SBI Securities reduced average customer-centre call time by 1.6 minutes using RightTouch's QANT Speak AI operator. Applying that measured reduction to the major telephone lines' call volumes gives an estimated saving of about 2,933 hours per month across the centre. This monthly total is an extrapolation, not a separately measured reduction across every call.
The project addressed rising inquiry volumes following the new NISA programme and stronger online-trading security requirements. Recruitment and training could not keep pace. Traditional telephone menus became harder for customers to navigate as more options were added, and calls to the wrong line created transfers and repeat explanations. SBI Securities already used RightTouch's QANT Connect to match web visitors to the right agent from their on-site behaviour, but many customers found the phone number through a search engine and called directly.
SBI Securities compared competing services and prioritised natural conversation and the ability to change flows as needs evolved. QANT Speak asks short follow-up questions to distinguish issues such as forgotten credentials and two-factor authentication. It can adjust its response when a caller speaks while the AI is talking. During identity verification it reads details back, and when unsure of a name it offers several candidates for the caller to confirm.
An early proof of concept revealed that agents were often asking customers to repeat information already collected by the AI. Summaries had been accessible only through a link to a separate page, which was awkward to open as a call arrived. The team integrated QANT Speak with Genesys Cloud so that essential information appeared directly on the incoming-call screen.
Supervisors at the Yokohama customer centre helped identify operational problems. The project first concentrated AI-routed calls among experienced agents, configuring skills so the same line received both AI-routed and directly routed calls for comparison, and gradually expanded the approach. Staff shared screen-use practices and ways to lead the conversation without making customers repeat their request.
After roughly a month of revising inquiry categories, call tags and flows, with repeated automated testing, routing accuracy exceeded 90%. For some inquiry reasons, the AI achieved an 85% self-resolution rate, defined as no transfer during the interaction and no repeat call within 24 hours. Full implementation began in June 2026.
The company plans to extend the approach beyond technical inquiries, reduce transfers and expand self-resolution. Future CRM integration and personalised matching based on customer history remain plans, not achieved results. The interview also identifies operator concerns about workload and job security as challenges requiring explanation, training and continuing operational support.