Agency Huge Crunched Hundreds of Beer Reviews to Craft the Ultimate Data-Infused Recipe

Since 2013, Brooklyn-based agency Huge has held a weekly meeting of beer geeks who gather to taste and rate unique and rare brews from around the world. Over that time, the Huge Brews team has sampled more than a thousand different beers. Then the tech-savvy group started wondering what would happen if they took all that tasting data, fermented it with some machine learning and tried to formulate the ultimate, metrics-driven beer recipe. For the past six weeks, members of Huge Brews have been brewing their own beer based on those years of tasting notes, along with data from two popular beer review websites. After looking internally at what beers they considered their favorites—which ended up being similar to a juicy IPA—the team then scraped user-generated reviews for about 250 beers, pulling in about 40 website reviews per beer. The end result: two distinct batches including a juicy IPA with blackberry and another with raspberry and lactose—an ingredient that has grown in popularity over the past year with the rise of what's known as the Milkshake IPA. (Huge Brews worked with a home-brewing startup in Brooklyn called Bitter and Esters, which provided brewing expertise and equipment.) Huge Brews is calling its new line of brews Off-Brand IPA. [kickout id="1"] The process obviously began by looking at which beers the Huge Brews team had liked most—a data point they soon realized had been changing over time as tastes and trends evolved. “The ingredients have shifted over the years,” said Kenny Chung, Huge’s director of SEO and the leader of Huge Brews. “We were looking back at the first year or two of ratings and the beers that won back then were just a really good standard IPA and sometimes even an American pilsner. These days, I doubt we even have a standard IPA winning on a weekly basis.” Huge’s analytics team then used natural language processing and machine learning to identify core attributes for what makes the “ideal” beer. After that, the data was broken down into word buckets—based on categories like fruit, color, body, opacity, and alcohol by volume—and then organized in a way similar to how a wine aroma chart might organize grape varieties. Once they came up with a general type of beer they wanted to brew, the team looked back at the categories to see which were rated highly, then used that as a jumping of point. Based on the data, that meant the recipe should be full of citrus, tropical notes and berries along with with a flavor that was both sour and also “oak slash aged,” according to Chung. [kickout id="2"] "[Veteran brewers] did tell us that the most important thing when brewing beer is note taking,” he said. “And that’s how most people refine the recipes, so it was definitely in line with how we were approaching it with a data-informed approach.” The beer, which was finally ready for tasting earlier this week, was then bottled and capped with labeling designs headed up by Huge’s chief design officer, Derek Fridman. “We actually did an internal taste test and everyone gave it a thumbs up,” Chung said. “We’re careful to not be too experimental and did things a little more broadly.” While the agency isn't licensed to sell the beer, it does plan to let a larger group of employees try it at a staff gathering on Friday.