Wine marketers are betting millions that algorithms will help us find wine we like
Most computer-generated wine recommendations have traditionally been pretty much worthless. There’s this, which we’ve noted before on the blog. But there’s also this, from wine.com, which arrived last month: The email said I would like a $64 bottle of Veuve Clicquot (I think not), as well as a $20 Kim Crawford sauvignon blanc (I think not, again), as well as a $12 California pinot noir and a $12 sweet red blend (I really, really think not).
That’s because the formulas used to put these recommendations together depend on things that aren’t necessarily related to wine – like income, age, and gender — and use past purchase information probably more than they should. Just because I bought wine A doesn’t mean I’m going to like it enough to buy wine B. And AI-generated recommendations have been notoriously lousy at using price to determine what we’ll like.
In addition, most algorithms have not been able to effectively process the wine-related criteria that do matter, like style, alcohol levels, and so forth. And I’m not sure that marketers understand the challenge. One highly-touted AI effort in 2021 said the the goal was “to personalize wine recommendations based on individual preferences, for example: Is the subscriber a red wine or do they prefer white (or maybe a little of both)?” Which seems more of a 1990s approach than something in 2021.
Another, the BottleBird app, asks a series of questions to find out what we like, including “How do you feel about the smell of flowers?” How we answer, says its publicity, will make sure we “never buy a bad bottle of wine again.” So, of course, when I tried the app, I got a 97 percent match with two Fre non-alcoholic reds and two buttery California chardonnays. Hardly the WC’s kind of wines, are they?
The process is so difficult, in fact, that a friend of mine who has written extensively about the subject once told me the only way AI wine recommendations could be successful is if they included information from human tasting panels. Which hardly sounds very AI.
So why have wine marketers spent millions of dollars trying to find a way to teach machines to make wine recommendations? My hunch is that it’s not so much about helping us find wine we like as it is selling us wine. Which, of course, is not necessarily a bad thing – if they work. But when they recommend wines I wouldn’t buy even if they were free, what’s the point?