Breaking language barriers: Our journey to perfect Microsoft 365 Copilot localization

We in Microsoft Digital, the company’s IT organization, are at the forefront of innovation, constantly striving to enhance the experiences of both our employees and external customers. As “Customer Zero,” we are the first to use and test our own products such as Microsoft 365 Copilot, ensuring they meet the highest standards before they reach the market. This unique position allows us to identify potential issues, gather valuable feedback, and make necessary improvements, all while experiencing our products just like our customers do.

What Microsoft Digital is seeing around the world

Language-related issues in technology products are a significant challenge globally, especially when introducing new technologies like Copilot. Identifying and taking action on these issues is one of our main responsibilities in our role as the company’s Customer Zero.

We have encountered numerous language and culture-related cases that are impacting our customers and employees, putting a spotlight on language support for Copilot.

Issues include incoherent meeting transcripts and recaps when switching between languages, misidentification of languages and dialects, and degraded performance and translation quality in languages other than English. These problems become urgent in the current global environment where multiple languages can often be used in a single meeting, and translation tools and AI models struggle to keep pace.

Enhancing localization for Microsoft 365 Copilot in Japan

In our mission to empower every person and organization on the planet to achieve more, it’s crucial to meet people where they are and enable them to be productive in their context. While Microsoft has made significant strides in providing localized access to our products and services, products like Copilot demand even higher reliability and accuracy of localization to be embraced globally. Here we look at the experience of a single market—Japan—and use this case along with global feedback to suggest remedies for language-related issues impacting Copilot’s success in non-English-speaking markets.

Japan ranks 92nd out of 116 countries in the English Proficiency Index (EF). Less than 30% of the population is able to speak English and less than 10% are fluent speakers. This highlights the importance of reliable localization services that ensure our products are inclusive and accessible to all.

Market opportunity for generative AI in Japan

The generative AI market in Japan is projected to reach approximately $1.31 billion in 2025, with a compound annual growth rate of 46.54% from 2024 to 2030. Japanese companies are increasingly leveraging generative AI to streamline operations, improve efficiency, and deliver innovative products and services, particularly in the manufacturing, healthcare, and customer service sectors. The Japanese government’s active support for AI research and development through funding programs and strategic initiatives is expected to drive further growth in this market.

Challenges and opportunities for Copilot in Japan

Despite the promising market opportunity, Copilot faces several challenges in Japan. The internal Net Satisfaction (NSAT) score for Copilot in Japan is lower than it is in English-speaking regions. The key challenges include:

• Complexity of collaboration: The involvement of multiple teams across multiple organizations in the localization effort for Copilot in Japanese has made collaboration challenging. This fragmentation leads to confusion over the prioritization of requests and multiple escalation paths for the same issues, as well as duplication of effort and a lack of transparency in triaging competing improvement requests.
• User experience and quality of responses: Customers and employees have expressed the need for improved localization for Japanese. Users are often dissatisfied with the quality and accuracy of Copilot’s responses in Japanese, noting that the output often doesn’t meet expectations or is less helpful than free versions of similar tools. A number of functions are available in English but not yet in Japanese, which creates barriers to adoption and difficulty in proposing the product to customers in Japan.
• Linguistic complexity and cultural differences: The Japanese language has a unique linguistic structure, including the absence of word separation in written Japanese, differing grammar constructs, and heavy reliance on context for meaning. Japanese also has several levels of politeness that must be considered in addition to other cultural customs. For example, the honorific “さん” (san) is often misinterpreted as part of a person’s name, confusing translation engines.
• Performance and reliability: Frequent errors and slow responses in Japanese are significant concerns for users, which negatively affects trust and satisfaction.

What is Microsoft Digital doing to address localization support?

As part of the effort to improve Copilot, our team is constantly collecting feedback through various channels such as built-in tools within the apps, internal surveys and focus groups, Teams channels dedicated to collecting feedback, and meetings with customers, and we pass these findings on to the engineering groups. When we identify issues or feature requests, either internally or from customers, we log and escalate them in a customized internal tool.

For each product, we have a lead business program manager who reviews each entry, clarifies the details with the employee who entered them if necessary, and then directs them to the appropriate product group, who will prioritize and triage the requests. Our lead business program managers work directly with the product groups to follow up on requests and provide additional information as necessary until the request has been resolved. We then follow up with the employee or customer who raised the issue or request to confirm that they are satisfied with the result.

At the local level, we also meet with customers, often IT to IT, to not only learn from customers but also to share our experiences with them. We discuss the challenges we have faced at Microsoft when deploying products and solutions so that customers can more quickly and easily overcome similar challenges should they arise in their deployment efforts.

Strategies to improve language support

Some of the techniques we’re applying to address these challenges include:

• Creating a complete overview of localization efforts: Establishing a clear roadmap for languages across all products and streamlining feedback channels to significantly improve collaboration and enhance the efficiency of localization efforts.
• Developing a localization playbook: Incorporating language and localization considerations from the very beginning of product design to ensure that localization is seamlessly integrated into the development process, leading to better outcomes.
• Enhancing linguistic accuracy: Collaborating with native Japanese-speaking translators and localization experts to improve the quality of translations and ensure linguistic accuracy.
• Involving the Customer Zero team earlier: Engaging Microsoft employees to use new products and give linguistic and cultural feedback in the early stages of product development to identify and resolve issues before the product reaches general availability, leading to a smoother launch with customers.
• Improving translation quality: Expanding and customizing the Japanese-language datasets used to train AI models to account for cultural and linguistic differences, along with improving translation algorithms both internally and in collaboration with customers to enhance the overall quality of translations.

By addressing these challenges and implementing these actions, we can enhance the localization of Copilot in Japan and other non-English-speaking countries and ensure a more inclusive and productive experience for all users.