Otsuka Foods

Search

News Release

June 11, 2026
Otsuka Foods Co., Ltd.

Otsuka Foods’ Biwako Research Institute Develops “Oishisa LENS,”
an AI System That Predicts Flavor Profiles and Recipes
to Secure Expertise Transfer and Enhance Product Design
58 Years After the Brand’s Launch,
Preserving Bon Curry’s Signature Taste for the Next Generation

TOKYO, Japan, June 11, 2026—Otsuka Foods Co., Ltd. (Head Office: Chuo Ward, Osaka; President: Goro Ikeuchi) announced today that it has developed an AI system that learns from and predicts flavor profiles and recipes. The new Oishisa LENS (Logical Exploration Navigation System) is named after the Japanese term for “deliciousness” or “great taste,” oishisa, and was developed at the Company’s Biwako Research Institute in Otsu, Shiga Prefecture. The Company has begun applying the system to Bon Curry, a long-selling brand now in its 58th year on the market. By turning decades of product development knowledge into usable data, this effort seeks to preserve Bon Curry’s signature taste and pass it on to future generations, while also creating new possibilities of great taste.

Background and Challenges
Amid Japan’s declining birthrate, aging population, and labor shortages, the difficulty of passing on skills to the next generation has become a society-wide challenge. For Otsuka Foods, continuing to deliver the great taste customers have always known is a responsibility and mission owed to society and the loyal customers who have supported the brand for so many years. Accordingly, finding ways to pass down flavor development expertise to the next generation has become an important priority.

At the same time, recent sharp increases in raw-material prices and uncertainty around procurement, driven by climate change and geopolitical risk are challenges shared across the food industry as manufacturers work to maintain consistent taste and quality.

Otsuka Foods’ Biwako Research Institute has a culture in which researchers work tenaciously on a product until they achieve a taste with which they are fully satisfied, often refining the flavor direction and formulation across hundreds of trial batches. However, much of the judgment and ingenuity gained via this trial-and-error process tends to accumulate within individual researchers, and there has been concern that valuable techniques and insights may not be fully passed on when this generation hands over the work to the next.

Even as the environment for procuring raw materials changes dramatically, maintaining the taste customers know and love requires a system that can manage and harness recipe and formulation know-how in an institutional manner.

In response to these challenges, and driven by researchers’ desire to “inherit the precious development assets of the experts who came before us and preserve the flavors people have loved,” the Biwako Research Institute launched a DX project: one that transforms intangible techniques and knowledge into a managed asset and uses AI to support both the transfer of taste-making expertise and greater efficiency in development.

Establishing an Original Predictive Model and Development Platform
■Visualizing complex flavors and preparing data for AI training

  • At the heart of this initiative was the task of converting “deliciousness” — something underpinned by the development team’s experience and intuition, refined over many years — into data that AI could learn from.

    Curry is a food that combines many different ingredients and whose taste and aroma change through heating and simmering, so organizing “deliciousness” into data requires careful design. The team therefore began by standardizing the vocabulary and criteria used in sensory evaluation, which researchers assess flavor and aroma through taste testing.

    For example, the term “curry character,” which is frequently used in sensory evaluation of curry, was understood by some researchers as “the overall impression of the sauce, including the cooked-down richness from simmering,” while others took it to mean “the sense of a complex spice blend, like that of curry powder.”

    Working through such differences in interpretation one by one, the team created a list of 218 phrases to describe “deliciousness” for Bon Curry Gold, the flagship line of the Bon Curry brand. Over about six months, they then interviewed each researcher and prepared more than 1,000 evaluation basis samples, narrowing the list to 16 “taste evaluation attributes” that everyone could use as a common language for evaluation.

    As a result, taste evaluations can now be recorded and visualized as numerical scores, attribute by attribute, accumulating into data available for future use.

■Building an original AI predictive model

  • Oishisa LENS is an AI model developed by combining sensory-evaluation data from these taste attributes with trial recipe data created by researchers. This allows the system to learn both the sensory qualities of delicious food and how it is made, enabling it to predict flavor profiles and recipes.

    Instead of depending on external systems, the model was developed internally, assembling the Company’s own expertise and data. The name embodies the idea of “bringing what could not previously be seen into sharp focus, as if through a lens, and exploring new dimensions of deliciousness.”

    In model validation, Oishisa LENS not only generated predictions that aligned with the intuition of experienced researchers but also suggested novel trial-recipe ideas that the researchers themselves had not considered. On that basis, the system is now being put to practical use in product development work.

■Building a database-driven development platform

  • In parallel, the team built a development platform that combines development expertise with AI training data. Some 40 years’ worth of paper records — hundreds of thousands of pages — were digitized using AI OCR, thereby unifying into a single database the extensive collection of product development records and materials created at the Biwako Research Institute, starting with Bon Curry. Researchers can now search and reference this database instantly whenever needed. Additionally, newly added trial recipes and sensory evaluation data are designed to feed directly into Oishisa LENS as training data, creating a foundation that allows everyone in the organization to utilize these development assets.

Oishisa LENS Capabilities
Oishisa LENS allows, for example, quantification of how much each ingredient in the product contributes to what makes Bon Curry taste like “Bon Curry.” If an ingredient’s supply changes or a flavor needs refining, the AI can analyze data to predict “what recipes might work” and “what they would taste like,” enabling the team to explore alternative ingredients and formulations — and thus continue delivering the taste customers have known and loved for decades.

One researcher remarked, “By verifying the recipes we inherited from earlier generations in numerical terms, we have proof of which ingredients are decisive for the flavor, making the priorities for our trial batches much clearer.”


Furthermore, when consumer preference surveys are integrated into the system as data, it can also predict which consumer segments are likely to prefer a given prototype and the reasons why. By allowing researchers’ intuition and a variety of data types to be handled on a single common platform, the team can now narrow down and propose trial product directions based on data, which will also accelerate the development process.

Looking Ahead
Otsuka Foods is taking the visualization of deliciousness further by combining collected data and predictive technology with measurements from taste sensors, aroma sensors, texture analyzers (instruments that measure mouthfeel), and other tools. The Company plans to extend this initiative from Bon Curry to other brands, leveraging technology while continuing to value human experience and intuition — preserving the flavors people have come to love, passing them on to future generations, and pursuing new forms of deliciousness suited to the changing times.

PDF