Maintaining Quality in Financial Report Translation—Beyond What Generative AI Alone Can Deliver

A translation support tool that enables timely English-language publication and supports human judgment

Okasan Securities Co., Ltd. (hereinafter, “Okasan Securities”) continues to provide high-quality market information to investors in Japan and overseas. As the company strengthened its communications with international investors, the timely publication of analyst reports in English posed a major challenge, particularly from the perspective of fair disclosure.

Previously, manual translation and revision were the norm, resulting in a delay of several days between the release of the Japanese version and the English version. In financial markets, where the freshness of information is critical, this conventional manual translation process was reaching its limits in both speed and quality.

To overcome this challenge and build a system for accurate and rapid information delivery, Okasan Securities introduced Yaraku Translate, a generative AI translation platform. We spoke with the two people who led this transformation on the front lines.

Interviewees

■ Background

The limits of manual work seen firsthand by a professional translator

“Soon after joining the company, I noticed several ways we could improve our workflow,” says A, who joined Okasan Securities’ Global Research division in 2024.

At the time, demand for English translations of Japanese-language reports written by analysts was rising rapidly at Okasan Securities, but no translation tool had yet been introduced. The standard process was to paste the source text into Word and translate and revise it manually, one line at a time.

What was especially exhausting for the team was the sheer amount of manual work required even before the translation itself.

「Previously, we pasted the source text into Word and translated it manually, one line at a time. Because we had a fixed template, we also had to adjust fonts and format the document as we worked. It was difficult to keep translating without breaking our train of thought, which made the process inefficient.」（T）

Translating a single report took considerable time, and for longer reports, the delay before publication could approach a week. Transforming the workflow had become an urgent priority to achieve efficient, timely English-language publication.

■ Why they chose Yaraku Translate

“Unrequested summaries” show a lack of respect for analysts | Why a financial institution chose a dedicated tool over general-purpose generative AI

Secure generative AI solutions for enterprises have become increasingly common. Even so, Okasan Securities deliberately chose Yaraku Translate, reflecting the exacting standards of accuracy required of financial institutions.

T points out the following risks of using general-purpose generative AI.

When we used general-purpose generative AI on its own, it sometimes made simple calculation errors or produced false figures simply because a unit—such as wattage—had changed. Numerical errors in financial reports can be disastrous.

A was also concerned about certain tendencies unique to AI.

Generative AI sometimes creates unsolicited summaries in an attempt to be helpful or adds information not present in the source. But we respect our analysts and want to preserve 100% of the intent behind the words they have carefully written. That is why AI that paraphrases or summarizes without being asked can actually get in the way of our work.

Yaraku Translate, by contrast, can display translation results from multiple AI engines side by side for comparison.

「Rather than simply trusting a translation produced by AI, we can compare several options and choose the one that best reflects the source text’s intent. This process of comparing and evaluating alternatives was exactly the kind of quality control we were looking for.」（A）

■ How they use it

Fundamentally different from one-off chat AI | Turning individual expertise into a company asset and helping AI improve

Since adopting Yaraku Translate, the two have made full use of its features to stabilize translation quality. Above all, they place the greatest emphasis on turning knowledge into a reusable company asset.

1. An environment that “gets smarter the more you use it” through term registration

With general-purpose chat AI, even if you teach it the correct terminology, that knowledge is often forgotten after the conversation ends. With Yaraku Translate, however, registered terms accumulate as reusable assets.

For example, to prevent “Kimetsu no Yaiba” from being translated arbitrarily and ensure that the official English title, “Demon Slayer,” is used, we research the official websites of companies such as Toho and Toei and register the terms in the Glossary. We now have more than 1,400 registered terms. Entering 1,400 rules into an AI tool every time would be impossible, but with Yaraku Translate, we only need to register them once for the AI to become smarter on its own.

This mechanism helps ensure consistent quality regardless of who performs the translation and has become essential to the organization’s translation workflow.

2. Automating “custom rules” with the prompt feature

One feature T uses particularly often is the ability to preconfigure instructions for the AI, or prompts.

「We used to build Excel macros to convert notation for amounts (“JPY 100 billion → JPY100.0 billion”) and fiscal years (“FY25/3”). Now we configure these rules in advance in the tool. The AI then applies them automatically during translation and consistently produces output in the correct format.」

■ Results and future outlook

From “leaving translation to AI” to “working with AI” | Pursuing timely information delivery with a trusted partner

Introducing Yaraku Translate dramatically changed the translation workflow. The team no longer spends unproductive time translating from scratch by hand or fixing formatting. Instead, they start with draft translations suggested by AI and focus on rewriting, fine-tuning, and reviewing.

Having suggestions from multiple engines has significantly reduced the time we spend thinking about grammar and structure. Most importantly, we have been able to publish information faster while maintaining a level of quality that makes us confident saying, “This is ready to publish.”

Looking ahead, the team hopes to see further enhancements, including more accurate text recognition in images, and plans to expand its use of Yaraku Translate with the goal of publishing analyst reports simultaneously in Japanese and English.

■ Post-interview note

After the interview: The value of human expertise stands out even more in the age of AI

“It’s scary to leave everything to AI. That’s why a human has to make the final check.” This was something A and T said repeatedly during the interview, and it left a strong impression on us.

While much of the conversation around AI focuses on automation, on the front lines of finance—where an error of even one yen is unacceptable—AI remains an “excellent assistant,” but human judgment remains central. Their comments conveyed a strong sense of professional pride: respect for the analysts and a determination to deliver accurate information to investors.

Beyond the financial industry, we are entering an era in which a company’s competitiveness will increasingly depend on how effectively it uses AI. Using Yaraku Translate, Okasan Securities is combining technology with human expertise to deliver new value to investors around the world. Its efforts may well offer a model for the future of global business.