[Session Report] Product Session: Voice AI Powered by Twilio ConversationRelay #SIGNALCONF

Introduction

This is a report on a session I attended at SIGNAL World Tour Tokyo.

The session had two parts: a live demo by Mr. Nakamura of Twilio Japan in the first half, and a Q&A with Mr. Kunimoto of Ururu Co., Ltd. in the second half. This report first follows how the AI actually behaved in the demo, and then introduces how Ururu has grown its phone answering service with a small team.

The session overview is as follows.

• Title: Product Session: Voice AI Realized with Twilio Conversation Relay
• Speakers:
  • Mitsuharu Nakamura (Twilio Japan Sales Engineer)
  • Yasuhiro Kunimoto (Product Manager, Ururu Co., Ltd.)
• Moderator (second half): Kazuki Goto (Twilio Account Executive)
• Date and time: October 8, 2026, 18:10 to 18:30

https://signal.twilio.com/tokyo26

Aim of the Demo

Mr. Nakamura began by saying that troubles like "my child has a fever" or "I locked my keys in the car" invariably happen on holidays or in the middle of the night, and that no matter where you call, you can't get through. AI can work at any time of day. So the aim of the demo was to have an AI answer emergency calls and, even if not perfectly, guide the caller through conversation alone to the point of, for example, stopping a water leak.

Flow of the Live Demo

The demo proceeded by placing a call from a browser and talking to the AI.

1. As soon as the call came in, the AI answered by calling out "Mr. Nakamura" by name. It used customer information registered in advance. Mr. Nakamura said that just being addressed by name gives the caller a bit of reassurance first
2. When he said "water is leaking hard from the faucet," the AI, while keeping the call going, sent a request to the registered LINE account asking him to send a photo of the model number sticker attached to the faucet
3. When he sent the photo, the AI read the model number, replied on LINE with a diagram showing the location of the shutoff valve matching that model number, and instructed him to turn the shutoff valve all the way to the right
4. When he said "the water has stopped," the AI offered visit slots of 9:00 and 13:00 the next day, and arranged the repair for 9:00
5. After the call, a summary of the key points of the conversation arrived on LINE

Mr. Nakamura pointed out that explanations like crawling under the sink to turn a valve are hard to convey by voice alone. Using images together makes it obvious at a glance which valve to close. He also mentioned that the AI could read the tiny model number text from a roughly taken photo. The final step of sending the key points is so that a person calling in a panic can check later even if they missed "who is coming and when." At the bottom of the screen, the results of instant analysis of the conversation content were also displayed.

Mr. Nakamura concluded the demo by saying that while it can't solve every problem, if the water stops and they know a technician will come the next morning, that person can sleep peacefully through the night.

Twilio Features Behind the Demo

The foundation of this demo is Twilio Conversations. Mapping Mr. Nakamura's explanation to the role each feature played in the demo gives the following.

In the demo, ConversationRelay, which is part of Agent Connect, generated the voice that said "Mr. Nakamura." ConversationRelay handles, in a single product, the roles of converting the other party's voice to text, passing it to the LLM of the user's choice, and converting the LLM's reply back into voice.

Ururu's Case: fondesk, Expanded While Staying Small

In the second half, Twilio's Mr. Kazuki Goto served as moderator and spoke with Mr. Kunimoto of Ururu.

fondesk and fondesk IVR

fondesk is a phone answering service in which operators answer calls coming in to a company on its behalf and notify the company of the contents via Slack, LINE, Teams, and so on. fondesk IVR is an automated answering service in which users can build and change the response flow themselves. Pricing starts at 12,000 yen per month for fondesk and 2,980 yen per month for fondesk IVR, and you can start taking calls right away after about 5 minutes of procedures on the web. Features such as obtaining phone numbers, managing operator availability, routing incoming calls, making calls, and sending SMS are all implemented with Twilio, they said.

Reaching the Top Industry Share with a Small Team

fondesk started in 2019 and became the top-share phone answering service in the industry in three years. It is now used by more than 7,500 companies, including fondesk IVR. Even so, the team running it remains about 10 to 20 people in total across development and sales.

According to Mr. Kunimoto, fondesk was not originally something expected within the company; it was a small experiment that began from an idea on the front lines. It was started by four people, including Mr. Kunimoto, alongside their regular work, and they were not in a position to ask the company for "50 million yen for development." He said that what let him turn it into a product, building it little by little on his own, was Twilio, which lets you use phone numbers and lines only as much as needed with no initial investment and scales automatically as the business grows. The configuration of the foundation has barely changed since the time when there were only about 10 contracts. The development team has now grown to five people, but he said that not a single member, including himself, is knowledgeable about the public switched telephone network.

AI Operator in Two Weeks

In spring 2026, a feature called AI Operator was added to fondesk IVR. It is positioned as a third option, in which AI handles the calls, following fondesk, where humans answer, and fondesk IVR, where a machine responds in a fixed flow. The only thing the user configures is the greeting part, "Thank you for calling. This is ○○," and they can start using it without any other preparation.

AI Operator is implemented with ConversationRelay, and Mr. Kunimoto alone brought it to release in about two weeks from starting work. If they were to build the same thing in-house, they would have to prepare the voice-related processing separately and stitch it together, and take on everything up to overall control and infrastructure construction themselves. Mr. Kunimoto highly valued having been able to release it in a very short period compared with that.

So far, fondesk has focused on basic responses with high demand. Going forward, it aims to go a step further and provide responses equipped with specialized knowledge about the business of the companies using it. He said they want to use Conversation Intelligence and Conversation Memory so that it can answer while taking into account the caller's past inquiries.

Impressions

I thought that because the demo narrowed its goal to "getting them to the point where they can sleep through the night" rather than demanding perfection from the AI, the meaning of the flow going back and forth between the phone and LINE was easy to understand. In Ururu's case, a product that started with four people working on it as a side task has expanded its business while staying small, and new features are being released by a single person in a short time, which concretely conveyed the benefit of entrusting the foundation to a platform.