06:14
Nicholas
The founder and CEO of palette club, which is a new platform created around your taste using artificial intelligence to match the best wines and premium foods to your personal tastes. We use 200 traits found in tasting wine to make calculations and algorithms to figure out your personal taste.
06:41
Chris
Let’s get into that, like AI and taste. How did you get into that?
06:45
Nicholas
Well, when I moved to California six years ago, as a Frenchman, I stopped, I knew a lot about wine as you could expect from French people. I suddenly was lost in front of the shelf. It’s like an idiot in front of, the wines from all over the world. I figured that’s actually a main problem in the wine industry in general. Since I built quite a lot of e-commerce companies and data science related recommendation engines in the past, I asked a bunch of mathematicians, data scientists, and wine experts. If we could come up with a real model where we would solve the problem of choice and, in France, for instance, to start there, French people drink French wines and Italian people drink Italian wines. You have a limited selection in the U S of those. At the same time, you have wines from all over the world, literally a million different bottles in the world.
07:43
Nicholas
Just in the U S alone, 225,000. I find that the TTB every year. Even if you’re a professional summit liaise, I work with they taste max like four or 5,000 bottles a year. Even if you’re a pro and you do some blind tasting, you’re just going to get very often wrong or not really know about every seat. The other reason why I figured that it was a main problem is distribution and supply chain in the U S because you have that St Joe’s system have a lot of middleman. It’s basically a product pushed industry, which means in the end on the shelf, the wines, which are decided by all the people done so upon, the best margin for them and not necessarily about what’s the best value for you. It should look at it as an industry on the whole, it’s a very large $360 billion industry in the world.
08:38
Nicholas
It’s also the most fragmented of that size, where in the end, if you accumulate the fact that the distribution is not driven by your personal interests and the sale is never about your taste, or it’s really hard to figure out actually your taste and particularly the guide setting, bind to you, that is actually a very, it’s very hard to get the good value. Basically, I would say 90% of the cases, even when you think, or you going to go for that cab or Sharnay, they’re not making a good deal. This is what we solve with product club,
09:17
Chris
I can see a number of hard problems that you’ve got to solve. One, you’ve got to know lots of different things about all of these different wines. You talked about, how many millions of bottles there are with however many moments, the more there are coming every year. You’ve got to also figure out what people like and do, what do you market segment those people as well? I mean, how do you wedge into that in terms of defining like an AI model?
09:42
Nicholas
This is where we came up with a complete different approach than the industry, because that’s a long-lasting problem. I would say you can overdue problem in the wine industry. Everybody came with all sorts of gimmicks, right? Wine critics calls and whenever, competition fans, gold metals, this and that. Also grapes like in the us, particularly wine is very defined by grape. The reality is if this is not a very good indicator, if it took two wine experts, I mean, considering premium wines of all of like $15 price points, the grape is only like maybe 25% of the final taste. It can define a lot of things obviously, but, and then the blends and then there’s vintages. Before anything else, the winemakers touch, which I call artists when they do their real good job is made of, dozens of decisions during the year. Obviously during the wine making and fermentation and all the different processes and steps, there’s also a lot of unknowns, such as, 73 additives are actually authorized in the S for about 40 in Europe.
10:53
Nicholas
These are not indicated in the bottle either. There’s so many ways to play with what is inside the ball. To reply, to question more precisely the way we considered is we came up with this model of 200 trades found in wine tastes, we reorganized everything, and we created a customer feedback loop using blind tasting. Why we use blind tasting is not only because it’s fun. You could say, yeah, that’s a gimmick, but it’s obvious that people are very biased in their ratings and their consideration of whether, if the wine is good, if they know the label before. The way it works is we ship starting kids with four half bottles. You can also take a quiz, but that’s not as precise as tasting the wines, right. That starting kit comes in buying taste mode. We cover the labor with a small tissue and secret well number.
11:45
Nicholas
It’s actually a fun experience to do with friends. With the only thing we ask from customers is to rate the wines from one to five and in the app, it’s a very simple, straightforward process. That is enough information for us to stop putting you on the map. I will get back to that, the map of taste, but why we have very precise information as opposed to any other type of things out there is because previously before we ship the ball, they’re tasted by professional seminars. We enter in our data model about, on average 30 traits precisely about those wines. This is done by these two different, so many days. So they have different tastes, et cetera. That gives us a lot of information because if you consider these four bottles to start with, of course, very different and demand too. There’s always like some, you will like not like, and have like, very extreme traits, usually matter things like a CDT alcohol level sweetness, tannins, Oak aromas, and so on.
12:55
Nicholas
With those four ratings, we put you on the maps. So how is the map? The map is now more than 12,000 greetings from customers, thousands of profiles. We have seen with the data clusters and we’ve defined those clusters. We’ve actually given them funny names, which is the acid tripper or the Italian grandmother, or no. Again, so the experience is more fun. We, after those four ratings, we basically explained to you that, your affinity to certain cluster and basically also your main cluster, it doesn’t mean we put you just in one cluster and that’s it. We consider you have an affinity with different clusters, right? The other thing we do is we show your main sensitivity to the main traits. If you rate more miles than actually show secondary traits such as cherry or black food, or, king of spice, et cetera, and we also show your percentage of match to different classical wines as they are supposed to be.
14:02
Nicholas
That’s more or less a percentage match to the 140 different types of wines in the world as this is another type of mapping, basically. So, so that, where you are. That’s the first step basically, to try to understand your taste.
14:19
Chris
What you want and when you’ve got four bottles and that puts you on the map, how does that work then? Do you then get into like, here’s the next range? Here’s another four bottles to see if you like these ones. It allows us to push you in one direction or another to refine. Is that how it works? Or,
14:35
Nicholas
Yeah. You have two ways to go after that with us. And that’s the business side. You can either buy wines to the shop. Every wine we have in the portfolio has been duly expertised basic by the somebody A’s. We indicate the percentage of match for every bottle, using all those traits and using your profile like Netflix wine, if you want, but this is a gain. It’s like a very vertical approach. It’s not like, oh, we’re going to give you a percentage of match for millions of balls. We’re like very precise on the balls we have in the portfolio. So, and the main way we do, and this is where we come up, came up as a business model actually is subscription-based. On your terms, so you define the price point, the number of balls, the frequency of shipping, you can change it anytime. It’s actually the algorithms choosing the balls for you.
15:32
Nicholas
You don’t have to do anything. You just have to choose your parameters, basically. It’s as if you had a private, so many AI digital tastes that in your pocket, which chooses wines for you, but knowing your taste and using that eScience. It’s basically for it’s, even for connoisseurs, which are actually in most cases, pretty bluffed about what we came out with a taste it’s for everybody likes, good wine and just wants to be completely taken care of. This is it’s our vision, basically of the world of tomorrow of taste and premium foods. Obviously we want to apply this, not only to wine, it’s a global approach for things like chocolate, coffee, whiskey, even perfume, every, I mean, wine is before anything else taste is before anything else you’re linked to your smell and the same traits you have in mind, you can find in many other things in life.
16:32
Nicholas
Think of us like the first digital tastes type in the sense that we try to measure your taste, literally with data science applied to wine at first. To solve that wine issue, that wine buying problem in general, bringing new value and then applicable to many other the things.
16:52
Samuel
Well, what’s that engagement like when it comes to people actually giving that feedback. Cause it’s very easy because I know this from experience on something like a, I’m not sure if you have it over there in the U S but we have something called hello, fresh, where you get food delivery and it’s all packaged for you. And you’re encouraged to rate those delivery. I don’t think there’s any kind of AI behind the scenes, but they always ask for feedback on the meals and I barely ever rate them, although I should do, because it allows me to remember what meals are I liked and enjoyed? What is the engagement like? Do you struggle with kind of getting people to rate every box that they get? If you don’t, how have you encouraged that engagement?
17:35
Nicholas
Yeah. So this is very good question. I know how to pressure, and obviously I’m very familiar with recommendation systems and, Amazon is probably one of the pioneers there about kind of trying to get your ratings and reviews from customers, et cetera, everybody does it. To reply straight more straight to a question, we have 80% people re rating the boats because it’s different from all the, everybody requesting us to stay in a hotel room. I would use what this hotel or whatever the reason is because our entire experience is based on that. The entire like journey and promise of the product is based on that. People become customers, they know, this is basically about it, incentivize it with a loyalty program, the more your rate, basically you get coins and then you can get more free bouts. I don’t even think that this is like the main private, the main driver is because people understand this is really ultra personalized.
18:33
Nicholas
The difference with just rating for other people is that you really get a tasting profile. You really get a reward and output generally from the app. So you learn about your tastes. It’s, it’s a educational side of it, I think which is important for people. Also because you’re solving a problem, which is like so many bowels around there, but there is an interesting component in your question I want to venture it is, you know, the ratings, obviously it’s great for companies to get feedback from, you know, the products they deliver and the kids have had a fresh, but either do that well, and you know that around now for awhile. So here is a very important thing. Usually people don’t mention too much is there’s a big difference between being centralized only on your like super personalized, so on your personal taste and of course, customer satisfaction and considering as opposed to that ratings from all the people.
19:37
Nicholas
So this is where categories matter. If you look at music or taste, for instance, I think it’s very similar. The reason is it’s just because it’s your own personal data doesn’t mean that if you’re very good friends with someone and you have the same social graph that you’re going to like exactly the same wine or the same perfume, all the same music, not at all, it’s very personal. Recommendation systems are built on global, what are the people think on these type of categories, which are very personalized, that is not very meaningful. In my opinion, it’s more like, to try to drive more sense about, people’s purchases, et cetera. So, obviously if it’s, a movie or a restaurant there’s components of social graph and people who are like you, that’s totally fine. I think it makes sense there on these type of categories and things like, again, like wine and taste and music, you should be really focused on that person, that particular person.
20:46
Nicholas
And this is what we do. This is quite different from other products. Again, I don’t have a particular opinion on the rating system or how it goes with our refresh, but most companies, they just get feedback from the customers. They put that into they use that data obviously as well. It’s, it’s good to measure satisfaction. It’s good to increase retention and to understand the customer and even to somehow understand the customer tastes, I believe they were factored in, but I also believe it’s pretty far away from a pure, that assigned smaller, which is when you focused on so many traits, what we do.
21:26
Samuel
What I’m getting from that then is because AI and it is such a key driver in your business model, not necessarily the business model, but you’re almost your, it’s the reason why people would go to, it’s almost like if you don’t engage with the platform, if you don’t provide that feedback, you’re not going to progress as a customer. I’m not going to progress or get what’s right. For me is just not going to know anything about me. Is that, is that right?
21:55
Nicholas
Yeah. We have a few customers, there’s different types of customers. Some of them, they’re just happy that we pick up wines and they don’t engage too much into the ratings. They’re just happy with the ones. And that’s it because we took rates. I mean, it’s a double curation. If you want, we really like great wines from smaller autism mind makers, which make better value. There’s the creation and matching to your place. So, but most people do rates, they start rating a lot. Once they have a full profile that kind of, they don’t need to write that much, but they still do, and occasionally, and it’s still fun. I think, in data science, it’s where you apply it, to what. If you consider the rhythms of, data science, machine learning, artificial intelligence, or whatever, how you name it, we are doing, we’re still applying something which is quite simple and which doesn’t have that much data.
22:56
Nicholas
If you ask some areas of course, of tech, where it’s way beyond we might get there actually, because there is no limit in the number of things you can measure in matter of taste, it’s actually a very complicated saying, you can also take factors into account such as, weather or, season and try to get some information about context even the time, and of course, obviously food pairing and all these of there’s like a lot of type of data you can build in. I think what makes sense for the consumer is that it’s very simple for the consumer that it brings value to them. This is what we’re trying to do. There’s one important thing here on the business model side, somehow it’s, I want to explain as well, is Sanks to that model. Thanks to the fact that we figure out people’s personal tastes. That enables also a compete, what I call a reverse supply chain model.
24:00
Nicholas
If you look at foods in general and particularly wine, as I mentioned before, it’s a product push industry. You have like wineries produce wine that try to sell it directly to the consumer, which they don’t necessarily know. They mostly push it through different distribution channels. Again, people make decisions for the customer, basically why they don’t know the customer, right? That maximizes margin until the shelf and you pay actually the maximum price in almo it’s the other way around you figure out people’s personal tastes. The more you have, and particularly in the subscription model, we already know how many balls, which price points, which type of clustered strain you to ship. We can literally go buy the wines according to people’s taste. We taste a new wine, a winery and we enter the data in our backend, we instantly know how many people that matches and everything is driven by the data.
24:59
Nicholas
I went further to from your question, but I think in my world, it’s more about, how can the data, the AI be applied so that it makes sense and brings value to the customer. That’s every single bottle brand is that so, this is what we want to do. I see it. This is how you can scribe.
25:23
Chris
Why are you feeding that data back to the actual winemakers then as well to say like, should we’d like more of this or more or less of that,
25:32
Nicholas
Absolutely. We don’t share private data such as, customers tasting profiles, emails, or whatever that the remains, we never sell it. We never share it. Yes, on anonymous space we started to share and winemakers are actually way more interested than salt in the first space to have that data because they don’t have it even like, premium wineries have been operated or breeding for so many years. They use a lot of data in even, withdrawns and measuring levels. They use a lot of data in the wine making process, but they don’t know the data, the clusters, the things we start to figure out now from a customer perspective. And so they’re very interested. We haven’t, this was not the starting business model to try to sell it to them. We’re more like we want to be considered as partners and bring them, valuable data, the type of data which they’re interested in is literally linked to the clusters as well.
26:40
Nicholas
To explain more what means it’s not about, oh, there is like 80% of people, like from CBT, it’s more complicated than that. It’s clusters of data. A matter of taste is combinations. If I take a very simple example is, I like chocolate and I like, or blueberries, but I don’t like when you have dessert with chocolate dessert with blueberries, the respirators combined. So, and this is not many people like that is because, the chocolate tastes with whatever the acidity of the berries doesn’t mix well with my taste. In Singh’s wine, you have sink of, you have dozens of components. And so it’s a mix. Of course have some components from way descriptors are way more important than others, but let’s say one main cluster is people like strong acidity with medium tannins and earthiness. You figure out actually a sub gestures that say just a rough number of 10% of the people really like that type of text.
27:44
Nicholas
Okay. So then you will find sub clusters. You will actually find that 60% of the people like that. They like it better if there is a lot of black fruits. Okay. Once you have more of these information and data, we see you can start to put people on the map. I can tell you one thing, which is before I started this, I was kind of wondering, oh, what if, everybody has the same taste and then every single,
28:10
Chris
Oh, I was going to ask you that, whether there’s like, is there like a, a bell curve where there’s a big cluster of people in the middle and then it fans out,
28:17
Nicholas
There’s absolutely not. It’s actually the opposite of what we saw. It’s completely, you know, it’s scattered. But again, there is like clusters. So, as we’ve done numerous tastings as events as well, and they make, and now again, where I saw people, even couples like fighting on the ratings of the wines because, the two first wines, the first person who would put one star and the other five star, and then they would, the second one they would do five-star one-star exactly the reverse mode. It’s a very different from one person to the next. Again, there are similar tastes. Good. Give you one number for instance, which is, I used a couple of times, there was robot conquer and people who are listeners here, they know of course that name if they been into wine. He was the most famous wine critic, particularly in the nineties. You had a very good of course, a sense of taste.
29:18
Nicholas
The states was they give like bold wines, mostly, and, very structured, strong body, et cetera. So you rated them very high. That changed things in the industry, because if you gave a very good rating to a wine, that wine would literally sell much more, whatever, because of its basically your, or a celebrity type of way. He’s like retired for why, but there’s all these wine critics there, but this is one particular taste. We measure that and among Americans or Europeans, and we consider how many people really like that type of both taste, which you could be defined as, the Napa cab or the kind of heavier left bank bottle, the shuttering of stupid type of products, it’s actually only 25% of people will tend to go more to those types of clusters. That means, the industry is being like biased for a very long time by that type of taste.
30:21
Nicholas
It’s not the right taste for 75% of the people. It’s not, it’s not their favorite wines. Of course now things have changed a lot of people like burgundies and more balanced wines from, and all sorts of, again, like more specific tastes would more spice more this, more that, but anyway, this is what we figure out.
30:41
Chris
This a totally unique idea then in the wine industry to apply data science, you’re the first people doing this?
30:48
Nicholas
I would say we’re the first people to do it with that type of approach. We’re not the first people. I mean, that would be very arrogant on my side to say, we’re the only ones using data. Obviously, there are some famous apps which have been around for a long time. They have a lot of customer ratings about, but they have a more horizontal approach, which is like, okay, y’all and millions of bottles. This is what people rate and depending on your ratings and what you buy, we’re going to define more or less what you like for me, that’s more the old school of reviews and ratings, but again, they have a lot of data. There they’re also have very interesting data. There are people have worked on taste as well, more particularly. I mean, there’s a lot of research on taste, which is goes way deeper than we actually do.
31:39
Nicholas
There’s a lot of data as well on the wine, all this things, food by the industry where they use all sorts of data. What is unique about us is that feedback that we use number of data points. We get that unbiased blind taste thing exactly on the balls where we have the maximum information. So that’s to my knowledge, nobody else.
32:02
Chris
And I like the blind tasting thing. I saw that on your website, especially with the party as well, hosting a blind tasting party, that sounds like a pretty good way of gate gaining customers. Right.
32:13
Nicholas
It’s funny because we started this really generally as a, saying, Hey, we need to use Bryant isn’t because we have all those proofs from surveys and stuff that there’s so much bias if you already know the labor, right. He said, we need to use blind tasting. Obviously we figured the customers love the experience and it’s fun. They love to do it for their friends and compare the profiles. Many people came to us, oh, you should mix this with the dating apps or people that the same cluster and saying things like that. So, so obviously we are continuing and we put it as a core feature now because I mean, customers like it and to go back to one of your prior questions there about, how many people, right? This is also because it’s core of the product is are people understand that it makes sense to write it.
33:04
Nicholas
So people writing. So, no, it’s so far, I think were not going to change that type of feature or aspect of the product for a while because it works. It makes people happy. I have a good time.
33:18
Samuel
There’s wine dating on the horizon,
33:23
Nicholas
It’s more it’s getting Steven my priority list, I would say. Yeah. I mean, I haven’t talked with dating apps yet about that, but that might be a thing in the cost structure.
33:36
Samuel
With what you were sending them to customers. W you mentioned you put a bit of tissue over the label. That for the parties or is that via this, the subscription service or I’ve got that wrong?
33:48
Nicholas
Everybody we ship is PRI is in our logistics center wrapped by a purple tissue for red and yellow tissue for whites and roses for Rosa. I mean pink for it.
34:00
Samuel
They are able to see then the label, if they want to know what the.
34:04
Nicholas
Oh yeah, of course. I mean, you can cheat and you, and any way we don’t produce wine, we are not doing what a lot of fake wineries, somehow on the internet selling bulk juice. They buy like, 50 cent or dollar leader, then they stopped the labor on it. They should point, we don’t do that at all. We buy finished wines by artisan wine makers and we show you in the app, all the details, the tasting notes, the wine makes where it comes from. Obviously you can remove the tissue in a fraction of a second and see the label. It’s not about, we don’t want to substitute and say, oh, this is like a pedal wine. It’s, it’s curated from wine makers and buying tasting purpose is only for two things, obviously is data science to work better and for fun. So, no, of course we assist sticker with a secret bottle number, which comes with a tissue.
35:01
Nicholas
Basically your next shipment, let’s say you have wine number 2 39 a is auto-populated in the app. We already know of course that you have received that ball. You just go in the app and you see that 2 39 just have to rate from one to five. It’s as simple as that.
35:19
Samuel
Getting back on the data science thing again, in my head on I’m thinking of two major kind of points of interest when it comes to the data that you could cross reference to build up a profile. In my head, I’m calling these internal, external and internal factors would be stuff like time of year though, a wine was bottled or it could have been anomaly potentially, maybe that maybe the manufacturer maybe tried something new in that bottle. Well, there was something unique about that batch that went out, that they could determine it could be that they had a weird windy cold summer or something like that. These sorts of factors can change the characteristics of a batch of wine, the external factors, which you or the supply might not have control over, could be something like the things you mentioned earlier around that they like chocolate or blueberry.
36:19
Samuel
It feels like the customer has potentially a million different possibilities that could affect why they’ve rated it a certain way that you might not have, you might have control over when it comes to data collection or any kind of assessment on those factors are the wine manufacturers providing you with that, those kinds of data such as weirdly cold summer, and are the customers providing new some personal data, or if you’ve got your system and you’re just collecting data that plugs directly into the system that you’ve defined.
36:57
Nicholas
To start answering here. I would say, I don’t think perfection exists here. I mean, you can have the perfect wine with the perfect taste, whatever. At some point you have an epiphany and like, can’t get better, but more globally and things, data about what, the type of things we measure with wine and your personal tastes. We know, I mean, you can’t reach perfection, but basically you’re, you have to aim for it. Right? I will give you one number, which at least shows that it works. So, because we haven’t talked about that, we can measure the difference between when we don’t know anything about your taste, right. As when we do the first shipment, which can be a starting kit, or just because you’re right. Some of our bottles for the first time. The average rating, and, from on a scale of five is 3.2. You could stay people a bit more drastic, and it’s not that high because obviously the first ship are very different.
37:53
Nicholas
We ask people to rate low when you don’t like it, so that it works better once we have algorithmic fine. Basically when to choose bowels, because we have a starting profile from you and you rate bowels, and we can measure that as well on the same balls, as well as in same person. The average rating goes to 4.1 out of five. It’s a significant increase, which tells us at least doing works. We’re talking here about thousands of ratings and profiles. It’s, I think the numbers, I mean, four already. We know we’re never going to be able to get to five just because, different people rate differently. To get back to your, other of the questions is there are numerous data points we can use. Your question about vintages, obviously the vintage and the Ava, and all sorts of other components, such as, very factual components, such as, the grapes use then, and even the type of wine maker, which usually have their specific twist it’s the trucks are in the system.
38:59
Nicholas
The main thing is when the wine is tasted by two, so many days that’s a 15 minute process, whether it basically tastes several times, it’s the same wine to enter what they think about the data. Again, these are professionally trained, so they usually have a sense, a much higher sense of things. I will explain to you just to one trade, for instance, okay. A professional, so many tasting wines that makes the difference between fruitiness and sweetness, somebody who just likes to drink wine doesn’t necessarily make that difference. Okay. The other part of your question is, the data about context of, obviously there are things we will never know when customers red wines, it makes a difference if they eat food at the same time, we recommend to start with not to do it. It doesn’t mean we’re not interested in food, pairing on the country, but then things get, a dozen time, more complicated than they already are, and people’s tastes react, will react differently.
40:04
Nicholas
It’s also, if you drink the same wine in the winter or in the summer and the outside temperature is not the same and also the, which temperature the wine is served, it makes also differences. There’s, there’s like zillions of elements and data points. Some of which we could guess, because if we ship to starting kids in January, we, and the ratings are bad in January. We, we know that. The assessments, the magic of data science is also that you can start to grasp things from global data as well. Let’s say, give you an example of that. We know the average rating of a bottle, and we know the average rating of a starter kit, the average rating of, what people basically do. As soon as you raised rated, like for, let’s say eight or 10 bowls, we know if you rate high or not, as opposed to other, but some people just treat everything from three to five and others, rate pretty low from one to three.
41:07
Nicholas
So we take those factors into account. This is, again, these are mathematics. To actually make a difference, and re-evaluate exactly where you are. Again, the more data you have, the more you can factor in. If you’re doing a good job there, then, just have a better output. It’s, it’s kind of unlimited again, can’t reach perfection, just trying to, assets more and more that points to make your system better.
41:37
Chris
That’s a really interesting point though, on the, in the, so many things that you’ve said that I want to get into, but like the in terms of the ratings alone, and just trying to normalize, I guess, between someone who rates within a very small range and someone who rates within a very large range, how do you deal with leveling that? And do you learn? Cause you can browse your wines through the pallet club website and they’ve got ratings on them, right? You’re using the customer rating for that. Are you using the adjusted rating? I mean, how do you deal with that?
42:08
Nicholas
We don’t want to confuse the customer and we think it’s not relevant to show what other people think about,
42:18
Chris
Oh yeah, you haven’t got those, sorry, my mistake, you’ve got the match point on it because I think this is a problem that has been existed in recommendations for a very long time, in terms of like, you can go on Amazon and you can see there’ll be a mixture of star ratings. For a given product, someone might have like a, a three-star rating for the product saying that delivery was really slow. Actually that’s a misunderstanding of what that recommendation was for, it was for the product, not for the delivery from Amazon. Actually your approach to it where you’re not showing that, but you’re dealing with matching in stat is a really interesting approach. I wonder if it could be applied to something like Amazon, but that might be a bigger challenge for those guys.
42:59
Nicholas
Well, again, I think it depends on categories, for me, that’s kind of the category of products, right? There are products which, where the crowdsourced reviews or your social graph basically just makes sense, such as, most, a lot of the categories in Amazon, a lot of, Yelp type of stuff, TripAdvisor to cetera. There are categories where it’s the experts, which count, like if you buy electronics for instance, or maybe, certain games or things like that. The category such as music or taste, which it’s very personalized. It doesn’t mean the other information, a completely irrelevant. It’s just, what is the primary score rating, which counts and not to mention, there’s a very big problem in all those ratings reviews, Amazon included fake reviews, which costs them so much money that actually you can read articles about it. That’s a thousands of people to combat fake reviews and also regulators as they’re coming in because of that.
44:06
Nicholas
It’s a very large and complicated world out there. I’m actually happy that we focused one thing on images like your particular taste and personalization, and then match that to the products without too much fuss around. It doesn’t mean, again, we use global data to make sense of everything we use. Also things we don’t, the customers. You don’t want to explain everything because then you get sick. Oh my God, that’s too much information. And you’ve got.
44:38
Chris
To leave a bit of magic.
44:41
Nicholas
Things which define right. The, the logic of that behind it. So.
44:48
Chris
She’s still quite reliant on the I’m going to completely butchered that word, but you’re still quite reliant on them to be able to categorize that wine off the bat, I suppose. Yeah,
44:59
Nicholas
Yeah. Which is again, in so many years are humans as well. They have also their flaws in matter of what they assess and rate and how they taste. If your question is about, could went venture out and use a lab? The answer is, yes, of course you could do that as well. There are many ways maybe we come to the combination of both, if you consider as perfect as it is, it can be, would be a combination between molecules and lab assessment that are, and some of the human assessment data. Yes, this is maybe where we are going in the future. You could have all sorts of other data combined. I’ve asked, I’ve been asked many times about DNA. Obviously your DNA is a component of what your taste is today. When I dug into that, there were two problems, getting your DNA’s for all these platforms is quite a lot of issues with privacy.
46:02
Nicholas
The second thing is we figured from research is that your starting DNA with tastes will of course, open up close certain doors. If you have the bitterness gene or the, whatever, the silent Regina very well known, like in 20% of people have that. Of course that can be defined by DNA, but then in the end, it’s really about all these tastes cycle. The fact that you like certain things at certain moment in your life is very, it’s still a very unknown field. There are people working on that, but it’s, your taste of today, if you’re like 35 is defined by, what you’ve been exposed to a childhood, the context and the DNA is of course a factor, but it’s only one among a lot of other things. So, we stick to as close as possible to the truth and as simple as possible for the consumer.
46:55
Nicholas
This is why we came up with, we do the heavy lifting with the, some of the apartments one end. Then, we just ask consumer to rate from one to five and by in test mode, again, it produces results which are already pretty satisfying. So.
47:10
Chris
That’s an endless list of possibilities of what you could do if you start introducing DNA into the field, I suppose, but that would be fascinating to see what the data would tell you. I know my question around this Familia was more to do with, that’s probably a limiting factor in the number of Hawaiians you can have on the site, I guess, is your intention to try and keep that quite boutique or, do you want to be able to categorize every wine? What’s where are you pitching it?
47:34
Nicholas
No, our intention is definitely very ambitious and to make a dent in the world of wine, much further than, just our own customers. We want to open our IP to others and maybe as a platform as well. There’s, the training of that, there are thousands of days in the world and I mean, they can be trained very well and can taste balls every rest. So, so basically they have a pro we could even then, have a specific data about the profile of assimilate to also avoid some certain biases. The logic is, we could give access to thousands of someone years as a backend, and then they rate, as many bottles as needed on the business smallest side, we are more of a Tacoma than a horizontal model. If you look at the main problem we’re addressing here, which is choice the paradox of choice. If you enter to a classical wine store, you might have 200 different balls, which is already very complicated.
48:34
Nicholas
If you go online that on sites like wine.com or whatever, the problem is multiplied by hundred, and now you suddenly have whatever 20,000 different balls. So, you have all sorts of ways to try to choose them, but it makes things very complicated. Our approach is more that we have a rotating inventory much faster with against smaller winemakers or bringing better value. And, you have always like 500 or thousand maybe different skews, but again, they’re completely curated towards your tastes. Everything becomes easy again to the percentage of match. Also the fact that you just let the algorithms choose for you with a satisfaction guarantee, by the way,
49:18
Samuel
Where does the data show where the customer’s success is? It your ability to, just get your customers and provide them the wine that like, yes, I love this. Or is it opening them up to kind of new wines or yeah, just surprising them.
49:38
Nicholas
Well, it’s hard to answer that question because it’s a mix and we don’t have enough. I mean, you don’t exist since long enough to have done surveys and measure that precisely what we measure is that customer satisfaction increases thanks to the algorithms okay. Taught before so that we know is true. And, and it’s very obvious in our data that those customers, they stay and just happy because they rate the balls well, and they say, well, you’re doing a great job. Thank you. So.
50:10
Samuel
Just one last thing on the day of science stuff, are you cross-referencing profiles of people in order to potentially, I mean, the first thing that came into my mind, which is why I asked around while it was cars and suggestions is that if you have two data profiles that are very similar and you notice actually these two people, they’re on different sides of the country. They’re very similar, but this person really liked this one. We think that person’s gonna, so you just throw a wildcard in there and you just put in a wine that is maybe slight different, but is that ever something you do basically by cross-referencing the data or does, or do you silo every individual person’s profile independently?
50:55
Nicholas
Yeah. Every profile is independent and is particular in our database. And, but we use these clusters, right? We have 10 clusters in red wine, for instance. If two people belong in the same cluster, what might come out is that, of course the shipments might be similar, but since the shipments are not defined only by a cluster, but they’re defined bottled by Powell. Let’s say you have a thousand profiles in sizes in bold that’s means that you have a million different matching points, right? Every scene is really based on your personal taste profile. So, but maybe there’s another way to answer your question. That what comes into your boxes also by the algorithms is also defined by yourself? Because there is a, one of the features there I didn’t mention is you can tweak the algorithm. You can say, basically, I want more wines. I like, or more balances or more wines to discover it.
51:57
Nicholas
If should say more wines to discover its stretches. If you had two, exactly same profiles, right. Which doesn’t happen, but let’s say it would, and one says more wines. I like the other more wines to discover. That’s going to receive the same wines in the box. Obviously the second one is going to have more risky stuff, but it will also give us more information about the test once they rate them. So.
52:25
Chris
It’s a fascinating business model and I love how you’re a pro you’re applying the AI and the data science to an industry like wine, where you really are demystifying all of that process that sits underneath that. I think it’s really interesting. Why would like to cover before we have to wrap up, is this, isn’t your first startup, right? This is what number six. I think.
52:47
Nicholas
Number five, this.
52:48
Chris
Is number five. Okay.
52:50
Nicholas
So fast.
52:52
Chris
Jumping out.
52:53
Nicholas
I wonder what the number six is going to be. That’s number five.
52:57
Chris
Well, it’s on the cards. I think at the pace you seem to be going. So how did you get into this? What made you, what made you treat the teams that take this direction of, you’re not going to be an employee, you’re going to go and build startups and you’re going to get stuff off your ground, off the ground of your own. You’ve been very successful in it. How, just talk to us about how you got started doing that.
53:20
Nicholas
Yeah. Thank you. Well, when I got out of business school in Paris at that time, in the mid nineties there wasn’t, I mean, I don’t know if that one was called even startup or business angels were not, it was not an award around and now obviously everybody wants to build businesses, but I was one of the only issue of my year. Graduating then started a company because I wanted to be whatever I wanted to build companies. In the journey is you build companies as a pure breed, entrepreneurial like me, because you find problems around you. At some point you can’t sleep at night. We want to solve that problem. Right? The paddock club game at a moment where it’s in the middle of everything I did before I did e-commerce and travel business, I did the recommendation app. I did some data science, and this was a passion for a long time, but I was not in the wine industry at all.
54:20
Nicholas
I just think it’s better when you come from the outside, but you have, where you’re bringing in of new view, a disruptive view, of the market. I figured not only that the problem makes sense to be solved, but that the industry is really big and working the same way for decades. Right? If you think about, it’s only 5% of the market online in the U S and that is because of regulations. I think it’s also because innovation hasn’t really happened, basically the path which has moved online is mostly convenience. You have like a driver basically is sending wine up to you with not so much added value about, what is going to be picked. So, yeah. The answer precisely to your question, I think it’s when you start a venture like that as an entrepreneur, it’s a mix of things of what you’ve done before, what she will knowledge is, and market components where you figure well, might make sense that I spend the next five, 10 years on this.
55:22
Nicholas
Once you venture in your discover, of course, new things, like, the original idea was about wine, but I called it pack that club because I still don’t know. Maybe this is not going to be only about wine. The feedback from customer, I mean was tremendous. I can’t, I couldn’t ignore it when everybody’s telling me, oh, I want this for cheese and chocolate and whiskey and, whatever. So, so now it has become more like, okay, so we are, we’re continuing to make a better product and increase sales in wine. As soon as hopefully by the end of next year, one adventure as a marketplace with other premium foods and what appeared as well as in our smartphones taste is not there. Right. Everybody’s talking about the metaverse right now. Everybody’s talking about, virtual sings and virtual money and crypto and all these type of stuff, but, taste is the real world and it still needs to be, and data I’d like to taste.
56:25
Nicholas
I think it’s particularly interesting in 2021 to work on taste, it’s a frontier, which is like, there’s not much around there and it’s important.
56:37
Chris
Since this mist out of the metaverse, isn’t it? On that start-up point though, what’s harder. Start up number one, or start at number five.
56:48
Nicholas
That’s a super hard question.
56:50
Chris
It’s intentionally vague.
56:53
Nicholas
Yeah. Well, I, I would like to say startup number one is harder because you don’t know anything, but the reality is startup number five is harder because it’s a harder problem to solve with some way more ambitious problem that my startup one when I was 22 and also, you have more pressure about getting things, right? Because when you have 25 years of experience of building sings, et cetera, you somehow you’re supposed to know what you’re doing and have the right priorities, et cetera. Well, when you’re 22 and you start a company, the, your best guess is you’re gonna fail and it’s totally fine because you’re going to learn a lot with that. I didn’t fail my first one by the way, almost, but then we sold it in 1999,
57:46
Chris
Just in time, I guess then the.com.
57:49
Nicholas
Boom. There wasn’t a.com. My first.com was created in 99 and the travel industry, which is funny because the travel market was only 5% online at that time. Now it’s probably close to 100% and now the wine market, the food market’s only 5% a night. It’s kind of, you find these patterns it’s like now, and it’s supposed to move online 20% next February. There’s a really like big search awaiting next in the next few years. So.
58:18
Chris
When you’re talking about, you’re solving a different or more difficult problem for your fifth, when you were starting out at 22, did you have, like, I imagine from having spoken to you for an hour, that you were probably bubbling over with ideas, did you have to cherry pick which one you could actually do? With you’re young, you’re probably got limited capital. You’re trying to figure out just out of school, you’re trying to figure out how am I going to get this thing off the ground?
58:44
Nicholas
Yeah. Actually not, I, I ventured very quickly in the first ID, even though I wanted, I had a few other ideas, but the more I grew, the more ideas I have right now, I have like, I dunno, probably most of them are bad ideas probably, but I have like one idea every two days about, this and that business and not necessarily related to wine or whatever, since I’m doing right now, there are so many things to, you can solve or make better in this world. There’s so many great entrepreneurs doing it. So, no, I have way more difficulties of choice later on that in the early days. There’s one thing here is I wish when I started that this data and business angel world and everything in particularly the access to knowledge was as big a at the time as it is today for young entrepreneurs.
59:40
Nicholas
I mean, it’s amazing. They can get so much information out there on the internet and following, the best entrepreneurs, the best VCs, et cetera. We didn’t have that at that time. I had to figure it out yourself basically. So now it’s times have changed. It’s, it’s just grateful for younger entrepreneurs. It’s also great for, serial entrepreneurs like myself, because I also learn every day and I have new ideas every day. My God, if I wasn’t doing this, probably have a hard time to choose where I would start. But anyway,
01:00:18
Chris
Do you have any advice then for like either the person thinking about starting their own business or even the serial entrepreneur actually, can you advise a serial entrepreneurial.
01:00:30
Nicholas
Among entrepreneurs? We, we give each other advice and we listened to each other as much as possible. It’s learning as she is as is one of the main success of Silicon valley. That learning between that very open learning? Yeah. For younger entrepreneurs, there’s so much advice out there so will not like, be the typical ones and tell you, oh yeah, you should like go in and have a, this and that. I think the problem is because it’s so hype to build businesses and they read so many good stories. First of all, they need to be reminded. What you read on the internet about successes is only the 1% of success. The other thing is I see entrepreneurs which just go there because, there’s money or whatever, all they’re dreaming, it needs to remind them that not everyone is designed for that. You are entering a very hard journey.
01:01:27
Nicholas
Most of the time it’s, you need to be super resilient, have a lot of grit. It’s very hard out there. There’s only like a happy shoe, which suddenly, they are exactly the right timing, the right product. And, everybody’s sinks are over smart just because they had that. I mean, you need to be a certain level of smart obviously, but you need to have all these other components. And probably because it’s so high break. Now, there are many people who are trying to create a company and should do maybe something else. Again, there are cohorts of great entrepreneurs creating right now. It makes me very happy that, it continues to go for serial entrepreneurs. You know, what can I say? They know we all have different experiences, so would more be a conversation where I learn and they learn at the same time. I dunno. There’s like, it depends which field they’ve been in and yeah, but it.
01:02:19
Chris
Sounds like the sounds like the connections are key though. Right. Getting to speak to people, learning from experiences.
01:02:25
Nicholas
Oh yes. Oh yes. If you, if you are an entrepreneur and you just stay in your room and you don’t talk to people, it’s going to be much harder for you to get access to just the type of knowledge, the customer feedback, even in your mindset to product market fit. If you don’t talk to people and it’s going to be really hard on the other end, if you talk to the wrong people can be, people just like telling bad vibes and then not to do this and the wrong people can also be like talking only to your friends, which are just sharing your app and not giving you really good feedback about your product. You don’t see the, all the traps and things like that. You have to talk to many different people and ask for as much clear blunt advice as possible, no filters know as close to the truth as possible.
01:03:24
Nicholas
This is where you’re going to make progress. Obviously having a network it’s also important for business, for funding and everything else.
01:03:31
Chris
Okay. Good advice. I think we’ve covered a lot. We’ve covered an awful lot and I’m, I wish you every bit of luck with your, with the business. It sounds fantastic. I’m sure you’re going to learn an awful lot of stuff about wine.
01:03:42
Nicholas
Sure. I’m running always including about wine for sure. All.
01:03:47
Chris
Right. Great. Well, thank you for being on the show.
