Somm.ai is what the name implies. It is a virtual sommelier. That doesn’t mean we try to replace sommeliers. Sommeliers are super important to how we work. We think that sommeliers are the best source of information for wines, even more so than critics. Why? Critics are really good at what they do but they don’t actually have to sell the product, and they don’t have to take it back when people don’t like it.

At the heart of what Somm does is it tries to find the best wines by looking at winelists. A winelist is not an arbitrary list of wines. It was constructed by someone. Not all winelists are created equal. Some winelists are crammed down by the distributor, while others are the result of years and years of collecting, selection and research.

When a wine is on a good wine list, that is an upvote for the wine. When a wine is on a bad list, that’s a downvote for the wine. It’s that simple. (How do we tell when a winelist is good? We’ll save that for another post because it’s super interesting.)

Let’s see some interesting conclusions that can be drawn when using this ranking system. Here is a chart that shows how somms rank wines (y axis) vs. how popular a wine is (x axis). Because these are “rankings”, the lower the value, the better.

So the wines in the bottom right of this chart are the wines that are under-rated, and demand your attention. The ones on the top left might be wines that get too much of your attention. You can see sommeliers clearly have a tilt. That’s fine, they’re the leading experts of wine and they’re telling you implicitly from their winelists what to drink. The most obvious categories are Burgundy, Champagne, and Germany. Rhone, Piedmont, Washington and Oregon might sneak through too. The sommeliers seem to be telling to drink less Bordeaux, Tuscany, Veneto and Mendoza (yes, that’s right, drink less Malbec!).

Obviously these are averages for a region, and within these regions there are specific wines that score really well, and some that score less well. These averages are aggregations of the data we use to recommend to you the best wines to drink (we also use lots of other data too, to account for other desirable qualities, like a good price vs. retail and vs. other restaurants).

How do we know this methodology is working? Well when we backtested our rankings, we found that wines that we ranked higher using wine lists more than 3 years old went up more in price in the last 3 years than the wines we ranked worse, at a statistically significant level. That means our algorithm is able to find “undervalued” wines before the market is able to.

Our ranking system changes a lot right now because our company is so new, and we’re still figuring a lot of stuff out. But you might be curious what some of our favourite wines are - we’ll start posting some our favourite producers soon. When you search our systems, you will see a 👩‍🌾 icon beside any wine produced by a particularly great producer according to this methodology.