New Thinking

To bring “good” taste to scale, first build trust

Share this post:

Brian Smith knows wine. The co-founder of Winc and the company’s Chief Wine Officer (“The title you get when you’re allowed to give yourself your own title”), he studied to become a sommelier and even worked in wine production in Provence and Argentina. But when pressed, he expresses resistance to the notion that one becomes a “better” wine drinker over time and with experience: “The whole idea of connoisseurship is kind of interesting,” he says euphemistically.

Wine can be intimidating, and Smith knows as well as anyone that good taste is subjective. As Spotify does for music, Stitch Fix does for clothing, and Netflix does for movies, Winc (formerly Club W) treads the tricky terrain of providing algorithmic recommendations to suit the individual tastes of its customers. But like its fellow AI recommendation engines, Winc responds directly to its customers’ wants, building consumer trust in the company as a reliable arbiter of good (read: your) taste. Then, on occasion, it leverages that trust to expand customers’ palates.

Typically, a service providing personalized recommendations runs on the basis of a machine-learning algorithm: Customer feedback allows these services to “learn” more about an individual’s preferences and tastes, enabling them to provide increasingly customized suggestions. In theory, your Spotify Discover Weekly playlist should sound better and better each time you download a new batch of songs, as the service accumulates data about what types of music you seek out and which artists you skip.

When Pandora’s meticulous Music Genome Project kicked off its first round of financing in early 2000, and for the better part of a decade thereafter, media coverage hailed the project as a sophisticated endeavor and a true service for music-lovers who are pressed for time. Today, the company’s trained employees still analyze individual songs for 450 different musical “genes,” and Pandora credits its Human-in-the-Loop (HITL) model with its ability to provide music for music people, much in the way Apple harnessed the tastes of real-life DJs to build Apple Music. “Algorithms are really great, of course, but they need a bit of a human touch in them, helping form the right sequence,” Apple’s Jimmy Iovine told The Guardian two years ago.

For services leveraging a Human-in-the-Loop model and those that rely entirely on human recommendations, emphasizing the work of real people lies at the heart of building customer trust. On its website, Stitch Fix makes little of its leading role in the world of AI fashion and personal styling, but customers are assured that (human) stylists will handpick the items that appear in their subscription boxes. (The company’s VP of Data Science, Brad Klingenberg, provides an excellent introduction to HITL and the Stitch Fix model in a video lecture available from Kaizen Data.) Mubi, a streaming service offering an evolving roster of 30 handpicked films, has positioned itself as a sort of anti-Netflix: In a 2015 interview with The Verge, Mubi CEO Efe Cakarel called the Netflix user experience “frustrating,” arguing that it takes customers too long to find an appealing movie and claiming that the algorithm “categorically doesn’t work.” The growth of AI-driven digital audio platforms like Spotify—and the continuous financial losses suffered by Pandora—have made conspicuous the challenges of scaling HITL recommendation algorithms. But successful recommendations, when they’re made, bolster a customer’s willingness to break the rules learned by a machine.

Because, as Smith notes, taste is a mutable thing: “Even people with good taste, it’s constantly evolving.” He explains that as a Winc member, “you can develop an understanding of wines that you like, you get to try things that maybe you don’t like, and you take that into consideration, and there’s also this opportunity to step way outside the line and try some really new things.”

Winc has evolved substantially since the company’s Club W days. Once a subscription service shipping out third-party wines, it now boasts an entirely proprietary product, with over 100 varietals of Winc-manufactured wines currently available for purchase. The company’s database of customer preferences allows Smith and his cohorts to develop new wines that their members will love, but may not know they’ll love. “Over the past two years, we’ve been seeing a trend that is somewhat contrary to the historical idea that Cabernet Sauvignon is king and [that] people like really big, juicy, ripe red blends,” Smith says, explaining that the company’s consumer behavior and interaction metrics have indicated otherwise. “We’ve been seeing that people are gravitating towards lighter reds that are… easy-drinking and fresh.”

But playing to the wants of customers also creates an opportunity for Winc to impress new wines upon them—and perhaps to expand their tastes. Smith notes that Winc’s primary objective is to provide members with wines they’ll enjoy, meaning that the company also tracks the wine trends that their members are likely to be excited about. “We’ve been able to build trust by having targeted recommendations,” Smith explains. However, he notes, “Once we build trust, we now have—on a more limited scale—we now have the freedom to create things that live outside what might normally be recommended to you.” Last year, one of Winc’s fastest sellers was a $22 orange Pinot Gris, a product of what the company called “méthode expérimentale” and something American consumers were unlikely to recognize.

“Wine is snobby, and that’s part of the problem,” Smith says. That “interesting” notion of connoisseurship, he explains, “has created a barrier between the customer and the product, and so we have an opportunity to kind of break that barrier down through our service.” Winc’s small-d democratic approach to producing and selling wines consumers are likely to enjoy removes the barrier to entry, but it also makes it easier for oenophiles like Smith to introduce customers to other things, to expand—perhaps, improve?—their tastes.

In the end, Smith says, it’s ease of exploration that makes it possible for a wine drinker to develop their palate and gives them the confidence to talk about wine and order it in a restaurant. “We revere the history and culture around wine and the craft of wine, there’s no doubt about that. Love it. Incredibly passionate about it,” Smith says. “But the opportunity to…re-write things on behalf of the customer is an amazing opportunity, and I think that’s at the core of Winc.”

Learn more about IBM iX

Add Comment
One Comment

Leave a Reply

Your email address will not be published.Required fields are marked *


Adams

Thanks for the valuable info. I have a small business and I’ll surely follow these footsteps to make Sure I build trust in the online industry.

Reply
More New Thinking stories

Why is VRChat so popular, and what’s it mean for the future of virtual reality?

There are precisely three things to do in the social VR app called VRChat. You can build a room. You can create an avatar to exist within that room and those created by others. Finally, you can take on the persona of this character and talk to other avatars. Apart from a few minigames, there […]

Continue reading

Three Ways Metalearning Will Change the Face of Artificial Intelligence

If you were going purely off of what is depicted in movies and television shows, you’d think that using AI was an extremely streamlined process. Got a problem? Just push a few buttons or simply say what you want to your computer—and a few moments later the AI will have the issue resolved. The reality […]

Continue reading

How Funny Should a Chatbot Be?

Question: Why do scuba divers always fall backwards out of the boat? Answer: If they fell forward they would still still be in the boat. Ba dum tsh. This dad joke comes courtesy of Microsoft’s Cortana. Chatbots, just like dads, want to be funny. As chatbots and other conversational interfaces become a part of our […]

Continue reading