Leading Japanese media group cuts localization time in half with Shisa Translation API

One of Japan's top media groups had spent years stitching together translation memory vendors and in-house reviewers to localize articles for English- and Chinese-speaking audiences. The result: an average of 72 hours between first draft and published translation — too slow for breaking news.

Shisa.ai fine-tuned Shisa Translation API on ten years of the group's archived articles and wired a domain-specific glossary (proper nouns, person names, idiomatic phrasing) into every API call. Editors can now kick off a draft translation with a single click from inside their existing CMS.

Three months in, English first-draft time is 38% shorter on average, and the in-house reviewer rejection rate has dropped by more than half. Multi-language publishing lead time fell from 72 hours to 36 — a 50% reduction — making simultaneous international release of breaking stories a realistic option for the first time.

The team is now piloting Shisa Speech API for auto-generated audio versions of articles and exploring Chotto as an editorial-desk assistant, with a production rollout planned for late 2026.