Beverage Makers Turn to AI for New Flavors

Artificial intelligence is moving from the lab into the tasting room, and beverage companies are using it to study flavor data, speed up product development and target younger drinkers who want new tastes. Distillers, brewers, wineries and retailers are testing machine learning systems that can read chemical profiles, consumer reviews and sales data, then suggest ingredient combinations that are more likely to work in the market.

The shift is part of a broader push across the alcoholic beverage industry to shorten research cycles and respond faster to changing demand. Global brands are under pressure from consumers who are drinking less often, choosing lower-alcohol products or looking for more distinctive flavors. At the same time, companies want tools that can help them decide which recipes to test before they spend money on pilot batches and sensory panels.

Some of the earliest high-profile examples came from spirits. In Sweden, Mackmyra worked with the Finnish artificial intelligence firm Fourkind to create Intelligens, a whisky blend developed with machine learning. The system was trained on the distillery’s past recipes, sales records and tasting feedback. Human blenders then refined the output before the whisky was released in 2019. The product later won a gold medal in a 2020 competition, giving the project credibility beyond its novelty value.

In Britain, Circumstance Distillery used a neural network called Ginette to help create Monker’s Garkel, which it described as an AI-made gin. The system was fed botanical profiles and gin recipes, along with naming ideas and label concepts. The distillery still made final adjustments by hand, including changes to alcohol content and botanical balance. The result showed how artificial intelligence can be used not only for formulation but also for branding.

Large companies are also investing. Diageo acquired Vivanda in 2022 to bring flavor-mapping technology in house. Vivanda’s FlavorPrint system asks consumers about their preferences for fruit, spice and herbal notes, then builds a taste profile that can be used for recommendations and product development. Diageo has said it wants to use such tools to better understand regional preferences and tailor products for different markets.

Flavor houses are doing the same. Givaudan has promoted its ATOM platform as a way to reduce trial-and-error in formulation. The company says the system helps identify which ingredients support or weaken a target taste and can suggest combinations that meet consumer expectations while reducing development time. In one case cited by the company, ATOM helped create a snack flavor with 33% less salt while preserving taste quality.

The technology depends on data that beverage companies have not always connected before. Sensory panels provide structured tasting scores. Chemical analysis shows what compounds are present in a wine, beer or spirit. Consumer reviews add language about aroma, body and finish. Sales data show what people actually buy. Artificial intelligence systems combine those inputs to look for patterns that humans may miss.

In practice, that means models can be trained on thousands of tasting notes or hundreds of chemical markers from gas chromatography tests. They can also process text from online reviews or social media posts to detect what consumers say they like. Some systems use gradient boosting or neural networks to predict whether a recipe will score well in blind tastings. Others use generative models to propose new blends by sampling from learned flavor patterns.

Academic research has helped push the field forward. At the University of Geneva, researchers used artificial intelligence on chemical data from Bordeaux wines and reported that their model could identify the estate behind each sample with 100% accuracy in a controlled test. In another study focused on beer, researchers combined chemical measurements, sensory data and consumer reviews from roughly 250 beers. Their model outperformed conventional statistical methods and helped identify compounds that improved blind-tasting scores when added to a beer recipe.

Brewers have been especially active because they already work with large volumes of process data. AB InBev has used machine learning to improve filtration consistency and speed at scale. The company said one pilot cut filtration time by 40% to 50% while improving consistency across batches. That kind of operational gain matters because even small improvements can save money across a global brewing network.

Retailers and recommendation platforms are also part of the trend. Wine apps such as Vivino use machine learning to match users with bottles they may enjoy based on ratings and past behavior. Other startups, including Preferabli and Tastry, build taste profiles from expert descriptors, chemistry data and consumer feedback so they can recommend wines or guide product placement.

Tastry has said it uses large datasets of consumer palates to predict how different drinkers will respond to a wine or beer. Aromyx takes a different approach by trying to digitize taste and smell through biosensors, then matching those profiles against shopper preferences. The goal is similar: turn subjective flavor choices into data that can be used for product design and sales.

The appeal is clear for companies trying to reach Gen Z consumers, who tend to favor novelty, flavor intensity and convenience formats such as canned cocktails and ready-to-drink drinks. Industry research has shown that younger legal-age drinkers often look for fruit-forward flavors, international influences and lower-alcohol options that still taste complete. That has pushed producers toward tropical fruit notes, spice blends and other combinations that feel fresh without being too unfamiliar.

But there are limits to what artificial intelligence can do in beverages. Taste is subjective and shaped by culture, age and region. A model trained mostly on Western consumer data may not work well in Asia or Latin America without adjustment. Companies also worry about privacy when they use customer surveys or review data to build taste profiles.

There are regulatory questions too. If an algorithm suggests a new botanical mix or ingredient combination, producers still have to meet food safety rules and labeling laws in each market where they sell it. There is no special legal category yet for AI-created recipes in most jurisdictions, so companies must treat them like any other new formulation.

Intellectual property is another issue. Beverage recipes are often protected as trade secrets rather than patents, but artificial intelligence complicates ownership when software helps create the formula. In most cases, the company commissioning the work or the human developer guiding it will control the rights, but lawyers say contracts need to be clear before projects begin.

For smaller producers, the barrier is lower than it once was because cloud-based tools now make machine learning more accessible. A winery or distillery does not need a full internal data science team to start experimenting with AI-assisted development. It can use outside platforms or partner with vendors that already have flavor databases and modeling tools.

That is one reason interest is spreading beyond large multinationals. Producers see artificial intelligence not as a replacement for winemakers, brewers or distillers but as a way to narrow choices faster and test more ideas with less waste. In an industry where one failed launch can cost months of work and significant capital, that promise is drawing attention quickly across categories from whisky and gin to beer, wine and retail recommendations.