Designing the "Deliciousness" of Beer with AI: An Era Where Kirin's Unique AI Identifies Ingredients and Proposes Flavor Creation

1. What is Happening? — AI Utilization Delving into "Flavor Design"

Kirin Holdings is fully utilizing its independently developed AI in beer brewing, aiming to reach a state where it can propose "how to manipulate which ingredients" to achieve the desired flavor. Okinawa Times + Plus+1

The key point is not so much that AI "automatically generates recipes," but rather that it breaks down the factors that constitute deliciousness and pinpoints areas for improvement.

Traditionally, creating the taste of beer involved measuring ingredients with analytical instruments, but it was challenging to link "how those numbers affect the impression when tasted," and ultimately, it relied heavily on the developer's experience and repeated prototyping. Okinawa Times + Plus+1

Kirin is tackling this "final hurdle" with data and AI.

2. What is the Unique AI "FJWLA (Fujiwara)"?

At the core of this initiative is the unique AI "FJWLA (Fujiwara)."

It is said to have been developed based on the ingredient data accumulated by Kirin and the results of **consumer surveys (about 20 years' worth)** from tasting samples. Okinawa Times + Plus+1

With this AI,

• which ingredients

• in what way

• contribute to deliciousness (evaluations of bitterness, richness, aroma, etc.)
  can be quantified, as reported. Okinawa Times + Plus+1

Furthermore, in cases where there are challenges with "bitterness" in new product development, it is reported that the AI suggests ingredient options for improvement, and the AI-utilized options received higher evaluations in actual comparative tastings (based on reports). Shimotsuke Shimbun Digital

The results are planned to be gradually reflected in beers released from March 2026 onwards. Okinawa Times + Plus+1

3. Why is the "Deliciousness" of Beer Difficult?

Beer becomes a vast collection of chemical components through the interaction of raw materials (malt, hops, water), yeast, and process conditions (temperature, time, oxygen, aging, etc.).

For example, even when it comes to bitterness, it is known that iso-alpha acids, which are produced by the isomerization of alpha acids derived from hops during boiling, are the main factor. PMC+1

Regarding aroma, esters and higher alcohols produced by yeast and hops influence the fruity/floral impression. PMC+1

The tricky part here is that

• when ingredient A increases, it improves, but if ingredient B is also high, it worsens

• some ingredients affect "aroma" but have the opposite effect on "aftertaste"

• the perception of the same ingredient changes with drinking temperature and carbonation
  , making it a "combination problem."

In other words, determining "what to increase or decrease to achieve the desired impression" is not a simple matter of addition and subtraction.

4. What Does FJWLA "Propose"? — Imagining Its Use on the Ground

Based on the expressions in reports, FJWLA is thought to play the following roles. Okinawa Times + Plus+1

4-1. Articulating Challenges: Translating Sensory Evaluations into "Ingredient Issues"

A common occurrence in the development field is the challenge of translating "impressions" like
"bitterness is too strong"
,
"richness is lacking"
, or

"aroma is weak"

into design variables.

FJWLA creates a starting point for discussion by presenting "significant contributing ingredients" and "balances to be adjusted" based on these impressions.

4-2. Generating Options: Shortening the Maze of Prototyping

Even adding just one hypothesis of "it might be better to change this" can alter the number of prototypes and the design of comparisons.

The report that AI suggests ingredient options for improvement is precisely here, where it is expected to narrow the exploratory space of prototyping. Shimotsuke Shimbun Digital

4-3. Reducing Dependency on Individuals: Turning Experience into "Forms"

Flavor creation inevitably leaves room for craftsmanship.
However, if the reasons for judgments and points of focus are not shared, they cannot be reproduced as a team.

If FJWLA can explain "why it works," it can also contribute to skill transmission.

5. Kirin's AI Utilization is Not "First Time"

In fact, Kirin has been advancing the foundation for using AI in the beer domain for some time.

5-1. "Brewing Takumi AI" — The Concept of Back-calculating Recipes from Targeted Flavors

Kirin has announced the addition of a function to its machine learning-based beer development support "Brewing Takumi AI" that generates recipe candidates by back-calculating from targeted flavors (numerical indicators). KIRIN - Kirin Holdings Company, Limited

This announcement indicates that while human senses are ultimately necessary, AI can expand options and contribute to development efficiency and skill transmission. KIRIN - Kirin Holdings Company, Limited

5-2. Introducing AI into Factory Planning — Supporting the "Brains of Production" with AI

Additionally, Kirin Brewery has jointly developed with NTT Data a system that automatically plans the brewing and fermentation processes (brewing and yeast planning) using AI, conducting trial operations at all nine factories, with an expected creation of over 1,000 hours annually. KIRIN - Kirin Holdings Company, Limited+1

What can be seen from this is

• research and development (recipes, flavor creation)

• production planning (brewing and fermentation scheduling)
  , showing that they have been accumulating AI implementation experience in these "two wheels."

FJWLA can be seen as an attempt to integrate "sensory x ingredients x consumer evaluation" and move to the next stage in this extension.

6. The Significance of Feeding "20 Years of Consumer Surveys" to AI

The core of FJWLA lies not only in ingredient data but also in having long-term data on the evaluations of people who drank the beer. Okinawa Times + Plus+1

Beer evaluations by expert sensory panels and general consumer preferences (likes and dislikes) do not always align.

Therefore, the more robust the consumer surveys,

• which demographic

• perceives what kind of flavor changes

• as "delicious"
  can be mapped out.

When this map is linked to ingredient movements, it becomes easier to design by back-calculating from the "targeted customer value."

7. When and Where Will It Be Reflected? — From Products Released After March 2026

According to reports, the results of FJWLA will be gradually reflected in beers released from March 2026 onwards. Okinawa Times + Plus+1

Here, "gradually" is important, as it suggests not dramatically changing the main flavors all at once,

• but rather in new products

• limited editions

##HTML_TAG_