Can an Algorithm Recommend Wine Better Than a Sommelier?

1. What problem is AI actually solving in wine?
2. Verve Wine — where the conversation started
3. How Vivino’s Match algorithm works
4. Tastry — the chemistry-first approach
5. What algorithms do better than sommeliers — and what they don’t
6. Frequently asked questions
7. AI wine recommendation tools — quick reference

What Problem Is AI Actually Solving in Wine?

The wine discovery problem is a scale problem before it’s anything else. There are approximately 160,000 commercially available wines in the United States. A given wine shop might stock 500 to 2,000 of them. A knowledgeable retail buyer will have tasted a meaningful fraction of their selection. But no individual human expert has tasted all 160,000, and no consumer has the time or money to explore the category blindly enough to develop real confidence without guidance.

The result is a familiar retail dynamic: most wine buyers default to a handful of grapes, regions, or price points they already know, while the vast middle of the wine world — the interesting, well-made, fairly priced bottle from a variety or producer they’ve never encountered — goes unpurchased because it feels like a gamble. This is the gap that wine recommendation technology is designed to close, and it’s a genuine consumer problem rather than a marketing framing.

Verve Wine — Where the Conversation Started

In late 2016, Dustin Wilson — one of roughly 230 master sommeliers in the world at the time, best known for his role as wine director at Eleven Madison Park and as a central figure in the documentary Somm — left the restaurant world to co-found Verve Wine with veteran wine merchant Derrick Mize. The original concept was straightforward and somewhat provocative: use an algorithm to personalize wine recommendations, starting with an in-store tasting Q&A that fed customer preferences into a custom software program, and then use those taste profiles to power an e-commerce platform that could ship wines tailored to the individual.

Wilson’s argument was clean: “Just because an employee thinks a wine is awesome doesn’t mean a customer is going to like it.” He was identifying the fundamental misalignment between how wine is sold — through expert recommendation projected onto the consumer — and how it’s actually experienced, which is personal, variable, and deeply individual. The algorithm, in his framing, wasn’t replacing the sommelier’s knowledge. It was replacing the sommelier’s assumption that their taste was a reliable proxy for the customer’s.

A decade later, Verve Wine has grown into a three-city operation with locations in New York (Tribeca), San Francisco (Pacific Heights), and Chicago, alongside a full e-commerce platform. The algorithmic personalization that defined the original concept is now one component of a broader identity: a shop focused on small, family-owned wineries that show “great respect to their lands and winemaking traditions,” with Dustin Wilson’s monthly picks as a curated overlay on top of the algorithmic recommendations. Wilson himself has expanded into a Tribeca restaurant (One White Street) and a hospitality consultancy (Apres Cru). The bet on the algorithm has paid off — not by replacing human curation, but by augmenting it.

How Vivino’s Match Algorithm Works

Vivino is the world’s largest wine marketplace by community size, with over 60 million users who have collectively generated more than 200 million wine ratings and reviews. That scale is the foundation of its recommendation engine — but the way the engine uses that data is more sophisticated than simple crowd-sourcing.

In 2021 Vivino launched “Match for You,” a personalized match score system built on machine learning that generates a unique recommendation score for every one of the approximately 13 million wines in its database — essentially every commercially available wine in the world — specifically for each individual user. The mechanics: once you’ve rated at least five wines, the algorithm begins mapping your taste profile against the community data. It identifies which flavor attributes, structural characteristics, and regional patterns correlate with your positive ratings, and which correlate with your negative ones, and then generates a match percentage for wines you haven’t tried yet. A 70-100% match is flagged as a “Great Match” — wines the algorithm predicts you’ll rate four stars or above.

Vivino’s approach is fundamentally social and historical: it learns from what millions of people have rated and bought, then uses those patterns to predict what an individual with your rating history is likely to enjoy. The system doesn’t know anything about the actual chemistry of the wine — it knows what people said about it and whether those people’s other ratings correlate with yours. This makes it powerful for wines with sufficient rating history and less reliable for obscure producers with few reviews.

Tastry — The Chemistry-First Approach

Tastry takes a fundamentally different approach to the same problem. Rather than starting with user ratings and working backwards to wine recommendations, Tastry starts with the molecular chemistry of the wine itself and works forward to predict which individual palates will respond to it.

Founded by Katerina Axelsson, a chemist who paid her way through Cal Poly San Luis Obispo working in the wine industry, Tastry submits wine samples to its TTB-certified laboratory and analyzes them for thousands of chemical compounds — pulling what Axelsson describes as over one million data points from a single bottle. The chemistry is then mapped to a database of consumer palate profiles, built from a brief quiz that asks respondents about sensory preferences across everyday flavor and smell experiences (coffee, tobacco, vinegar, and others) rather than wine-specific vocabulary. The AI matches the chemical fingerprint of a wine against individual palate profiles to predict preference scores.

The stated accuracy: 93% — meaning when Tastry predicts a high match between a wine’s chemistry and a consumer’s palate, that consumer rates the wine highly approximately 93% of the time. Tastry also claims it can predict a wine’s Vivino rating score with 93% accuracy based purely on the chemistry analysis, before any human has reviewed it.

What Algorithms Do Better Than Sommeliers — And What They Don’t

The honest answer to “can an algorithm recommend wine better than a sommelier?” is: it depends entirely on what you’re asking the recommendation to do.

Algorithms are genuinely superior in several specific contexts. They don’t have bad days. They don’t get tired at the end of a service. They don’t project their own preferences onto a customer who said “I like bold reds” and then recommend the bottle they’re currently enthusiastic about instead of the one that fits the stated profile. They can process pattern recognition across millions of data points — identifying that a customer who loves the tannic structure of Barolo tends to also respond well to aged Rioja Gran Reserva — in ways that no individual sommelier’s memory can replicate. And they scale: a recommendation engine serves the same customer in a supermarket aisle at 10pm as it does in a wine shop at noon on Saturday.

Where algorithms currently fall short is precisely in what makes a great sommelier exceptional rather than merely good. An experienced sommelier reads a table — the occasion, the mood, the dynamics between people, whether someone is celebrating or stressed or adventurous or in need of comfort. They ask follow-up questions that a quiz doesn’t ask. They know that the wine someone says they want and the wine that will actually make the evening special are sometimes different. They carry the cultural and historical context of wine in a way that a recommendation engine optimizing for a predicted rating score doesn’t capture.

The most thoughtful practitioners in the space — including Dustin Wilson himself, whose decade of operating Verve Wine has given him a real-world view — have landed in essentially the same place: AI and human expertise are more complementary than competitive. Tastry has described the relationship explicitly: “Data from a company like Tastry adds a layer of expertise and knowledge to the sommelier, which can result in better matches for the consumer.” The goal isn’t replacement. It’s augmentation — and particularly, it’s democratization. The consumer who doesn’t have access to a world-class sommelier benefits most from a well-built recommendation algorithm. The one sitting across the table from a great sommelier who knows their preferences already benefits from both.

Frequently Asked Questions About AI Wine Recommendation

🍷 AI Wine Recommendation Tools — Quick Reference

The major platforms, how they work, and who they’re best for