Our wine recommendation engine

2022-08-26

RENDER HTML
# **Our wine recommendation engine** **We have been hard at work tuning our precision wine recommendation engine. Our vision is a world where you never again open a bottle of wine and wish that you hadn't bought that bottle. A world where you only ever drink wine that is top wine based on your own unique palate.** Today's wine world is far from that vision. Rather, most wine drinkers resort to one of the below six strategies to select their wines: - **Price Signaling.** Buying expensive bottles of wine feeling that there must be signaling value in the price point. That is, assuming that "quality costs". - **Pretty Label.** Picking a label that looks nice off the shelf within the intended price range. - **Recommendation by a friend.** Buying wine based on the recommendations of friends only to figure out that they often have very different wine preferences. - **Wine Critic Scores.** Following the recommendations of wine Demi Gods like Robert Parker, Jeb Dunnuck and Jancis Robinson. - **Sommelier Recommendations.** Listening to the advise of a sommelier. Someone whom the wine drinker has often met for the first time and who knows next to nothing about their customers' taste preferences. - **Vivino Scores.** Selecting wines based on the average scores from the crowd intelligence of sites like Vivino or Cellar Tracker. The above has frustrated us for years and we have been surprised how the wine industry has been so slow to adopt new technology. In fact, looking at current study programs at MIT and at Stanford, the hottest new fields of study are no longer about "software" or about "hardware". Students instead flock to programs integrating data science / artificial intelligence with chemistry and biology. Vinod Khosla, of Khosla Ventures, has said that he expects the breakthroughs over the next decade in these areas to outpace the past 100 years once data is unleashed for analysis, prediction and to develop new products and treatments. The same forces can be harnessed also to develop a precision recommendation engine for wine. Our chosen approach is based on three pillars: - **Taste Preference Questionnaire.** A taste preference questionnaire as a tool intended to onboard new wine club members and to get a first indication of what type of wines they might prefer. - **Your Wine Scores.** Those of you who have come to our free tasting rooms know that we ask you to rate the wines. This is a bit like Facebook and Google giving you a service for free in return for your data. The data feedback loop has given us a lot of data so far to backtrain our data model. - **Analytical Chemistry.** We fingerprint every bottle of wine in the lab to obtain more than 200,000 datapoints. And the secret sauce is not in measuring the presence and concentration of a particular compound or characteristic of the wine but rather how these affect each other as the relative concentrations change. That is, the same compound may produce very different sensory effects based on what other compounds are present and at what concentrations. A few photos below from the lab analysis work. The goal of our R&D efforts is to be able to predict whether you will like a bottle of wine before you have even tasted it. And then we can completely wipe out the brand premium that many of you have been paying to producers over the years. Instead, you can feel confident being recommended private label wines at much lower price points. And in so doing, we will disrupt the wine industry and bring the industry into the 21st century. subimage1 Most of the lab results are difficult to visualize as the data feeds a long and complex algorithm. However, it is also possible to produce spider diagrams like the below to compare wines. During the weeks and months to come, we will be sharing more and more insights with you so that you get a feel for how our approach is evolving.