One AI Voice Now Answers and Dials for Japanese Call Centres

A Back Office Stuck on the Phone

At Houmiya Setsubi, an air-conditioning and water-heater installation company in Atsugi, a city on the western edge of the Tokyo metropolitan area, staff used to spend every weekday from morning through late afternoon answering calls from customers checking on installation dates and follow-up questions. That left little time for paperwork, and the calls did not stop at weekends. Hiring more people to cover the phones was not an option the company could make work.

Rabona AI, a Tokyo-based startup founded in April 2026 out of research at the University of Tokyo, built its newest product around that kind of bind. On 28 September 2026, the company introduced what it describes as a blend-type AI call centre, in which a single AI agent takes both incoming and outgoing calls for the same business, switching between the two as call volumes shift through the day.

Splitting Inbound from Outbound Was the Old Bottleneck

Most AI phone tools on the market today are built for one direction only, answering incoming questions or placing outgoing calls, but rarely both, according to Rabona AI. That split meant a customer who called in still had to be called back by a human once a reminder or follow-up was needed, and systems without caller identification made returning customers repeat information they had already given. Conventional interactive voice response systems could only answer set questions; logging the outcome into a company's own records remained manual work.

Rabona AI says its blend-type system closes that gap by giving one AI agent access to both call directions and to a company's customer records at the same time. The agent draws on kintone, HubSpot or Salesforce data to recognise a caller by phone number, pull up contract or inquiry history, and hold a conversation shaped around that specific customer rather than a fixed script. Once a call ends, it transcribes and summarises the conversation, files the outcome back into the company's systems, and can send a confirmation by SMS, without a person touching the record. The company points to voice technology developed at the University of Tokyo as the basis for speech it describes as natural enough that callers do not immediately notice they are speaking with a machine, including handling interruptions and casual phrasing rather than requiring a caller to press a number on the keypad.

Rabona AI's representative director, Sho Toribe, frames the goal as freeing staff for less mechanical work, saying the AI should “complete not just answers but full transaction processing.”

Rabona AI reports that its systems have handled more than five million calls across more than 50 corporate clients as of September 2026, the base of experience it says informed this latest release. Based on its own review of publicly listed AI phone and call centre vendors in Japan, the company describes the blend-type approach, combining both call directions with full post-call automation in a single agent, as the first of its kind in the country.

Two Clients, Two Different Metrics  

Houmiya Setsubi is one of the companies Rabona AI cites as an early user. The firm handed both its post-installation follow-up calls and its installation-date confirmation calls to the AI agent. Rabona AI says the system now completes roughly 90 per cent of Houmiya's inbound follow-up calls on its own, a figure the company frames as freeing staff from a task that previously consumed hundreds of hours and kept them tied to a desk through the working day. Rabona AI adds that the system reached a workable level of accuracy within about a week of deployment.

ZAP Corporation, a light commercial transport and logistics operator based in Sakai, Osaka, took a different route into the product. Its outbound calls confirm delivery dates and, if a scheduled date does not suit a customer, offer to leave the package in a designated spot or gather alternative dates for a human scheduler to finalise. On the inbound side, a missed delivery triggers an automated callback, and the AI restates delivery information when the customer rings back. Rabona AI says the change cut, by roughly 80 per cent, the volume of delivery-scheduling calls ZAP staff had to handle themselves, freeing drivers to focus on the road rather than the phone, and the company has since extended the same setup from its original Osaka operation to a second site in Saitama.

Charging by the Outcome, Not the Minute

Rabona AI has paired the launch with what it calls outcome-based pricing for select clients, an alternative to billing by user seat or call volume. Under this model, a client pays for specific results the AI agent completes, such as a confirmed delivery date, an accepted package placement, an inquiry resolved without human handoff, or an appointment booked for a sales or property viewing. The company says the exact criteria and rates are negotiated per client and per task, and that a standard monthly base fee plus usage charges remains available for clients who prefer it.

The shift echoes a broader argument Rabona AI is making with this release: that as AI call handling moves from simply answering to completing an entire transaction, the more relevant measure of value is what got done, not how long the call lasted or how many seats a company bought.

A Claim Worth Watching, Not Yet Independently Tested

Japan's tightening labour market has pushed a wave of AI phone and call centre vendors to market over the past two years, most pitching some version of the same promise: fewer people needed to keep the phones answered. What sets Rabona AI's pitch apart is the claim that one agent can competently hold both ends of that conversation at once, drawing on a shared memory of the customer rather than treating inbound and outbound as separate products bolted together.

That claim, along with the specific performance figures attached to it, currently rests on Rabona AI's own disclosure and its own two case studies rather than on independent measurement, and this account has not yet been corroborated against other reporting. Strong results from two early clients are a promising signal for a company not yet six months old, but it remains a small sample, and the “first in the country” framing is Rabona AI's own reading of a fast-moving, crowded market rather than a verified industry ranking. Whether the blend-type approach holds up as Rabona AI takes on larger clients with messier call volumes, and whether outcome-based pricing stays workable once success criteria have to be negotiated at scale, are the two questions likely to decide how far this product travels beyond its first two case studies.