Verbex's Real-Time Voice Conversational AI Selected for Full Implementation in Mail-Order Phone Reception at Oshimaya
Verbex has announced the full implementation of its real-time voice conversational AI for mail-order phone reception at food retailer Oshimaya, following a highly successful PoC.
📋 Article Processing Timeline
- 📰 Published: April 22, 2026 at 18:20
- 🔍 Collected: April 23, 2026 at 00:02 (5h 42m after Published)
- 🤖 AI Analyzed: April 23, 2026 at 10:47 (10h 44m after Collected)
Verbex Inc., which develops and provides a voice conversational AI platform, announced that its real-time voice conversational AI has been selected for full implementation for mail-order phone reception at Oshimaya Inc., a company engaged in food mail-order business.
Oshimaya, based in Kumamoto Prefecture, has long operated a mail-order business delivering fruits direct from farmers and Kumamoto specialties like horse sashimi nationwide. They pride themselves on a unique distribution channel connecting primary industry producers with consumers, running a business that demands high-quality customer touchpoints in food mail-order.
This full implementation was decided after a Proof of Concept (PoC) confirmed high results in suppressing lost order opportunities and improving the customer experience during temporary spikes in call volumes, which occur during peak seasons, sales of popular items, and infomercial broadcasts.
■ Background and Challenges
In the phone reception business of mail-order sales, incoming calls can temporarily concentrate around the timing of sales promotions for specific products or advertisements/broadcasts. During these peak times, "overflow calls" that disconnect without being answered easily occur due to line congestion or reaching the upper limit of operator capacity, leading to lost sales opportunities.
Phone orders in food mail-order, in particular, cannot be completed simply by mechanically accepting the order details. Customers are often in a higher age demographic, resulting in significant individual differences in speaking style and conversation tempo, and there is a need to accommodate regional phrasing and dialects. Additionally, because the accuracy of understanding names, addresses, product names, and quantities directly translates to order quality, it is a highly difficult business area that is hard to handle adequately with standard automated voice response systems.
Oshimaya advanced the PoC with Verbex to verify whether voice AI could simultaneously reduce lost order opportunities, expand sales, and alleviate frontline burden in such highly difficult order reception tasks.
■ Results Obtained in the PoC
In the PoC, they tackled the highly difficult challenge of mail-order reception via completely free conversation for merchandise that has many elderly customers and dialect speakers. As a result, it was confirmed that the AI handled approximately 60% of calls without needing a callback.
Furthermore, high success rates were recorded for the accuracy of order information acquisition recorded by the AI, including 98% for ordered items, 95% for order quantities, and 94% for advertising media.
These results confirmed that practical operation contributing to the suppression of lost sales opportunities and operational improvement is feasible on the actual business frontline, going beyond mere technical verification. This high technical accuracy and the continuous improvement and support system through the PoC were highly evaluated, leading to the decision for full implementation.
■ Value Aimed for with Full Implementation
The goal of this full implementation is not merely labor saving or automation. It is to support the experience of "being able to properly order the moment you want to buy," which is the most important thing in mail-order business.
By establishing a system capable of immediately answering more customer calls even when call volumes are concentrated, they aim to suppress lost order opportunities. Additionally, creating an environment where human operators do not have to handle all primary responses makes it easier for them to focus on tasks requiring more judgment and consideration, which is expected to raise the baseline of the overall customer experience.
Furthermore, by refining the order flow based on the premise of division of roles between AI and humans, they aim to contribute to stable operations during peak seasons and continuous sales expansion.
■ Future Initiatives - Development of a Unique AI Model Strong in Mail-Order Reception
Verbex will continue to proceed with the continuous improvement of voice conversational technology optimized for the highly difficult area of mail-order reception, in collaboration with Oshimaya, even after this full implementation.
Specifically, they will accelerate the improvement and implementation of unique technologies in the following areas:
1. Improvement of Japanese Conversation Control Accuracy
Verbex will proceed with improvements to its unique Japanese conversation control so that conversations can continue naturally even if the customer interrupts the conversation or provides an unexpected response. In particular, by improving the accuracy of interruption detection and utterance boundary detection, they will realize more natural responses tailored to each customer's speaking tempo and timing.
2. Improvement of Japanese Speech-to-Text (STT) Accuracy
In mail-order reception, it is necessary to accurately understand information where misrecognition directly impacts business, such as names, addresses, product names, quantities, and proper nouns. Verbex will aim for further improvement in accuracy sufficient for actual operation through the continuous improvement of Japanese speech recognition that is strong against addresses, names, proper nouns, specialized terms, and dialects.
3. Strengthened Response to Complex Conversations and Business Flows
On the frontline of mail-order reception, conversations with many branches occur, such as not only simple conversations ordering a single product, but also the addition of confirmation items, references to multiple products, mixing with inquiries, and changing topics midway. To handle such practical tasks, they will proceed with the implementation of branching flows capable of handling complex conversations and business flows, covering even more cases.
Oshimaya, based in Kumamoto Prefecture, has long operated a mail-order business delivering fruits direct from farmers and Kumamoto specialties like horse sashimi nationwide. They pride themselves on a unique distribution channel connecting primary industry producers with consumers, running a business that demands high-quality customer touchpoints in food mail-order.
This full implementation was decided after a Proof of Concept (PoC) confirmed high results in suppressing lost order opportunities and improving the customer experience during temporary spikes in call volumes, which occur during peak seasons, sales of popular items, and infomercial broadcasts.
■ Background and Challenges
In the phone reception business of mail-order sales, incoming calls can temporarily concentrate around the timing of sales promotions for specific products or advertisements/broadcasts. During these peak times, "overflow calls" that disconnect without being answered easily occur due to line congestion or reaching the upper limit of operator capacity, leading to lost sales opportunities.
Phone orders in food mail-order, in particular, cannot be completed simply by mechanically accepting the order details. Customers are often in a higher age demographic, resulting in significant individual differences in speaking style and conversation tempo, and there is a need to accommodate regional phrasing and dialects. Additionally, because the accuracy of understanding names, addresses, product names, and quantities directly translates to order quality, it is a highly difficult business area that is hard to handle adequately with standard automated voice response systems.
Oshimaya advanced the PoC with Verbex to verify whether voice AI could simultaneously reduce lost order opportunities, expand sales, and alleviate frontline burden in such highly difficult order reception tasks.
■ Results Obtained in the PoC
In the PoC, they tackled the highly difficult challenge of mail-order reception via completely free conversation for merchandise that has many elderly customers and dialect speakers. As a result, it was confirmed that the AI handled approximately 60% of calls without needing a callback.
Furthermore, high success rates were recorded for the accuracy of order information acquisition recorded by the AI, including 98% for ordered items, 95% for order quantities, and 94% for advertising media.
These results confirmed that practical operation contributing to the suppression of lost sales opportunities and operational improvement is feasible on the actual business frontline, going beyond mere technical verification. This high technical accuracy and the continuous improvement and support system through the PoC were highly evaluated, leading to the decision for full implementation.
■ Value Aimed for with Full Implementation
The goal of this full implementation is not merely labor saving or automation. It is to support the experience of "being able to properly order the moment you want to buy," which is the most important thing in mail-order business.
By establishing a system capable of immediately answering more customer calls even when call volumes are concentrated, they aim to suppress lost order opportunities. Additionally, creating an environment where human operators do not have to handle all primary responses makes it easier for them to focus on tasks requiring more judgment and consideration, which is expected to raise the baseline of the overall customer experience.
Furthermore, by refining the order flow based on the premise of division of roles between AI and humans, they aim to contribute to stable operations during peak seasons and continuous sales expansion.
■ Future Initiatives - Development of a Unique AI Model Strong in Mail-Order Reception
Verbex will continue to proceed with the continuous improvement of voice conversational technology optimized for the highly difficult area of mail-order reception, in collaboration with Oshimaya, even after this full implementation.
Specifically, they will accelerate the improvement and implementation of unique technologies in the following areas:
1. Improvement of Japanese Conversation Control Accuracy
Verbex will proceed with improvements to its unique Japanese conversation control so that conversations can continue naturally even if the customer interrupts the conversation or provides an unexpected response. In particular, by improving the accuracy of interruption detection and utterance boundary detection, they will realize more natural responses tailored to each customer's speaking tempo and timing.
2. Improvement of Japanese Speech-to-Text (STT) Accuracy
In mail-order reception, it is necessary to accurately understand information where misrecognition directly impacts business, such as names, addresses, product names, quantities, and proper nouns. Verbex will aim for further improvement in accuracy sufficient for actual operation through the continuous improvement of Japanese speech recognition that is strong against addresses, names, proper nouns, specialized terms, and dialects.
3. Strengthened Response to Complex Conversations and Business Flows
On the frontline of mail-order reception, conversations with many branches occur, such as not only simple conversations ordering a single product, but also the addition of confirmation items, references to multiple products, mixing with inquiries, and changing topics midway. To handle such practical tasks, they will proceed with the implementation of branching flows capable of handling complex conversations and business flows, covering even more cases.