telephone Services with Twilio: A Voicebot Tool that Realizes the "Democratization of AI"
Time to read:
Developing a unique dialogue engine that enables full support
AI Shift Inc. is an "AI dialogue company" established in 2019 as a subsidiary of CyberAgent. They develop and operate the AI-powered customer support chatbot service "AI Messenger Chatbot" and the automated telephone response service "AI Messenger Voicebot." While many services leave the operational aspects to the client side, both services provide consistent full support from design to operation, and are highly regarded for their problem-solving capabilities. According to Yuta Tomomatsu, a data scientist at AI Shift Inc, the reason for this success lies in the company's unique dialogue engine development. "A superior dialogue engine is one that’s able to accurately select the optimal route from the prepared routes. Since routes that are not prepared cannot be selected in the first place, it’s important to train it on as many anticipated routes as possible. While it is difficult to differentiate based on the performance of the AI itself, our service focuses on an operational system that efficiently enhances the 'comprehensiveness of responses.' Analyzing the unsuccessful interactions with a large amount of manpower and accumulating know-how from over a hundred companies is a significant advantage. Furthermore, based on this vast amount of data, we have also realized initiatives to automate operational support functions for training the dialogue engine by utilizing AI in the operational system itself. By using this operational function called 'AI Compass,' we can automatically provide support of quality comparable to that achieved through operations requiring extensive manpower."
The emergence of voicebot services that overcome the limitations of chatbots
The company, which had primarily focused on chatbot services, began developing voicebot services about two years ago. Why did they decide to develop voicebot services? Tsutomu Tajima, head of the AI Messenger Voicebot business within the AI Messenger Division, shared his insights. "One of the purposes for companies to implement chatbot services is to eliminate phone-based support for operational efficiency. However, since there are a certain number of people who choose the phone first when selecting a channel, there are areas where we cannot necessarily contribute. We considered ways to shift from phone to chat, but we believed that for a more effortless customer experience, a service that can respond directly on the phone was necessary. We are leveraging the strengths of the dialogue engine developed through our chatbot service to expand into voicebot services. The main purpose is for information gathering such as appointments, with about 70% of the implementation happening in general companies, but inquiries from local governments are also increasing. The main inquiries come from the sales organization for local governments of our parent company, CyberAgent, but it seems that more local governments are starting to automate telephone responses as part of their digital shift. For example, we have implemented the 'AI Messenger Voicebot' for handling inquiries about road regulation information in Fukui Prefecture, and provided technical support for the development of an AI appointment dial dedicated to COVID-19 vaccinations in Tama City, Tokyo. Public institutions receive inquiries from customers of various ages and genders, with the primary channel being telephone. The automated telephone response service is likely to become a much-anticipated tool for those in various local government offices who are extremely busy."
Speedy and Scalable: Smoothly Implementing Telephone Functions with Twilio
In this service, Twilio is used for core functions such as receiving calls, recording, sending SMS, and forwarding numbers. Mr. Tomomatsu spoke once again, sharing the background of implementing Twilio in 'AI Messenger Voicebot'. "Until now, I had almost no experience in development related to telephones. We searched for something that would allow us to develop telephone features based on our knowledge of regular web development and discovered Twilio. After checking the demo, we were able to smoothly create a prototype, and since there were no functional issues, it naturally led to adoption in the production environment as well. While there are other services that transcribe speech recognition results, the ability to respond in real-time via WebSockets is unique to Twilio. The ability to quickly issue phone numbers and create demos makes it easier to propose to clients. Recognizing the caller's voice, converting it to text, passing it to the dialogue module to generate responses, and synthesizing this into voice audio is a series of phases that can be easily integrated into our own services and can respond flexibly, which is also appealing. Looking ahead to future service development, our biggest reasons for choosing Twilio were its excellent functionality and scalability. The development period for basic automated response services via telephone was about six months. Twilio has a wealth of samples, and you can also ask questions in the Slack community. Initially, I was anxious because I couldn’t see whether this service would really be viable, but by using Twilio, I was able to see a clear path to implementation, and development progressed rapidly. The voicebot service using Twilio has a higher number of simultaneous connections than typical services, demonstrating processing capabilities similar to having a call center for hundreds of people. Even in projects with tens of thousands of inquiries within a few hours, we are currently able to operate without any issues. Even after implementation, we receive detailed follow-up from Twilio, such as advice when proposing to new clients. We are receiving substantial support in areas where we are not familiar with telephones, including holding study sessions, which has been extremely helpful."
Developing systems for industry-academia collaboration to make AI more accessible
"We value incorporating an academic perspective into product development, and we also approach medium- to long-term challenges through 'industry-academia collaboration,' not just the issues of the product itself. We are conducting joint research in collaboration with various university laboratories, which allows us, as a brand, to follow the latest research while universities can obtain real customer data, creating significant benefits for both parties. We are currently advancing the development of integrating chatbots and voicebots. In the future, we aim to create a system that shares knowledge gained from responses in both phone and chat channels, allowing both channels to grow. Moreover, both chatbots and voicebots require integration with other systems to complete the service. For example, when making reservations or cancellations, it is necessary to connect to each respective system. By connecting these to a single system that we provide, we envision scenarios that can be completed automatically. We expect this to become an even greater strength of the service in the future. While maintaining a long-term perspective, we believe that continuously addressing immediate challenges is essential to achieving our mission of 'democratization of AI.' We would like to continue expanding the new possibilities of AI to the world."
