I, sommelier
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
You only need to see the way a wine list gets hot-potatoed around a restaurant table to realise choosing wine still makes most people anxious. And there will never be enough good sommeliers and wine merchants in the world to make that pain go away. Pam Dillon, co-founder and CEO of wine app Preferabli, believes AI could hold the answer. And after more than $22mn of investment and a decade of development, the former investment banker is taking her idea to the world.
Dillon is an energetic New Yorker in jazzy feline glasses; her current drinks of choice are Armagnac Old Fashioneds, peated whisky and orange wine. “When I was at Goldman Sachs, I was at least as well-known for my ability to choose wine and spirits for the closing dinners as I was for any of the technology transactions I did,” she says. “I was always really focused on the preferences of individual clients, because I wanted to stand out in a landscape that was very competitive.”
Dillon had the idea for Preferabli in 2007, “long before anyone was talking about AI. I couldn’t shake the idea that we could bring together supply and demand in the world of wine and spirits in the context of individual preference. How could we codify the whole thing?”
Preferabli’s raw material is a proprietary database of more than two million (and counting) wines, spirits and beers, which have been evaluated by a team of real-life Masters of Wine and Master Sommeliers for more than 800 different characteristics. These range from the technical (grape variety, vineyard elevation, alcohol level) to qualitative (balance, finesse) and gustatory (floral, herbal, smoky). This “algorithm food”, as Dillon terms it, is then overlaid with a whole host of generative AI capabilities, which allow users to rate wines and spirits, and receive ever-more accurate recommendations based on their own personal tastes.
It’s access to this database that makes Preferabli superior to other wine apps such as Vivino, says Dillon, which “makes recommendations for wines based upon what other users think of them. Vivino also has deals with retailers and offers wines for sale based on revenue-sharing arrangements,” she adds.
I tried the free consumer app (which is still a work-in‑progress) and found it a bit clunky, but the champagne recommendations it gave me after I’d plugged in a few favourites were a good mix of grandes marques and more esoteric growers. When it came to Japanese malts, though, it stuck rather drily to a few big names. A paid-for version of the app, which unlocks the database’s full capabilities, will be forthcoming. For now, Preferabli’s focus is licensing its software to trade. Dillon envisions a world where “[hotels, restaurants, etcetera] have a profile of guests or diners when they check in – increasingly many hospitality venues and reservation websites do – which shows what they’ve ordered in the past and their preferences, based on their Preferabli interaction. So the sommelier can then go to the table with the ability to have a completely different conversation about the wine.
“Alternatively, the guest can use their smartphone to click on a QR code that takes them to a website that has the venue’s drinks list pre-loaded with all the Preferabli functionalities built in. This is an increasingly popular option partly because it’s very, very hard to get that kind of labour [ie, trained sommeliers] right now.” Preferabli can re-order a restaurant’s wine list in line with a guest’s own preferences, or take more adventurous drinkers off-piste. Excitingly, it will soon also be able to rate wines according to their environmental impact, using data such as air miles, carbon footprint, viticultural methods and bottle weight. (If a venue has wines not already on the Preferabli database, then the team will track them down and get them tasted – unless, I presume, it’s some kind of crazily rare and expensive wine, in which case a wine recommendation app is probably not what’s required.)
“We didn’t build this platform to replace sommeliers,” says Dillon, “but to extend what they do and to eliminate those things they don’t want to do. We still expect you to say to the sommelier, ‘Preferabli has suggested these three wines, now tell me what you think.’”
Preferabli has already been licensed to retailers and hospitality venues in more than 70 countries, and presented its tech at Harvard, Stanford, Cambridge and Oxford universities. Plans are now underway to roll the software out into other “sensory product” markets, including fragrance and cheese.
Personally, I’m a bit depressed by the idea of a dining room full of guests squinting at glowing screens – but it’s hard not to be inspired by Preferabli’s possibilities. Will it spell the end for the flesh-and-blood sommelier? “I sincerely hope not,” says Dillon. “I don’t believe – at least in our lifetime – that AI can ever replace that human ability to deeply connect with a person; to understand how they’re feeling and how they want to feel.” I must admit, when I hear this, I feel just a bit relieved.
Letter in response to this article:
Talk of Russian drink puns, I’ll have a ‘Eugene one gin’ / From Alastair Conan, London CR5, UK
Comments