Accelerating Wine Sales w/ Chemistry & AI w/ Kat Axelsson & Charles Slocum, Tastry

Hey listeners, just want to let you know that we've released the latest compilation of show notes in book form, covering episodes from 2022 to 2023. It's full of insights on sustainability, marketing, and even has a few celebrity sightings. Pick what up on Amazon and support the show. Welcome to X-Shadow. The podcast that navigates the business of wine with unique perspectives and insights with your host Robert Vernick and Peter Young. Welcome to this episode of X-Shadow. Today we are talking with Katarina, actress and founder and CEO of Tastery and Charles Locom, the Chief Business Officer of Tastery. We're talking about selling wine using chemistry and AI. Welcome to the show. Thank you so much for having us back. Yeah, welcome back to you, Katarina. You're on November 2023. Herbity who wants the detailed information about her background and Tastery's founding. Please listen to episode 157 and Charles, welcome. It's your first time. Thank you so much. I appreciate it. We already got Katarina's background in the previous episode that she was on. So Charles, maybe you could give us a brief overview of your background and you can give an introduction. Sure. So as Chief Business Officer at Tastery, I largely manage a number of scientific teams at Tastery. My job is to take the innovations coming out of those teams and commercialize those products to help our clients generate revenue and increase acceleration. My background largely has nothing to do with wine. I think like a lot of people on the industry, I spent my whole life in security, electronic and physical security industries from low level working with banks and police departments, law enforcement, the type of thing up to DOD. I'm working with various militaries around the world. That's kind of my background. And I fell into Tastery on a lot. Katarina and was helping with some business strategies very early on just over time folded more and more into the company until it became a full time gig. And here I am. I don't even know how many years it's been, Kat, say, five years, six years. Something like that. Back when it was a project and there were three people in the company and a bunch of scientists on the side. That's when I started. And along those lines, Kat, maybe for our listeners context, you could give a brief overview of what Tastery is and what it does. Well, our company tagline is we taught a computer how to taste. We have these two unique data sets we developed in house, one on chemistry and one on consumer palette preferences. We're able to predict the outcome of chemical interactions and use that data to see what wines or products should exist that don't or how to have the wine you've already made suit the market you're trying to sell to or make improvements on the wine you're trying to make. How we use that information varies greatly depending on who we're working with across the supply chain. So we work with a lot of wineries in production. We also work with sales and marketing teams and retailers and distributors. But the short version is is everyone is using this to predict and react to the every evolving consumer preferences in the wine industry. When we last spoke, Tastery was two years into commercialization after four years of R&D. What's happened in the last year for Tastery? The first two years, we very much focused on establishing trust and showing our efficacy with wine makers and production. And I think we did a really good job of that. We have a lot of endorsements from some very prolific brands and wine makers and we're somewhat integrated into the production process. So since we accomplished that milestone this year was all about focusing on the other end of the supply chain and making sure we could create a system or an ecosystem that could help sell all this amazing wine that's being made. So we've really refocused our attention to importers, distributors small and large and sales teams so that we can get this data and insight into the hands of buyers. We had more of a push focus, now we have a poll focus, right? Let's sell this wine. I think that plays in well with what we want to talk about today, which is one of the key things you've recently launched that connects the chemical profiling of wine with your consumer taste preference database, which is what you call the wine and consumer insights reporter. Maybe we'll call it WCI for sure that gets to be too long as we go on here. And for our listeners, Tastry has provided an example report of the WCI, the 2023 Hidden Sea Rose, which we can put a link to in our show notes and discuss here to give it a little bit more context. But first, I know Charles, you've worked a lot on this to help to create this as part of your role. Can you give us a brief overview of what the vision for the wine and consumer insights report is? Sure. So the wine and consumer insights report is just one small piece of a larger program called sales accelerator. About a couple of years ago, our winery clients started coming to us and saying, hey, we have all this amazing data and insight. We understand who's going to like it, how much they're going to like the wine. And we can identify that. So the question was, can we take the data we already have and use that to help us market and sell the wine, get it to the consumers who are going to love it. And most of our clients, I mean, we work with 23 of the 25 largest wineries, but we also work with hundreds of very small wineries and they have a very hard time getting visibility in the market. And they make amazing wine. If they get placements, it's hard to keep those placements because they just don't have the marketing dollars to expose it once it's there. The idea behind sales accelerator is to provide a platform where the wineries can enter in various metadata on their wine. And of course, we have visibility on the chemistry and the performance of the wine against consumer preferences. And then all of that is aggregated with our AI and then distributed throughout the entire supply chain. So it goes to distributors, it goes to retailers, it goes to on the ground sales people all the way into the hands of maybe a server in a restaurant, maybe someone who's stocking shelves in a grocery store. And so the idea is to shepherd the wine through the entire system, provide a maximum visibility and make sure that the wine is being characterized and sold properly and provide that material so that that can happen. That's the intention of sales accelerator and WCI is one piece of that ecosystem. Before we dive deep into the WCI, can you tell us what the other pieces of the sales accelerator ecosystem is? Sure, sales accelerator were integrated with distributors like R and D.C. And so R and D.C. utilizes our data online to curate recommendations to buyers based upon what they're trying to accomplish. And then also on the ground sales people in R and D.C. are using the WCI's to help them sell those wines into on-prem or off-prem placements. In addition to that, we provide the same data can be used to train the sales staff in-house, right, whether it's a server or as like I said. And so these same data are then curated for that. So all of that distribution, retail, sales people, all of this is part of sales accelerator. So as an example, if tomorrow one of our clients says, "Hey, we just got NAPA green certified," that can be important to a lot of buyers. If they enter that into our system, five minutes later the sales person who's making a sales pitch in Tennessee is going to see that, "Hey, they're NAPA green certified." So it creates an instant path between the winery and everybody in the supply chain. So that everybody all the way down to, as I said, all the way down to the server just before the consumer can see all this data. And then sometimes it's consumer facing as well. So we're doing integration with some big box. We're actually providing recommenders and retail. And this data is also available there to the consumer. So if the consumer says, "Hey, is hidden C certified to use this example?" It will know immediately that data will be real time. Just so I understand that how are, you know, like a server or a distributor wrap or whatnot seeing this data? Are they logging into TASRI or you're integrated into their own platforms and then it shows up? TASRI AI can be integrated into the actual platforms like ER and DC. Also, we have separate applications. So we have something called the Sales Accelerator Portal. It's a little app on your phone that a sales person can go to and say, "Hey, I'm walking into this on-prem location. I know my buyer. My buyer really cares about minority-owned wineries. If this is a minority-owned winery, they're going to care about that. That's going to help me make this sale." They can in 10 seconds on this app. They can look that up and find out any relevant information very quickly. So imagine you've got dozens of pages of metadata on a winery. They can search and find that instantly, the 60 seconds before they're walking in to make that sale. To get those bullet points, they're going to help them get that placement. That's constrained within their book, I guess, right? So within the distributors book and then whatever it is, which specific lines that they're trying to sell. That's exactly right. Whatever is in their book, we associate our data to that. Then we associate their pricing data. All of this comes together. So on the application that they might use, they're all options. Some sales people will use the WCI's and not the app. Some will use the app and not the WCI's. We're not trying to create or one application fits all. We're trying to cater to whoever is trying to sell this wine and provide them the data that they need at the time, not to shoehorn them into something that we're creating. So the WCI, as I said, is just one tool they might use. So the WCI, Katarina, provided as an example to your listeners, might also have pricing on it. So this could be a leave behind. You make the call, "Hey, this is the wine. These are the selling points." And then on there can be the pricing, the discount pricing. So even a leave behind that the sales people can use if that's helpful. Some people incorporate pricing, some people don't. It's designed to be very flexible to give their sales person what they need. Just so I'm clear, it's all the same information but packaged in different ways. One is an AI application so people can search versus the WCI as a specific wine. But the base data that's in the WCI is the same as the data that's feeding the app. The data that's in the WCI is a highly curated, very small portion of the data. The data that's available to the salespeople, so we also have a chat functionality. So the salesperson can just say, "Tell me about the sustainability of this winery." And the AI will generate that data. And it will generate it in a way that they can regurgitate that quickly. So it's not going to give them two paragraphs that they have to read, figure out, figure out how to word that. It's going to provide it to them in a form of, "If I'm talking to a buyer, what would I say about the sustainability?" The idea is to provide it as regurgitatable as possible so that they can move more quickly and communicate clearly in the way that the winery wants them to communicate that efficiently. I was asking this also in the context of, as we dive into the WCI and all its components, that that will have insight for listeners into what else they could do and think about in terms of using the sales accelerator. Absolutely. I think the WCI is just a little peak at the surface level, right? It's designed to be a very short document. It's very focused. So we can talk about those components and why every component on their matters, at least to our salespeople. This is again driven by them, not our salespeople, but salespeople out in the market. But yeah, it's just the surface of it. So as an example, just to throw something out there, we have wineries who will say, "Hey, I'm trying to pitch to X-crochet restore or X-restaurant chain. The AI can look at all of their data, all of their metadata that they provide." And then the AI will take that, look at the assortment of that retailer. And maybe it's a restaurant. It'll also look at the menu items. It'll look at what they're currently offering, how those menu items pair with those food items. And then it will identify what are the key arguments? Why should this wine be in that assortment? Is it going to pair better? Is it going to create a unique pairings that they can't currently support? Is it less expensive? Whatever it is? And then it's also going to pull in data like we have thousands of pages of market trend data. The AI can, for the sales team, in one or two seconds, search hundreds of documents and pull out those market trend components that matter to that buyer. An example would be, this actually is a real example. A client was going in, this large chain normally carries wines that retail for about $20 in retail. Their wine was $30. And the AI found that there's actually a trend in that type of restaurant with premiumized wine that it's actually increasing in sales, whereas most of the markets decreasing. So these $30 premium wines in given retail environments, restaurant environments, which this one was, that it actually has a 30-some-percent increase in growth. And so it's able to find that data and inject that into the sales pitch, which would take a sales person days or weeks if they could even find it trying to go through that. So we consider all of this 95% marketing. It's not going to give you everything. We're not trying to do that. What we're trying to do is let you filter through all the data, put the best argument together, 95% of it, and then the marketing sales team's taken from there. >> Just so I understand here, is the your AI just scraping the internet for all that data or are you plugged into other systems like some AI or wine searcher or other things which don't even have things like US grocery lists? >> Super good question. We don't scrape the internet. One of the prideful points that we take is that our AI will never hallucinate. It can't do that. We cannot ever allow it to make something up. So the market trend data, we feed it directly. This is data that is a reliable source. We feed those data into the AI. This is what you're allowed to look at. The metadata from the winery, what the winery says about their sustainability, about the critic scores they've gotten, whatever it is, all of that data is controlled by the winery. And that's the only thing that AI is allowed to look at. And then the next piece is the hard chemistry preference data that comes out of our ML. So the AI is allowed to look at those three things and those three things only. So through that means, we never have a hallucination situation like you would get with a chat GPT where it just starts making up stuff that is at best off target, at worst, just completely wrong. By controlling the data it's allowed to look at, we're actually using a combination of small language models where we're very tightly curating what it can look at and a large language model which allows it to be very creative in the way it presents the data, but all of the data is controlled. So no, we don't scrape the internet. So then do you have partnerships with other data providers that have a lot of that database and it's cleaned it up? Right now we are doing all of that in-house, but we are looking at partnerships. We have a very clear focus on trying to help wineries generate revenue and gain traction in the market. We don't want to become a data company necessarily. So to the degree we can partner, that's great. Our challenge right now is the quality of the data. So we're very anal about the quality of the data and if it doesn't meet our standards then we won't use it. That makes it a little bit harder for us, but we've created all of our solutions today and if we have to keep going down that road then we'll do that. But we are looking at some partnerships that look promising where it looks like they'll be able to meet the quality requirement. Got it. And so now focusing on the WCI. Who exactly is a WCI for and how do you see them using it? WCI is meant to address two particular readers. The first page of the WCI is intended to address the category buyer for a restaurant, retail. The second page is for that category buyer to provide this data, the second page data to their sales staff on the floor in aisle, the servers to help educate them. We can dive into it but that's the quick answer. It's meant to deal with those two individuals particularly, those two archetypes. And the customer who would buy the WCI and give it to those customers then would be the winery or distributor? Yeah, so existing winery clients of Tastry that are already having their winery analyzed for other reasons. This is all free. The whole sales accelerator system is free. The entire intent was to take data they've already paid for and how can we leverage that to drive sales and adoption of the product? Let's jump into the WCI and kind of get into its components in the different sections. So just for our listeners who can't probably, if they're driving especially or not looking at the document that's going to be linked in the show notes, essentially two slides that are fairly data rich showing the name of the wine, having some notes, having a score category on the first page, flavor profile, and then some retail talking points and food pairings and flavor profile on the second page. Where is the best place to start? I think we could start on the top left and just work through it. Every piece of this is highly curated. If you start on the top left we see a bottle shot. Everybody has a bottle shot. We found is that there's a huge correlation between really seeing the image of the bottle that creates higher retention for the reader. So usually bottle shots, we have one on this image, which is the full bottle, but we have a zoom in that the AI does on the actual label. And we found that that actually helps retention for people in store that might have to be learning five or ten or twenty wines this week or a server. It makes it easier for them to remember it. So we provide this bottle shot with a closeup of the label. Below that we have of course the name of the wine and we have the variety on the appellation. We're just listing that data as useful to the category buyer. Below that we have the wine category. So the AI curates a wine category for this wine. In this case it's a Rose 10 to 20 dollars. That category is the category against which it's creating a score, which we can talk about I suppose. So to the right of this in the center of this first page is a case tree category score. In this case it's a 129. The category score came out of retail when we're working with retailers going back quite a time now. What retailers were interested in when we were working with them as our clients or when we do work with them as our clients is they want to risk mitigate what they're putting on the show. So they try a bunch of wines. They might taste a bunch of wines. They are painfully aware that they're tasting group of one to three to five people is not representative of the American public and it's a huge risk. They might all love that wine and it just might not be right. Not that it's a bad wine. It might be the wrong environment. I like to say if you put a screaming eagle in a safe way at $25 it's going to fail horribly. It's just a wrong environment for that flavor profile for that wine regardless of the price. So it's not that the wine is bad but are they choosing the right wine for what they're trying to do. So largely the pastry category score is zero to 200 points and 100 basically means it's an average wine for that category. If you are at say around 100 or higher you're outperforming half the wines in that category. That's what matters to a retailer. They may or may not care about the 129 versus 139. Generally they're just looking to mitigate that okay I'm not buying the worst wine on the planet for this category am I that's what they're looking for. So we're trying to give the category buyer confidence. The retailers that we work with don't actually see it this way but this is where this came from. They do get the same sort of data in a different visual. So the idea was we were asked by salespeople and distributors to create a score. They said look we need something that's very simple. Someone can look at and go okay I'm confident I'm going to buy this and we were very hesitant to do that but that's what they wanted and so we thought well the best way to do that is to compare the wine against its category. So what's interesting about our category score is not a credit score on our WCI's if there are credit scores for the wine those will be presented as well. If Dunic has something to say or Wong has something to say or Wine enthusiast has something to say we definitely want to put that out there the intent of this is to help sell the wine but where a critic might be interested in say how well does it represent the tarwhar or how well does it represent the wine maker those types of characteristics. This score is not interested in that this score is interested in is this going to perform well against its category for consumers who might be buying this wine. And so if we could show two side by side I think cupcake Pinot Grigio this year I think because of 181, very high performing wine. Last year it was a 190 something. And so the AI is saying, look, people are going to love this cupcake Pino Grigio. And I think it was the number one or number two soap wine last year and white wine. At the same time, I can put like Chateau Lagrange up there, which is a 182. And in its category, in a very high-end Bordeaux, it also was going to perform well. So it's all about how is this going to perform with the consumers who are buying that wine in your retail establishment? Not what a critic does, which is how well does this represent the terroir of the variety. That's the big difference in our score. It's about giving the buyer confidence to take a risk on this, generally speaking, right? It's going to be an unknown label. They're not familiar with it. And it's risky for them to take a chance on a wine. They're much more likely to just stick with whatever's on the shelf. We have to give them confidence. This long tail item is going to perform well. People are going to love it. Try this smaller, less common label. That's the intent. The category of your pastry category scores an important one, what category they're in. Who determines the category that this wine is scored against? Is that pastry or is that the winery? It's pastry. It's the AI. So what the AI is doing is it's looking to put using an objective category, what's the best score we can put on the wine? The idea is to substantiate the sale of the wine, but with an objective limitations. So as an example, I have a client who makes a Napa Cab in Napa Valley, AVA. Last time I looked at it, they're about 82. They have an 82. So they're below average. That's a tough category, though, right? On the other hand, if we pull them out from Napa Valley to Napa County, well, now they're like a 143. So we would change the category to say Napa County and say, look, within Napa County, it performs this well. Whereas Napa Valley might be tighter. So we are curating a little bit. Doesn't that impact the retail price or is the price given and fixed from the winery? Because when you say Napa Valley versus Napa County, to me, that implies a much different comparison set in terms of price as well. So it doesn't necessarily help you. If your price is too high and you're all of a sudden Napa County, so it's also looking at the price. And all of that price, all of these factors go into the category. So there are wines that just are not going to get great scores. We're not going to whitewash that. It is what it is. But we do try the best we can to say, look within this, this wine is going to do well. That's how the categories are created. That's part of it. The other part of it is that sometimes we have to open the category up because there aren't enough wines. So we run into challenges there, right? So as an example, we have a client now in New Jersey. They have a San Marco. It's the first bottling of a brand new varietal in the US. There is no comp set for that wine. There are also wines that are made. We might have a gamma and a given county. There's no other gamma. Well, we can't give you a score until we get a representative set of data. So we might have to say, well, we can't use even your county. We're going to compare your gamma against gamma is in California because we just don't have a big enough competitive set to compare you. It's always constantly looking at all of those factors and creating a viable set. Because in the end, we want the retailer to have confidence that, yes, this is a viable set of wine. And in this category, it's going to perform well. And so it's very dynamic. But if I'm the winery, right? And I've set my price. Use your example from earlier. I'm screaming you go, right? But for some reason, I set my price at $25. Then you say, I have a really low score in this category. It should be in the $1,000 price point category, right? Would you create a WCI at $25 and $1 at $1,000? No. Well, that's never happened. And I would say our first answer would be no. I mean, I would always listen to our clients. The winery doesn't choose the price of the wine either. What we look at is what is the wine actually selling for in the market? So we don't want someone to say, oh, my wine is $10. They're actually selling it for $30 or vice versa, right? So we're actually looking at actual sales data and understanding what the actual price of that wine is in market. And that's what we use for the price. That's why we don't list the price on this sheet. The actual price of the wine we don't list. But we know what it's selling for and we drop it within that range. But what it's selling for in the market should be reflective of what their suggested retail price is for the most part. Generally, I would say that's true. But sometimes it works to their advantage. I'll give you an example. We had a client who was building a proposal using Taster AI to build a proposal. And when the AI did the WCI and it was building this pitch to a retailer, although it's a $20 wine and the pitch is to put it in the $20 wine kind of category that kind of space in this brick and mortar retailer, the AI actually saw that it was outperforming in $17 to $27 category. So one of the arguments that made it in the pitch was that actually this wine outperforms up to $27. So there's actually potential extra margin here. You could charge more for this, increase margin, and still make your clients happy. And so that was compelling for the winery and the buyer to see that, hey, we can actually maybe raise the price on this wine because of the way consumers are going to perceive it. And so with the category, how is the score calculated? So it's very simple. It's a percentile rank, right? So what it does is it looks at all the wines in that. And 50% are going to perform above. And 50% are going to perform below. So if we're looking at rosés, $10 to $20 in the US, this wine is performing 29% above the average. And you mentioned that some categories might be too small because they're unique and whatnot. How many wines are actually in the database that you're comparing against? Tens of thousands. But I mean, there are 160,000 labels just in the US. So tens of thousands isn't representative, right? So what we're doing right now, which is interesting, it hasn't come out yet. What we've seen is a problem. So as we work with say, chateaus and very high end wineries, they really don't have comp sets, or they don't have sufficient numbers of competitors to really give them a score. So right now, we're developing a variation of the WCI, which is catered towards rarer and more unique wines because we can't create comp sets like this. And so that's something we're working on now and we'll have something, I think, in the next week or two. So we've had some amazing clients with amazing wines, but there just aren't enough competitors to actually compare them to somebody. We don't want to take three or five competitors and call that a representative set. It's not statistically reasonable. So we have to come up with a different solution. Just a quick question on the 200 point scale, which obviously obviously skates it from the confusion with the 100 point. That's precisely why we did it. So it didn't get confused with the critics score. Just so I understand the math correctly, it's a distribution curve where all the wines in that category clawed anywhere from 0 to 200. If it's 129, this rose, does that mean it's 29% better than leading or does that mean it's only half of that because it's a 200 point scale doubling the range? - It's outperforming by 29% the average. So what we're doing is it's not actually a distribution, it's a rankings, right? So we're looking at how it performs in a rank order. Essentially that's a form of distribution for sure. It's outperforming by 29%. If we're looking at this here, so it's outperforming the average by 29%. - Okay. - That's accurate. - And so for wines that have a low score, 'cause that means that 50% of the wines are below 100. - Correct. - If they get a low score based on that category, what should a winery do to change their score or if they have a really high score, what do they do? Like I'm just curious, how do the wineries use that information if they score low or score high? - Sure, well if they score high, they're happy. That's a good thing. If they score low, like I said, the AI is trying to curate. So for instance, we have clients that well half the people in Napa Valley are underperforming the average, right? That's by definition. So the AI will pull back and say, well, okay, it's not really performing there, but maybe we go to Napa County, maybe we know to go to North Coast, you know, the type of thing. And sometimes it will back up to a point where it's not reasonable. So for instance, it might come back and say, well, against California Cabernet's, you're doing well. And they might say, yeah, but I don't want to even say that. When that happens, just like a critic score, they don't have to expose this score. They can opt to not use the WCI. So it's always up to them if they want to try to use it to market the wine or not. Just like a critic score, right? The wine enthusiast gives you 70 points. You're just going to drop that and say, well, okay, we'll have to work better in the next year, that type of thing. The unique thing about Tastry, I think, is that when we do get a situation like that, the benefit to Tastry is that the wine maker can say, or business can say, okay, what do we have to do to increase this? And we can actually provide the wine making team visibility on that. And then they can decide if they want to do that. Sometimes they'll decide to lower the price or raise the price. I mean, we very often find that a standard feature recently in the past 10 or 15 years, I think is that wine makers will say, I'm putting a $25 wine in a $15 bottle and I don't understand why I'm not getting better performance. And that can sometimes be good, but sometimes it can be a real detriment. Sometimes a $25 flavor profile to someone who's trying to buy a $15 wine is actually not what they're looking for. And so we've suggested wine makers lower the price and we've seen increase in velocity and we've actually suggested they increase the price. Both ways and seen increase in velocity. So you've got to make sure what the consumer is expecting is what you're giving them. And if you kind of mix match pricing, going back to the screaming eagle example, it generally doesn't work well. Sometimes it does, but generally doesn't work well. - What is the size of the data set you needed or to get into that statistically relevant number of wines for that category? Because that's the one part to understand like, hey, if you put me into California Cabernet and I have higher aspirations for my wine and I want to be more focused, how many data points do you need in order to the WCI kind of rank it in that category? - So I can tell you that it depends on all the data. So there isn't a set number. I can tell you that there's never less than 15. 15 is the absolute minimum for a very small AVA or something. If we have 15 wines and a little tiny AVA that are all the same variety, same price point, then we can say, okay, that's a reasonable set. And then depending on other factors, it might increase that number from there. So 15 is the absolute minimum. And then sometimes it will say, no, I need to help. of 50 of these, or I need to have 75 of these, depending on other data. There's a model that we use. It is kind of a first principle driven model, but it's determining the statistical significance of what we're saying. And the absolute minimum we go down to is 15, and that's a problem for a lot of wines because there aren't 15 of a lot of wines and a lot of varietals and given ventages, right? So it can be challenging. - How do you see the concept of terroir playing with the score? - Because theoretically, if a wine truly expresses its terroir very well, it'll have a unique profile and taste to it, which some people might value a lot and others may not. So how does that correlate with score or do more unique wines score worse or better? - More unique wines do not necessarily score works or better. So terroir as a representation of the terroir is not something that this score is identifying. That is something we do on the wine making side. So wine makers can evaluate their wine from that perspective. How representative of this, say, a classic, you know, and why from Southern Oregon somewhere, let's say. But this score does not take that into account. This score is purely focused on our consumers going to like this wine. Now, does it come into play though? Yes, even though we're not specifically doing that, what we will see is that, although this is not specifically driven, what the AI will do is generally speaking again, these are all generalities, that the lower the price, the less the terroir matters, the higher the price, the more it matters. So if I'm getting a low-die cab at $10, whether it tastes like a low-die cab is almost irrelevant to the buyer. If I have a low-die cab at $35, then it becomes more relevant. So because of the way it understands wine, that is incorporated into it, but it's not a specific set of rules we've built for this. - Yeah, it's going to looking for natural groupings that are relevant for that data set that you're curating essentially. - That's right. And if you're looking at a $200 boarder from France, it's looking for a very specific style. And if you're off style, it's going to ding you really hard. It's going to presume that people are not going to like this wine. But like I said, it's rocking that from its understanding that we do on the back end with winers. It's not a specific driver that we have built into the WCI, but it does incorporate that to that degree. - So there are some wines that, you know, in your example, when you said $200 boarder, that is not in that same style that popped into my head, 'cause I think of a wine like Armco Brian, which is using some rone techniques of maceration of the whole clusters, the stems into the wine to create a somewhat different profile. I think that comes out of the rest of the boarder and certainly from Pesock. So with the score, not score well, even though it had very high critical reception and critic scores are really, really high. - That's a good question. So to be honest with you, when it comes to the black box part of this and these very rare wines, we have seen very rare wine that the AI looks at and says, oh, this is kind of fascinating, is kind of the way that I interpret its result because it doesn't give it a bad score, even though it's sticking outside. So for instance, the Laurent Perrier Ultra Brute with zero dosage, right? It sees that is very different, but it doesn't characterize that as a bad wine. So exactly how it's interpreting that, again, the AI is not trained to say, oh, this chemistry marker, these chemistry markers are off by XML, therefore it's a problem. There's nothing mechanical like that where we feed the AI, but the AI is looking at is the balance of this going to be pleasing to consumers. And so if it generally comes from the area, exactly how far you'd have to be off before it might say this is not of that style, that's kind of black box gray area. It is something that we're looking at and it's something that right now we're working on talking about these rare wines. When we were looking at that San Marco, how do you evaluate the San Marco? When it's the first time it's been bottled in the US, the AI looks at it and they can tell you how many people are gonna like it, but whether or not it's on style or not, is a challenge. And so I think although we haven't seen any huge issues with it, that's why we're creating this other tool that is gonna look at unique wines differently, just so that we do separate that out so we can keep track of what's going on there. So we don't end up with something with A gunner face somewhere where some amazing wine is poorly rated or some damaged wine is highly rated. We haven't seen it. I don't expect it's a huge threat, but it's something we have to always pay attention to. - On the left-hand side of the WCI, there's also T-Stream Notes, which kind of read like a background of a bottle by a little bit more in depth, pulling in some of the data from the report. Can you guys explain to us a little bit more about how those notes are created? Is that marketing speakers that AI generated from your portfolio? - It's 100% AI generated from first person from T-Stream. We've asked the AI to look at the performance of the wine, specifically the T-Stream category score, and then look at the flavor profile and what the AI is doing is predicting how the average consumer will perceive the wine, the aggregate of consumers, right? So when the AI predicts how Catarina will perceive this wine and how I will perceive this wine, it would actually write different tasting notes. They vary depending on who's drinking it. So what it does is it says, well, if I look at the aggregate of consumers say in the US, what are the majority of people going to be consistent with tasting? And then it breaks into two groups, kind of less sophisticated tasteers in that they biologically and/or, experientially have a very limited ability to describe what they're drinking. And then it kind of also breaks into a little bit deeper. So you'll notice it kind of talks about generally, you're gonna taste this, this, this, and this in these kind of ratios. And then it says something like for more experienced drinkers, and that's pointing to people who have a more sophisticated palette that are able to detect and/or describe those things what they might detect. And so our goal there is to predict what the aggregate of consumers will taste so that the tasting notes are in line with what the majority of consumers are actually going to experience when they drink the wine. So one of the things that we've found is that there's very many wine makers and/or people in the winery are super tasteers. They're very experienced at tasting wine and what they taste is not what consumers taste. And sometimes that's not an issue, but sometimes it leads to experiences that are not what the average consumer gets and that can create dissatisfaction with the wine. - Is it targeting all aggregate consumers across the whole US just like hitting center mass or is it targeting the consumer that is the person buying the wine category or say 10 to $20 all varietals? - Right now it's hitting aggregate everybody. What we can do in recommenders is provide recommendations that it described the wine the way that specific consumer is going to taste it. So on the recommender side for a specific consumer, we can do that. But this is meant for the entire US, a set of notes that are gonna be the most representative. So if we were to have 100,000 people sit down and taste it and write out the notes, the AI is predicting what that would be, what the aggregate of those tasting notes would be in their percentages. If we move over to the right, we're looking at segmented consumer appeal. So this is interesting. This is where we talk about say high end wines versus crowd pleasers. If we wanna think about it that way, I mean, I kinda hate to characterize wines that way, but if we take again, say that cupcake and that Chateau Le Grange example, where they're both a little over 180, what you would see though is that 10X, the number of people will rate that Chateau Le Grange above 93 points. Whereas very few people would rate the cupcake that high, very tiny amount, but the number of people that would find the cupcake appealing to good is much larger than Le Grange. So what this does is it gives a little bit of insight to the buyer on the type of consumer and the way that this wine might be presented. Every wine has people that are gonna think that cupcake Pinagregeo is a five star wine best wine they've ever had. And there are people who think it's a one star wine, so where's the one they've ever had. And there are differences in distributions for every wine. It's a way to give an insight into the buyer. So buyers are sometimes very interested is this kind of more niche wine that some of my consumers are really gonna love, or is this something that I can put on sale on the weekend and I'm just gonna move a ton of these bottles and everyone's gonna be like, yeah, this is a good wine, right? So the number that we always use in Tasty, that retailers and wineries are often looking at is what we call an 85% match. And so to think about, I'm just giving you some insight into the way the retailers who work with us look at this, an 85% match in Tasty is about what a consumer on say, Vivino would rate, 3.9 to four stars. It's on a critic scale would be about 88 to 90 points. That's about an 85% match. And what we've found is that when a consumer gets an 85% or higher match, that's the level at which they recognize they notice that they like the wine. They notice that it's a good one. It's that point at which you might say, wow, what is this? It's also the point at which recurring sales start to happen. So a lot of people can find your wine. Hey, yeah, no, it's okay. It's good. They're never going to go search it. They're not going to go look for it. It's not important enough for them to do that. This helps give a little bit of a breakdown to this segmented consumer appeal to give the retailer some visibility on what that's going to look like. - This example, this exceptional 93 points are higher. It says it's a 1.3 million consumers, potentially. Those are pretty big numbers. And the very good is 15.2 and 88 for the average appeal. Is it dialed in anything more specific than again, this is still also kind of critical mass across all of the US, or do you hone that in by the category as well a little bit more? - This is critical mass across the US, the random consumer. So again, we do do what you're suggesting when we're working with retailers to help them really break down their client base and understand even store to store. We have retailers that restaurants and or brick and mortar stores that are very close together will have very different consumers walking in them with very different preferences. We've done an exercise with our whole del hay's where two stores, 15 miles apart in one city just had completely different preferences walking in the door. And so understanding a middle upper class or middle to middle upper middle class clientele in a given region can be very different than the middle to a little bit lower middle class. And we can see dramatic different. differences in the preferences. And part of that is price. They've got a different price category they're looking in. And part of that is their experience with wine. All of those things are important. - How would a salesperson use this segment and consumer appeal exactly? - So a salesperson might look at this and it depends on who they're selling to. If they're selling the wine and this is meant to be a crowd please, or so let's say a salesperson is selling a wine for a wine by the glass at a large restaurant. That category buyer, they wanna mitigate the people that are gonna have a bad experience. They're not trying to create the best experience for everybody. They wanna wind a whole lot of people can drink and go yeah, that's pretty good. On the other hand, if they're selling a premium or super premium or ultra premium wine, they're gonna lean into hey, look at the number of people that are gonna rate this wine, 93 points or higher. This is how it's gonna pair with your menu. It's gonna go great with this filet, with whatever it is. And they're gonna build the argument around that. So we're trying to provide the material that a salesperson can look at this and then help make the argument that whatever they're trying to make. And again, every one of those pitches is different. When we first went to salespeople, they were adamant. I need to be able to merge whatever data you can give me with my tribal knowledge of this buyer. I know this buyer. I just need the data that's gonna help. And so we try to provide a lot of data points that they can leverage. - Is there a way for a salesperson to figure out who of their customer base would rate that wine exceptional so that they can match the preferences effectively? - Absolutely. So if a client of ours, which could be a restaurant, it could be a brick and mortar, launches the pastry recommender, then we can recommend precisely the wines that specific consumers are going to like to those specific consumers. And what we found in retail, that's compelling to retail. In every place we've launched this and we've launched it from the East Coast to West Coast and lucky save Mar Aho del Hay Stores. What we found is that generally, we've always seen an increase of at least 3% in gross revenue in about 90 days in the wine aisle. And sometimes as high as 12%. And so there is a tree's ability to recommend wines that consumers are gonna like is super powerful in retail. If a retailer launches it because, if we just think through this, if I know that we have collected the palette preferences of half the people walking in the store, so I can recommend a half the store. And I know what my conversion rate is on recommendations. And I know what the success rate of those, which is about 80% on recommendations. I can pretty much predict how many wines I'm gonna sell. If I know this many people are walking in, these metrics that we provide in retail would be metrics that show this wine is going to be in the top 10 recommendations in 1.6% of consumers walking in the store. In the top 10 of say 1000. So that's pretty high. They can look at that and say, well, if it's 1.6 in top 10 and I know what my conversion rate is, and we can actually tell them how often we're gonna recommend it because we know the consumers walking in there. They can almost predict precisely what's gonna sell when they do that. - Interesting. And so the last section of the first page of the WCI is a flavor profile, which has elements like fruitiness, sweetness and oakiness along a spectrum, and it's ranked against the category range. So that definitely helps people. I did find the style of the wine. What's your vision on how people use this section? - I don't know that they use it so much, but I have heard is that they might look at this and they might see, well, the fruitiness on this wine is below average for this category. The aromatics are a little bit below average. The sweetness is below average. Any of those things could be good or bad things. I think that the sales people use this sometimes. So we have had a winery as an example. I've had a wine maker, they could make this argument here. It's less, oh, that wine maker made the argument. Look, my wine is less adulterated. I'm putting less flavors on it because it's a fantastic wine. So they might use that as an argument. So how they use this and the sales, it's up to them. It's both material for the sales person and also for the buyer and how they use it as plethora. They all do different things. - And does the flavor profile the data in that underline or underlie elements of the pastry score or the consumer taste preferences at all? - No. So let me just make sure I understand your question. Is fruitiness higher or lower or aromatics higher or lower? Does that directly drive likeability? No, there is no direct correlation to that. There are indirect correlations in the overall flavor matrix. But no, it doesn't mean it's better or worse. As you know, if I'm pairing the wine with a certain meal, I have a certain type of general preference. I may prefer sweeter wines or less sweet wines. That may be relevant if I'm selling a wine as an example, if I were selling a certain wine in a certain area of the country, and I happen to know that in this area of the country, or particularly in this area of the city, that because of the cultural background there, these people are going to prefer sweeter wines. That might be an argument I would make and say, hey, this rib wine is sweeter and that's actually what your consumers like. Where if I tried to do that in Beverly Hills and I tried to sell a sweeter rib wine that probably wouldn't go over well, I would make a different argument, right? But there isn't a direct correlation. Jumping onto the second page, which has flavor profile again, retail talking points, and that was all feeding into a group of food pairings. I am curious how do you see people using these elements and how are they generated? Sure. Again, we have our image to try to get people to remember it. Again, this document here is really designed as a training tool for the person selling the wine to the consumer. Food pairings, these are just basic food pairings, goat cheese, fresh berries, grilled chicken. What we're trying to do is give something that's very easy for someone to read to understand how they might talk about this wine. Retail talking points, same thing. The AI is looking at sometimes, certainly dozens, if not, scores of documents, provided by the wine maker or data. And then it's pulling out what the AI believes is going to be most impactful, a range of comments that we most impactful to sell the wine in a given environment, in this case, the United States. There are many more that the sales team can have access to dynamically through the sales accelerator portal we were talking about. They can really curate this. But these are just some generic talking points. You know, if I'm going to hand this to my servers and say, hey, we're carrying this wine starting Monday, yet familiar, these are some basic things they can talk about that might be meaningful to consumers that they can regurgitate. And then the last one is the flavor profile. And the idea here is instead of making a server or trying to read a paragraph or a critic review, it allows them to look at this and they can see, oh, strawberry minerals, red fruit, raspberry, I know how to talk about this wine very quickly. In one second, they can kind of understand this. So the idea being to give them the minimum data necessary for them to sound competent about the wine and not mischaracterize it to the consumer and drive it that experience. - And so there's a bunch of colored boxes on the flavor profile section and their different sizes. Do the sizes mean something or does their positioning left to right mean anything? Usually there's a run or reason for why infographics are produced, I'm just curious. - Absolutely, so this is called a tree graph. So if we want to look up a tree graph, you can look at that. This is the percentage of people, so if you think about this as percentages of consumers that would recognize these notes. So the majority of people are gonna recognize strawberry. The next largest majority of people are gonna recognize minerality and they would call it that strawberry minerals. The next largest is red fruit. They're not gonna characterize it specifically as strawberry, they're gonna say red fruit. And the next one is raspberry. So it's giving you an idea from largest to smallest with the majority of people are going to say. So as a server, if I say, oh, it's got this really cool minerality and a lot of strawberry, that's all they need to say. And that is what the consumer is gonna experience if they try this wine. Now they can go deeper if they'd like, but of course we're only showing a small, we're going down to blood orange and watermelon. The AI is actually predicting probably on average 200, 250 notes for each wine that people will say, right? There are people who are gonna say graphite, lead pencil, but they're very small number of people. So what we're looking at are what are the majority of people gonna say when they taste this wine? And to give them again, to make sure that what the server is recommending, they get excited about it, the consumer, when they try it. Oh yeah, strawberry and minerals, I get that. And so it's consistent. It's in line with what the experience is going to be. So differentiate that from the flavor profile above. The one on the second page is more about perception and about how large of groups will perceive. These flavors in the one above is more like the analytical, the actual what the AI model tasted or what the ML model tasted for this wine. So remember, we're always pretty much consumers are gonna say, this tree graph is predicting the percentages that people are going to perceive these flavors. If you look at the pastry notes, it'll be very consistent with this. It's basically saying the same thing in a paragraph. And then the flavor profile, where we're talking about fruityus and so forth is also consistent with this, but we're just not talking about flavors. But the fruityness and so forth is consistent. - And you mentioned that the sales accelerator is included as part of your subscription to the pastry. For those who aren't familiar, what's the general pricing for pastry? - Real simply, pastry has a yearly subscription right now, a $1,580 I think. Forgive me, I'm not a salesperson. That gives you access to the dashboard and all of your data lives there and all of your interfaces live there. And then every time we analyze, say a wine. So if we wanted to have all of these data and access to the entire sales accelerator system for a wine, that's $370 per year. So you send in the wine, it's $370 to analyze it. And then all of this is available to you. And then there will be, I think starting about next week or the week following. So pretty shortly here, there will be a new page. The marketing teams can access where they are feeding all of this data to the AI. So right now, this is kind of beta and we're working with our clients to manually enter this data into the AI. But there will be a page within the program where they can go in there and they can maintain this data real time. So that the marketing team is consistent and all of this data is real-time across the country for anybody who's accessing it. As we were saying earlier, whether it's a live interaction with ER and DC or whether it's a real-time job, whether it's somebody using the sales accelerator portal, a sales person or whether it's a WCI, all of these things will be up to date real time. - And so as we sit here recording in September of 2024, there's a lot of macro issues impacting the wine industry, younger generations like Gen Z, or drinking less, people are drinking less in general, even baby boomers, government agencies are saying that no one should drink alcohol or zero amount of alcohol is good for you, but focus on health and wellness in general. How do you think, "Testory" can help wineries, specifically navigate these conditions more effectively? - I think there's two things that we can do that we are doing. One is, as far as consumers go, and their perception. One is, we are working with wineries to create low alcohol and or no alcohol wine. I think that personally, and I think as a company, we don't think that normal alcohol wine is going anywhere. We have fluctuations in the market. I think there are pullbacks for lots of reasons. I don't think the wine industry is dying personally. I think that's overblown, but I think that providing some offerings that are addressing consumers that want an 8% alcohol, that can be really difficult for traditional wine makers. It's just not an area most have played, and Tastory can help them engage that in a really thoughtful way and put a good product on the shelf quickly. And so I think that we can help winemakers where they have less experience there. The other thing that I think that we can do is our marketing tools through sales accelerator. So one of the things that we're really working on our clients with through sales accelerator and all of these tools is marketing and interaction with wine that is addressing the younger audience. So as an example, when we do food pairings or recipe pairings, we can do classic pairings. The AI can say, oh, this red wine's going to go great with the beef bergenjon and a blah, blah, blah. Yeah, we can do that, and that's great. But what we find is that when we have wineries that are addressing younger crowds, they're more interested in, hey, this wine will go great with the kale salad and frozen pizza rolls. The AI is doing really cool pairings that address kind of that age group and kind of where they are in life and where they are in their experience. And I think also there's less money. The younger generation, they have more issues with debt. They tend to live in groups. It's actually very communal, very social. They might have three or four roommates and they kind of get together once a week and they do something. And so I think we are trying to help our clients to really embrace that younger culture. They kind of want something, I gave that example because it's something that popped up in one of our recommendations. They kind of want to be healthy, kale salad, but they kind of like frozen pizza rolls too, right? And if that's going to go well with this penal war, then fantastic. And it allows us to engage them in a fun way and not such an austere manner. And I think that the wine industry, it's not going anywhere. It doesn't have to become something else. But I do think that the RTDs and the RTSs, and particularly RTSs, these are all these ready to serve cocktails. I think that we can plan that space as an industry very well if we do it properly. And I think we just see the be a little bit more fun about the way we speak to those consumers and it seems to be working very well. - I'm just waiting for the AI to use Gen Z emojis that are right there while it's future. - It's probably coming. - In wine industry as a whole, what do you think are two to three key things that it should focus on in the next year or two? - I think everyone is trying to fight for market share right now. Everyone is trying to get a way to wear this for their brand and sell their wine. And I think that's the right thing to be focusing on, obviously, but as we're doing that, I think we should be a little bit more stringent on trying new approaches and trying to sell wine in a different way and to the younger consumer. I think in that vein, and this is not self-serving at all, but I would say the wine industry could focus on using technology to do all of those things. Sales going to market, understanding their consumers, they could use technology to do that much faster and get a lot more visibility and to do it a lot more efficiently. So let's focus as an industry on how we could move beyond just talking about technology and understanding consumers and how we needed to let's try our hand at some of these tools. - And maybe for both of you, Kat and Charles, what are you most excited about for the future of Tastery? Maybe to start with Kat? There's a lot to be excited about. I think this year more than ever before, it's becoming apparent that it really is possible for us to create an ecosystem that would benefit everyone in that ecosystem. So I'm excited that we're growing relationships outside of production. I'm excited because I could see a path toward how those relationships will benefit our current client base even more. - I think I would second that. I think what's exciting about Tastery is every time we do something, it takes the AI 60 seconds to do and they come back and say, yeah, we got a contract for 5,000 cases, that's exciting to me. I think to the degree that we can help the industry sell the wine and I think kind of tying another point together that you made a second ago, I think the really impactful efficiency here is right now every wine maker, every winery has to market every wine to everybody. That's the only method available to them. And that's super expensive to try to market everybody. And what we're trying to do is curate that list down so you can market to the people that are going to like your product and close those deals. So spend less money marketing to make more people happy. And I think that just drives huge efficiency, right? If we have 10% of the people are really going to love this product, let's find those 10%. Let's not market to 100% of the people. It's just very expensive and inefficient. - You're right, it can be more expensive to market just to those people who will buy it, but if their hit rate is a lot higher, it's worth it. - Exactly right, if we get those recurring sales, then it pays for itself over and over again. - Thank you both for explaining the WCI and what Tastory's been up to in the last year. We'd like to wrap up each episode on a personal note. So for each of you, what is the most coveted wine in your cellar and when do you plan on drinking it? - I'll go first, I'm going to answer this two ways. So one, I'm a big fan of there. Some wine makers doing amazing things, making $20 bottles of wine that are very just pure and very untouched. It's actually very exciting to try that. And I like those wines and there's more and more why makers doing that. They're actually moving more away from kind of a constructed model and I find that interesting. So I would point to like an almond pinon noir is just a very pure pinon noir, inexpensive. I like that. As far as exotic wines, I have the luxury of trying a lot. I would say that I have a 1943 mosquito from Croatia. I have the luxury of trying that wine with Michelle Jacques who has a very large wine collection in France. It was a fascinating wine because it's a sweet muskato, but imagine a sweet muskato that has the absolute essence. I mean, yes, tastes like Louis XIII cognac combined. And so I'm looking forward to trying that again. It was just a very interesting flavor profile in that wine. - And in my collection, I mean, I have a Laurent Perrier Grand Sequel and I was recently visiting my friend Juan at Antenori Napa Valley and I have a bottle of the proficio, I believe it is. It's like a cab-solve cab-fronk blend and I don't anticipate I will be drinking it anytime soon. So I'm not sure if that answers your question. - That works. Sometimes you have to sit on those wines. Well, thank you both. We appreciate sharing all your information and telling us what's been up to date with T-Street. - Thanks for having us. - Thank you so much. - Thank you very much. - Thanks for joining us. If you loved this episode of X Chateau, we'd love for you to subscribe, rate, and give a review on iTunes or wherever you get your podcast. Until next time, cheers. (upbeat music)