Working Surfacelink · 6 Nov 2026
04
Many loose paper notes blowing toward a single letterbox in a wooden door, and behind the door a sorting rack of pigeonholes where each note drops into exactly one slot, the whole scene drawn with wide empty margins on every side.
Microsoft Inside Track

Beyond translation: How we’re making Microsoft Copilot work better in Japanese

Kevin Kotani · 5 November 2026

How-to · localisation · collaboration · Human approval

Read the original
Takeaway

Microsoft gave Copilot's Japanese-language problems one reporting route with named owners and had native speakers test early; daily use in Japan then peaked at 93%.

Summary

Kevin Kotani, a digital experience lead in Microsoft Digital, Microsoft's IT organisation, described how the company made Copilot, its AI assistant, work better in Japanese. Staff in Japan found that accurate translation was not enough: Copilot also had to get names right, follow meetings that switch languages and use polite business Japanese. Problems were reported across many teams, which caused duplicate bug reports and competing priorities.

Key points
  • A localisation team in Japan holds regular meetings where field staff, the Copilot product group and language experts review issues, agree priorities and give each problem one owner.
  • More than 300 native speakers tested a preview in July 2025 through Microsoft Elite, the employee early-access programme, before broad rollout.
  • Speech AI model updates raised name recognition in Japanese meeting transcripts from about 40% to 66%, and a fix stopped Copilot reading the honorific "san" as part of a name.
  • Daily use among staff in Japan peaked at 93%, against about 71% across the company, and the team now tracks how many language issues it catches before release.
Implication

A product team shipping an AI assistant in several languages can give language problems one reporting route with named owners, and recruit native speakers to test before release. This is Microsoft's account of its own product, written by its internal IT organisation.

Plays
Product

Before releasing an AI assistant in a new language, have employees who are native speakers of it test a preview and report errors such as misread names or the wrong level of politeness.

Product

Route every language-quality report through one shared channel where field staff, product owners and language experts agree priorities and name one owner for each issue.

Derived by Working Surface from the article; more in the Playbook. Source line: Find issues early with native speakers, then scale fixes in a managed way.

Source issue

6 November 2026: teams scaling AI agents are replacing case-by-case checking with one repeatable path that every piece of agent work passes through.