Major JP retailer cuts contact-center running costs 40% with Shisa ASR API

A major Japanese retailer runs nationwide contact centers that handle more than 10,000 Japanese-language calls a day. They had been using a global cloud ASR vendor, but accuracy on Kansai and Kyushu dialects, and on in-house product and campaign names, was consistently falling short.

Shisa.ai domain-adapted Shisa ASR API on the customer's own consented call logs and wired in a custom dictionary covering the full product catalog, store names, and active campaign vocabulary. We paired that with a ShisaLLM-based post-call summary step that automates most of the operator's after-call write-up.

Ninety days into production, word error rate was 32% better than the previous vendor, and average after-call work time dropped by 2 minutes 15 seconds per call. Combined ASR spend plus labor cost per handled call fell roughly 40%, letting the retailer hold peak-hour answer rates steady with the same headcount.

A phase-two rollout will add real-time call summaries and a ShisaLLM-powered quality-monitoring feed for supervisors, with every site scheduled to be live by late 2026.