Sommelier vs Artificial Intelligence
I'm a huge fan of technology, and the AI craze has fully absorbed me. GPT-4 is my tool of choice, though I’m glad there’s a date on this post because that comment won’t age as well as Pichon Baron ‘19. I think the promise of AI in retail drinks is awesome, and I'm a huge fan of the OpenAI team. Is GPT-4 perfect? Far from it! However, I've yet to meet anyone who has experienced these latest LLMs and hasn't been impressed in some way.
A few months ago, we decided to introduce GPT-4 to our catalog to see what would happen. It was intriguing, prompting us to start building an integration for our homepage. That's when things took an unexpected turn. Immediately, there was a group that loved the information, straightforward recommendations, and entertainment value. But there was also a faction that seemed to think we were breaking an unwritten rule of wine retail: "Thou shalt not use technology to undermine the omniscient sommelier."
To put things in perspective, my opinion is undoubtedly biased. I believe K&L has assembled the most extensive group of genuine wine experts in the country. Combining our current team and alumni, I'm convinced there hasn’t been a larger assembly of wine and spirits expertise in the states. Ever. This might be a bold claim, but it's how I genuinely see it. Our success stems from this expertise, and it fuels the exciting growth plans we'll be announcing soon for the coming years. We've crafted a great business anchored in genuine wine and spirits knowledge, complemented by a dedication to treating our customers well and investing in technology. However, without our expertise in wine and spirits, we'd be lost.
Do I believe a machine learning model can surpass a seasoned expert who delves deep into understanding your palate, preferences, past favorites, and current desires? Currently AI isn’t analyzing wine by pouring it into a glass and letting the model assess its aroma, color, body, or taste from start to finish. That's as romantic as playing a love song on a kazoo. Pinpointing all these nuances and offering the perfect match, paired with a genuine human connection, is where the sommelier or K&L associate will always outshine any machine. But for fun, let’s trigger the haters for a sec.
Do you personally know many wine connoisseurs? Within that circle, how many have in-depth knowledge of every region and varietal, and are always on standby to text you back immediately, any time of day? Can they deftly sift through 10,000 ever-changing options? Imagine a unique bottle, impeccably suited to your palate, that just became available in a distant warehouse but will be gone in a mere 45 minutes. How would they know?
More importantly, let’s get out of the mentality of it being us (the wine pros) versus the machines. That’s ridiculous. We use these tools to improve the customer experience, not to get rid of human connection and relationships but instead to provide something useful and incrementally fun.
Imagine if we could document our team of experts during their staff tastings, capturing all the knowledge they exchange, their opinions and stories. What if we transcribed these discussions and used them to train our models? Suppose we organized a roundtable to discuss a few decades of birth years, debating the best and worst regions and vintages. Or visited one of our direct import producers and recorded their insights to train our model. Sounds intriguing, right?
We possess decades of invaluable data, including hundreds of thousands of multi-paragraph notes and reviews from staff and critics. None of that info is fully integrated into our model yet, but it will be!
Our goal isn't to replace wine experts. Rest assured, industry professionals, your roles remain as crucial as ever. However, these novel tools add a unique layer of enjoyment for the public, hinting at even greater innovations to come. I'm excited to explore their potential.
A special thanks to SF Chronicle's senior editor, Janelle Bitker, for her insightful article this week. Without throwing us under the bus, she noted that our model missed the mark in 3 out of 4 recommendations for her team. I’m glad it got one right! I value your coverage and hope you'd concur that our real strength lies in our staff, not machines. We remain committed to this philosophy. But stay tuned, as the advancements we're planning might just take your breath away. Our journey has just begun!