about AI? Then don't miss the Premier AI Conference.
At GTC, choose from over 600 sessions on dozens of hot topics.
Get hands-on training, connect with experts from Google Stanford Amazon Apple Facebook Epic Microsoft MIT Pixar City BMW and hundreds more It all starts March 17th at the GPU Technology Conference in Silicon Valley.
To learn more, go to gputechconf.com. That's gputechconf.com.
Then save 20% on registration with code GMPODCAST.
That's all caps, GMPODCAST. We'll see you there.
Welcome to the NVIDIA AI Podcast. I'm your host, Noah Kravitz.
A quick note before today's episode. If you enjoy the podcast, please subscribe.
Did that sound earnest enough? Please subscribe.
Just search for the NVIDIA AI Podcast on CastBox, Apple Podcasts, or wherever you listen and hit the subscribe button.
We'd love your feedback too. Of course, you can give us a five-star review on any or all of those networks, but also tell us what you think.
Tell us who you'd like to hear interviewed and what topics you'd like us to explore.
You can drop us a line via email at AIpodcast at nvidia.com.
All right then, on to today's guest. AI, and all of computer science really, runs on data.
So let's start today's show with some data.
$95 billion. That's the projected size of the global athletic footwear market, sneakers to you and me, by the year 2025, according to Grandview Research.
$1 billion. That's the estimated value of the current resale market for sneakers, according to Business Insider, amongst other sources.
And finally... $50,000. That's the asking price, as of my recording this, for the Adidas 3D Runner Men's Size 14 on online sneaker marketplace GOAT.com.
With so much money on the line, it's hardly surprising that a robust fake sneaker industry has sprung up in recent years.
Here to talk sneakers, sneakerheads, and using AI to sniff out fakes and stamp out frauds is Michael Hall, Director of Data at Goat Group.
Goat Group is a Southern California-based sneaker marketplace that's a member of NVIDIA's Inception program for promising AI startups.
Michael, thanks for coming on the NVIDIA AI podcast.
Thanks for having me, Noah. So, and I'll call you Mike, we were talking offline, I was told to call you Mike, so of course I called you Michael three times already, because that's how I roll.
Let's start the stage here for the uninitiated.
What's a sneakerhead? And when and how did sneakers become such big business?
And then finally, it's a three-parter to get you going.
Why is the resale market in particular worth so much right now?
Yeah, great questions. To start off, a sneakerhead is what we usually affectionately call someone who's obsessed with sneakers.
They are our core users of the app and they appreciate not just the aesthetic value of sneakers, but also their monetary value.
I cherry picked, you know, I went on to goat.com.
And as you mentioned, there's also an app.
I went on to GOAT.com and searched price high to low.
And so I picked out the $50,000 Adidas, which is maybe an anomaly.
But sneakers on yours and other resale marketplaces routinely go into the high three and four figures, right?
Yeah, I mean, I was flabbergasted when I got to go at how big this market is.
I would consider myself a sneakerhead now, but less so when I joined. maybe two pairs of sneakers.
Uh, one of which was a Jordan one and the other one was the much lesser known Onitsuka.
And, uh, When I got here and saw that we had over 12 million users buying and selling over 35,000 different styles of sneakers, And the company has now grown to over 500 employees on both coasts.
And we've raised over $100 million in VC funding. to continue the mission of the company.
All of this just staggered me. Now, like you, a little bit of a, I'm sure, lesser than you, but kind of an amateur sneakerhead, or I've always been a fan, but, you know...
I've never built out a collection or anything, but I also had a pair of Jordan ones back in the day, really into them, followed them.
But like, How did this blow up so big? Obviously, we got to talk about Yeezys, but is it just that You know, when people start collecting, that kind of creates a marketplace and fuels the desire or what happened?
I think we should take it back to two different time periods.
The first one is really the 1980s with the Jordan one, right?
Michael Jordan completely turned the sneaker world on its head when he first introduced the Jordan 1.
And he already at that point had started kind of creating this mythos around sneakers that he wasn't allowed to wear on the court because it didn't have the right colors for the NBA regulations. and really started marketing these high and higher end sneakers.
But really the modern sneaker hub If you fast forward to 2005, started with the Pigeon Dunks.
These are the Jeff Staple shoes that people lined up for days outside of his shop for.
And where the police actually had to be called in to break up and people were bringing weapons.
The enthusiasm for these sneakers was so extreme that people from all walks of life were willing to spend potentially the legal viability of their future to try to acquire them.
Um, and then, yeah, now you fast forward to now and look at, you know, specifically probably two of the bigger brands at the moment, Kanye West, easy line, which is our bread and butter at Goat, as well as Nike's recent reinvention with Virgil Abloh and and the Off-White brand.
The scarcity mixed with just the aesthetic beauty of these sneakers really is a draw for people to spend you know, their disposable income and more.
Yeah. Is it the scarcity? I assume that these designer collaborations, the Off-Whites and the Pigeons, there's a new pigeon that's just coming out now, right?
They re-upped it? Yeah, they did. And it's hard to compare height between 2019 and 2005, but it's a it's a hotly desired sneaker and I think it's a beautiful thing.
Yeah, it's cool. But I promised I wouldn't get too off track talking about the sneakers themselves. assume when you mentioned scarcity, these are kind of limited edition runs, which is kind of what sets up this marketplace for the scarcity driving the resale potential.
Yeah, absolutely. And scarcity is half of it.
But also, due to the popularity of these shoes, Nike and Adidas and other brands have really focused on these retro restocks so that people can get, again, the shoes that Michael Jordan was talking about or showing off in the 1980s, that they remember and maybe had us as kids, you can get them again. with slight variance so that you can really buy into that nostalgia that brought you back to when you first encountered the pair of sneakers that you cared about.
And I mean, honestly, how the company started, you know, we're now the world's largest marketplace for buying and selling authentic sneakers.
And the impetus for the company in the first place was our co-founder, Daishan.
He had... a pair of Jordan 5 briefs when he was a kid.
And he'd been into sneakers when he came up with the idea for the company for probably 25 years.
And so he decided he wanted to recapture those Jordan 5 grapes that he'd had as a kid. and went online, went to the biggest marketplace for sneakers at the time, which was eBay, and was amazed at how many different offerings there were on eBay of just this one SKU in his size.
Turns out that that was actually a dangerous problem.
He, between all of them, tried to suss out which ones were real, which ones are fake, which is a pretty common story a few years ago. and finally decided on one of them, purchased it, and received the shoe.
When he got the shoe, it turns out it was completely fake.
The stitching was wrong. The Jumpman logo was off.
It didn't even smell like an actual TV shoot.
You laugh, but that's a, that can be a big tell, you know, speakers have a specific smell.
No, I've been there. I actually had a similar experience with the Air Max 95 Neons.
And I bought a pair off eBay, got them, and yeah.
They were nothing. They were felt. They were nothing like the originals.
So let's get into that. How does Goat Group work?
What does your website actually do? Yeah, so we started in 2015 and we have the mission to solve the problem of of counterfeit sneakers in the marketplace.
These are high and luxury items that people care a lot about.
The company in a nutshell, the basic model is we pioneered this ship to verify model.
We have a marketplace, we have buyers, we have sellers.
Once the buyer decides on the shoe they want to buy, we consider them matched up and the seller sends us the shoe and the buyer gives us the money.
And once we receive that shoe, it goes through a very rigorous authentication process.
We have manual authenticators who get the shoe in their hands, And these are domain experts.
These are people that can tell you 30 plus years worth of Jordans. what exactly the stitching on a Jordan 11 should look like between the patent leather and the regular leather and the sole of the shoe.
Every single detail of these shoes. And these are, and again, for the uninitiated, you know, this isn't just the, like we joked about like, I ordered a leather shoe and what I got was made out of cheap felt or didn't smell right.
These are like fine details that might be lost on... the average consumer, but for the connoisseur collector, this is where the devil lies in these details.
Absolutely. I mean, Nike now at this point is the most counterfeited brand in the world.
Over 10% of the sneakers that are sold online are counterfeits.
As you said, they're very, very, very close to the real thing.
And one of the things that I found out when I got here, and maybe I'm jumping ahead, is that when we started trying to tackle this automated authentication problem, It turns out that a lot of these fake sneakers are actually made in the exact same places with the exact same materials as the real sneakers.
Right. Yeah, get into that a little bit.
I saw an online documentary about this and went down a little bit of a rabbit hole myself.
But for the listeners, how does that happen?
You know, we kind of have to piece that together ourselves.
We obviously can't be on the warehouse floor when these are made.
But what we can say definitively is that a lot of the... approaches that we initially wanted to take.
But when I came in, you know, I came into this company with a pretty strong hubris, actually. thinking that I'd be able to come up with some kind of magic wand that these authenticators could point at the shoes and beep boop, it's real, it's fake.
I had a lot of things I wanted to try and almost all of them immediately failed, including microscopic pictures of the materials.
Like I said, the materials are the same. in a lot of cases.
Dimensions, size, weight, all of these features that we just needed to start collecting because no one had this data.
So to get an idea of the scale of the problem here, you mentioned 12 million users.
How many pairs of shoes are you guys authenticating or kind of getting and holding in escrow and shipping out to buyers? on a daily, monthly, yearly basis.
Yeah, we actually can't share that. That's not public information at this point.
Fair enough. Okay, that's fine. but enough that there's a problem to be dealt with.
Absolutely. From personal experience, one of the first shoes I bought when I got to the company You know, we have the benefit of actually just being able to go and pick up our shoes at the warehouse ourselves after they're done authenticating them.
And I was anticipating this pretty strongly.
It was one of my shoes that I'd always wanted.
It was in one top three. And I got the email the day it was here that they were fake shoes.
And This actually helped pretty significantly because the shoes that we were using to test the authentication process were Jordan ones at that point.
And so we were actually hunting for fake crushed me personally.
Give one, give one to the team. So understanding, you know, and, and for the listener, hopefully the stage is set here.
This is This is big business. This is shoes and it's fun and it's nostalgic and we're laughing and we're talking about it.
But there's a lot on the line here money-wise, let alone for the collector.
Like you just said, your first pair turned out to be fake and it was kind of crushing.
So a lot on the line monetarily and emotionally, personally here.
So- Understandable, you got to keep some secrets close to the vest, but how did you start using AI to try to solve the problem?
Yeah, I mean... Machine learning and AI is not new to the marketplace community, right?
These are problems that are sometimes really well researched and sometimes not.
We have a similar breadth of problems as other marketplaces.
This includes recommendations and discovery, automating work and operations flows, discovering fraud of buyers and sellers, and that added problem of authentication. of the shoes and so you know we really affect the streamlined process for the user at at every single step in the user flow, right?
You can picture the top of the funnel where we're trying to get people to be aware of GOAT and get them on the platform We help with marketing placement and attribution and testing and all of that.
You can picture the seller, now that we have as many shoes as we have, coming in to be listed on the platform, content moderation actually becomes a significant problem. not fun that we were equipped to deal with before.
So there's a lot of work there to be done around kind of hot dog, not hot dog and skew verification. we can train models regardless of whether it's a authentic or fake shoe to at least be able to identify the skew.
Right. then we can kind of bypass some of that manual content moderation in favor of automated.
And then on the buyer side, right, you have the normal discovery and personalization problems that you have on marketplaces.
Recommendations is becoming an increasing problem.
We have a unique audience at GOAT. Sneakerheads will be the first to tell you that they know exactly what they want when they come on our site.
Yes, indeed. They're not coming in to be introduced to some sneaker that didn't exist.
They know what's out there and they know what they want.
For the record, I want the undercover ex-Nike React Element 87.
I don't just want the straight Nike. I want the undercover version.
Yeah, of course. And we can table that and talk about that on the side.
Perfect. Yeah. So these users know exactly what they want when they come on the site.
But as we grow, we need to open this up to users that may be newer to the sneaker community.
And so recommendations and discovery is becoming an increasing problem there so that we can really expose them to the breadth of sneakers that are on our platform that they may be interested in.
And then on top of that, you add the more normal marketplace problems like credit card fraud and chargebacks. shipping routing, inventory management.
Once we merged with Flight Club, we've learned a lot from their consignment model, and we now have a massive inventory of sneakers in our warehouses already pre-authenticated, ready to be sent out to users.
And so inventory management between the flight club stores themselves, the GOAT warehouses, and any cold storage we may have, it becomes an increasing problem and an increasing cost.
Obviously that leads us finally to the authentication problem, which is I think probably the most interesting problem that we have to deal with and unique problem.
We're talking to Mike Hall. Mike is the Director of Data for Goat Group.
Goat Group is... an online marketplace for buying and reselling sneakers.
As Mike mentioned, they recently acquired Flight Club, a boutique sneaker shop that, I'm probably going to be wrong here, but started in New York City?
Is that right? That it did, yeah. We still have a flagship store on Broadway.
Excellent. I miss those days, but not in January.
I'm glad to be out here in California. And you guys are also in California down south in Culver City.
Yes. So let's back up a little bit because you mentioned being a little bit of a fledgling sneakerhead when you got to GOAT.
What were you doing before GOAT, and how did you get into computer science, machine learning, and all that good stuff?
Yeah, I come from a family of computer scientists, and there was really no doubt that I was going to go into computers myself.
I finished most of my grad work at UCLA about 10 years ago.
Yeah, 10 years ago now, 2009. And it was an amazing environment to be in there with, AI greats like Richard Korf and Adnan Darwish and these people that really from the 80s on made major contributions.
And as soon as I left there, I jumped into my first startup.
And that led to a series of startups, mostly around ad tech.
And then I moved into consumer apps, which eventually landed me at Tinder.
At Tinder, I led the early recommendations team.
So we decided on... the set of people to put in front of users for them to swipe on.
And when I got there, like I said, it was very early at Tinder.
When I got there, it was slightly better than random.
And we made huge advances in both getting users that were already doing okay on the platform, more matches, and then also trying to kind of increase the number of users that got at least one match every day, right?
We wanted to make sure that not only the users that were already doing okay were being more successful, but also that we could spread some of that success across the entire user base.
And so that was kind of the goal there. As a side note, I may steal Slightly Better Than Random as the title for my autobiography if I ever write one.
So I just want to mark that here on the record.
As long as you're okay with me putting it on my grade zone, too.
I think that works. That's a good thing.
Cool. Two years ago, I transitioned over to Goat.
I was the first data hire here, and they wanted me to hire up a team to solve you know, kind of all the data science related problems as well as data engineering analytics and kind of the rest.
And so I met with the co-founders here and really appreciated what they were doing.
I really appreciated their love for the product and for sneakers, but also just their desire to protect consumers and kind of create that streamlined process for the users that we're used to this a ridiculously chaotic process of trying to find authentic sneakers online.
So let's talk a little bit more about the authentication problem with sneakers.
So you spoke about the problem of authenticating sneakers and how many different sort of variations there are that come, you know, these different factors that weigh into how the fakes now are so close to the real thing, but they're not quite You know, again, without divulging any trade secrets, how are you going about the problem, solving that problem and finding success?
Yeah, so one of the benefits of the failures is that we now have the biggest set, as far as we know, the biggest set of captured data on authentic sneakers that anyone has.
We've been collecting data including photos from every possible angle of both the sneaker and the box it comes weight, who's touched it at every single step of the process.
We have all of this information and it's really helped.
But one of the things that makes this problem especially hard is the imbalance between the classes of inauthentic and authentic sneakers.
We have tons and tons and tons of examples of authentic sneakers, right?
We see them every day, they go through our warehouse and that covers the majority of SKUs in our warehouse.
However, even with some of the more counterfeited brands, people have stopped sending us inauthentic sneakers.
They've stopped sending us fakes because they know how good we are. by catching them, right?
And so we want to capture enough of a ground truth set for that fake class We're just not able to get enough samples to properly train a classifier.
So we really had to start going to more of an outlier detection, unsupervised learning approach.
We've really utilized human in the loop machine learning.
Focusing less on that magic wand to be able to say whether something's authentic or inauthentic. in peeling off traffic from the authenticators so that they know what our probability is that the shoe is inauthentic.
So they can go through a different set of steps for shoes that we see very frequently and we have a very good sense in automatic authentication. whether they're real or not, and really spend the majority of their time on the ones that we suspect are inauthentic.
It's fascinating. It's like you've gotten so good at the first, you know, it's the last mile problem in a way, right?
You've gotten so good at getting up to whatever percent.
That then getting to that last percentage is more difficult, but in part because the fraudsters, you know, know not to mess with you.
Yeah. Yeah, it's definitely declining returns, but we're still working hard on the problem.
So what is the or what are some of the most faked sneakers on the market right now?
Jordan 11s, I think, are the number one fake shoe on the market right now.
In a particular colorway or just all of them?
All colorways, but obviously the fakes tend to come in maybe two to three months after the initial release.
Right, right. Sometimes they're ahead of their game and sometimes if it's not hyped enough release, it won't pop until later.
And so we'll see the inauthentic sneakers, you know, once the hype level grows.
And for the person who's inspired by this podcast, they're looking at GOAT right now with their credit card in hand.
What's a good starting sneaker for the fledgling sneakerhead?
Well, that's kind of a personal question.
Sure. I think that my favorite starter shoes for my friends and for myself are...
Air Max 97s, I think are a great starter too.
It's not super chunky, it's pretty sleek, it's got the nice curves on the outside, and it's got infinite colorways. you can always find something that you're interested in.
I mentioned before the Onitsuka Tigers. That was actually my first sneaker that I bought post starting at GOAT was the original blue, red and white color way.
I can see it in my head. Yeah, they're beautiful sneakers and the crosshatches on them are fantastic.
And then obviously, if you want to jump into the slightly more hyped end.
Jordan 1 highs and lows are a good place to start because again, ton of colorways and instantly recognizable by anyone who's seen a sneaker.
Or the new Spider-Man movie. I was going to say, my older son has a poster in his room right now. of Spider-Man rocking the OG colorways of the ones.
My life is frustrated to no end that I won't stop mentioning Jordan 1's whenever we talk about Spider-Man now.
I'm also frustrated at our finances now that I have this addiction to sneakers.
I think my collection is over 40 now. and it's prominently displayed inside of our bedroom.
How many of those do you wear? I would say I rotate through about 25.
Okay. And are there some that are just purely collect, you know, kind of investment or collector's items that are... Do the kids still say on ice?
Or Deadstock is another brand we use. Yeah, I think my two favorites that I... where very infrequently are my Shawn Witherspoon Air Max 97 ones. was designed by a creator here in Los Angeles who won a contest to be able to create a shoe with with Nike and they're beautiful.
They're made out of corduroy. I love them.
And then the other ones are the Air Jordan one shattered backwards.
It's the orange colorway of Air Jordan. Air Jordan 1s.
They're way outside of my price range, but I did some off-hours work buying and selling sneakers myself to actually...
So with enough GOAT funds to be able to purchase it without breaking our budget.
I don't want to get you in trouble with the feds, but is that insider trading?
No comment. I hope not. Definitely not.
And, and, you know, it's, It's nice. We've released a product this year called Instant Ship, which is kind of what I was mentioning, our consigned product.
Mm-hmm. And what's extremely helpful is anything that you buy on Goat. you can actually put in what we call good storage.
So we'll just hang on to the shoe. It's already been authenticated.
And then you can either keep it there until you want to bring it out and wear it, or you can resell it. and it can be shipped out immediately instead of having to wait for the ship to verify model.
Right, right, right. So that's actually helped everyone be able to kind of turn over these shoes faster if your whole goal is to make money.
And we've seen these I kind of pictured these sneaker resellers before I knew better as kids that were making a little extra money to buy sneakers themselves that they couldn't afford. or individuals, but people are building many enterprises on our platform, right?
Like our top sellers have sold... I think they sold last year more than our very, very top sellers individually sold more than $2 million of sneakers.
Wow. And those same top sellers. sold more than collectively as much as $10 million, which is just staggering to me.
No, it really is. My son and I are both big Kendrick Lamar fans.
And last year, the tour of his his label TDE, I'm saying this more for the listeners than I'm sure you know about this, Mike, but his there, there, the championship tour rolled through Oakland.
And it was sponsored or he sponsored whatever the official relationship is by Nike.
And so they had a pop-up shop in Oakland with some limited edition Nike merchandise for the tour.
And I went down and waited in line to get him a hoodie and was standing in line next to somebody who works at a sneaker boutique in San Francisco.
And so we got to talking about it and that gave me a very brief glimpse between that and then they had the Nike app Our city didn't qualify, but I guess there were stops on the tour where like during the concert, a push notification went out to the app that like, you can buy the Kendrick shoe right now.
And now it's just kind of, you know, my moment to what you were talking about of like, this is big industry.
You know, this is, This is serious stuff on the line here.
It is. Last question before I let you go.
And again, apologies to the data scientists out there, but it's a sneaker question, not a data question.
No, I'll follow up with one more industry question, but the newly released, I believe they're called the Nike Adapt BB, the self-lacing shoes.
Do you have a take on these? I haven't held them in my hands yet, so it's hard for me to give a personal opinion.
But the hype is growing around them for sure.
So purely as an investor, might I want to try to get a pair and put them into goat storage?
I would not guide you away from that. That's a very well-worded answer.
So looking ahead here, whatever you can speak to, whether it's GOAT or the industry at large or just the problem of... the diminishing returns of authentication around fake goods.
Where do you see all of this headed? You know, the work that you're doing and where Goat's going and the industry –
How is machine learning and the related technologies going to play a part in this, you know, five years down the line or further out?
I can't dig straight into our roadmap, but I can say discovery and personalization is huge on our radar.
As I said, we're trying to welcome people who are who are new to the sneaker community and maybe don't know exactly what they want. machine learning is at the core of discovery personalization and really making sure that we're showing the right shoes to the right people. and my team is going to tackle a big portion of that with the rest of our engineering team and product.
Then on top of that, just really global expansion.
We have, like I said, over 12 million users, but a big portion of them are in the US.
We're trying to broaden the market inside of the US and continue to get sneakerheads to use our platform as much as we can and provide great services for them.
But really, we have a lot of fans outside of the US that we have not been servicing as well as inside of the US.
And so we'd like to put a focus on international pretty strongly and specifically you know china's a massive market for us Australia, the UK, these are all markets that people are spending It's spending quite a bit of money to be able to use our platform on top of the money they're already spending on our platform.
So we want to provide a better service for them moving forward.
Excellent. Anything that supports sneakerheads, I'm all for.
Mike Hall, thank you so much. Thank you so much for taking the time to talk with us.
Obviously GOAT.com is the website. You mentioned an app, you mentioned Flight Club.
If people want to check out the sneakers that are for sale or possibly learn a little bit more about the work you're doing, where should they go look?
I would recommend if you're in Los Angeles, New York, or Miami, go to the fly club stores in those three cities.
Get your hands on the shoes. Turn them around, look at the quality, look at the stitching, look at the colors, see what you like.
Look at the things that no one can get their hands on that are in the kind of Locked down cases that weren't even released.
And then beyond that, download the app. browse.
We're trying to improve, as I said, discovery and personalization every day.
And there's something for everyone on there.
Excellent. Well, best of luck to you, Mike.
And again, thanks for coming on the podcast.
Thanks so much, Noah. Thank you. Eager to learn about AI?
Then don't miss the Premier AI Conference.
At GTC, choose from over 600 sessions on dozens of hot topics.
Get hands on training. Connect with experts from Google, Stanford, Amazon, Apple, Facebook, Epic, Microsoft, MIT, Pixar, Citi, BMW, and hundreds more.
It all starts March 17th at the GPU Technology Conference in Silicon Valley.
To learn more, go to gputechconf.com. That's gputechconf.com.
Then save 20% on registration with code GMPODCAST.
That's all caps, GMPODCAST. We'll see you there.
