[Session Report] Voice AI and Telephony DX: Achieving Next-Generation Customer Engagement #SIGNALCONF

Introduction

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

Three companies — AI Shift, Mobilus, and IVRy — that sell voice AI (AI that handles responses by voice) as their own products took the stage. The three companies also compete against each other on deals, and they spoke frankly about their genuine outlook and approach to product development. This report introduces the key points each company emphasizes in building their products, followed by a discussion of their outlook on AI adoption and the division of roles between humans and AI.

The session overview is as follows:

• Title: Voice AI and Telephony DX: Realizing Next-Generation Customer Engagement
• Speakers:
  • Tsutomu Tajima (Executive Officer, AI Shift Inc.)
  • Tomohiro Ishii (President and CEO, Mobilus Inc.)
  • Ryoga Okunishi (CEO and Representative Director, IVRy Inc.)
• Moderator: Kei Nagai (Twilio Account Executive)
• Date: October 8, 2026, 17:50–18:10

https://signal.twilio.com/tokyo26

Key Points Emphasized in Product Development

Toward the end of the session, moderator Nagai shared his personal experience as a user — noting moments where something felt "a little off" — and asked each company what they focus on to dispel the anxieties of companies considering adoption. The answers from the three companies each reflected a different perspective.

Resolving the Matter on the Spot

Tajima noted that while evaluation tends to focus on how natural the conversation sounds, what AI Shift truly values is the completion of the task from start to finish. For a reservation call, for example, AI is built to handle making, canceling, and rescheduling reservations, with integration into back-end business systems. While natural pacing and accurate speech recognition are necessary, he believes that what matters most is building the system to fully resolve the customer's request on the spot. He acknowledged that this often requires custom development for each project, but said the company has committed to tackling exactly that.

Keeping the Cost per Interaction Down

Ishii pointed out that beyond the costs of building and operating the system, the cost per interaction compounds over time. While there is a temptation to use the highest-performing models for better accuracy, that approach is not financially viable. He explained that their system is designed to mix and match within a single interaction: using small, inexpensive models where sufficient, switching to larger models when needed, and not using AI at all in certain situations.

Accessible to Anyone, and Turning Calls into Data

Okunishi explained that since IVRy is an AI phone service available from a few thousand yen per month, the company is committed to making it easy for anyone to get started immediately. His other focus is on calls coming into company main lines, departmental numbers, and store numbers. Unlike call centers, these calls have historically not even been recorded. He described how IVRy builds systems that, as a byproduct of streamlining responses, also accumulate call data and use AI to analyze previously invisible customer feedback for business use.

Nagai added that at Twilio, voice and interaction history are referred to as context data, and that Twilio works to make this data easy to manage and reuse safely.

Outlook on AI Adoption

Early in the discussion, Ishii asked the other two panelists for their estimates on the percentage of call volume that would be replaced by AI, and how many years it would take to get there.

Ishii himself said he publicly states "80% within the next four years." Okunishi, taking into account the wide range of clients from small ramen shops to major corporations, shared his sense that including small and medium-sized businesses, the figure would be around 50 to 70 percent over ten years.

Tajima said he thinks about it in two stages. He expects it will take around ten years for conversational performance to improve enough to delegate the majority of calls. On the other hand, he noted that even with current technology, about 40 percent of interactions could already be handled by AI — and the reason that potential isn't being realized is not the technology itself, but rather workflows and integration with existing systems. He said he wants to achieve that 40 percent within three years.

Earlier in the session, Tajima also touched on why AI adoption has been slow. Much of the AI in use is developed in the West, and Japanese companies tend to look to overseas cases as models for adoption. However, when Tajima attended SIGNAL in San Francisco in May, he was struck again by the fundamental differences in context between those overseas models and Japan. Japan is the most rapidly aging country among major nations, and Japanese is a language used only in Japan. If the situation where overseas AI leads continues for the foreseeable future, Tajima said, closing that gap is precisely where Japanese companies can create value.

Division of Roles Between Humans and AI

Nagai asked for opinions on which model suits the Japanese market better: one where everything is handled entirely by AI, or one where AI prepares the groundwork and hands off to a human.

Okunishi used the example of email continuing to be used even after Slack and LINE became widespread, noting that it takes a long time for people to change their behavior and communication methods. While conversations with a clear purpose are easy to delegate to AI, there are also conversations where the person doesn't yet know what they want to ask. His view is that even as the range of what AI can handle expands, the diversity of interactions itself will remain.

Tajima's perspective differs slightly from the other two. AI Shift has little involvement with products that assist human operators. Tajima noted that typing is, in a sense, a specialized skill. He believes it is easier for people to resolve their needs entirely through voice — a mode of communication anyone can use — and sees a state where AI alone can handle interactions as the ideal. He added that in reality, a combination of humans and AI will continue for some time.

Ishii noted, half-jokingly, that from his company's business perspective, he hoped there would remain some room for human involvement.

Why Outsource the Communications Infrastructure

What all three companies have in common is that they entrust the telephone line infrastructure to Twilio. Ishii described the reason as "leave it to the experts" (literally, "rice cakes to a rice cake shop"), and Okunishi used the same expression.

Mobilus started as a company built around chatbots and live chat. According to Ishii, they did try handling telephony in-house, but at the time they could not find engineers with expertise in Asterisk. What surprised them when they ventured into voice was how much more demanding the required uptime was compared to chat. Customers demanded "no downtime, no matter what," and for large enterprises, the system needed to handle hundreds of lines. Ishii said that what Mobilus wants to invest in is the management interface and UI for AI agents, and that is where they want to differentiate.

Okunishi talked about how IVRy expanded from serving small businesses to large enterprises. When Ishii asked what had changed as they moved into large enterprise accounts, Okunishi answered that in pursuing ease of use and satisfying quality for small businesses, they had ended up with a product that also satisfied large enterprises. He said the growing interest in AI among large enterprise call centers, which led to more inquiries than expected, also helped. He noted that being able to focus development entirely on usability and AI refinement — even without deep telephony expertise — supported their growth, and introduced their current cumulative account count of approximately 70,000.

Impressions

It was informative to see three competing companies refining their products from different angles — task completion, cost, and leveraging call data — which made the concrete challenges of building voice AI that works in the real world clearly visible. I was also struck by how all three companies shared the same approach of outsourcing the telephony infrastructure and concentrating on their own strengths.