How GOAT uses machine learning, cloud to get you the greatest sneakers
It takes a lot of machine learning and computer vision to ensure that a pair of high-end sneakers is authentic.
Just ask GOAT. GOAT is the largest sneaker marketplace and specializes in selling authentic goods. Specifically, GOAT provides buyers and sellers of sneakers an authenticity guarantee with a “ship to verify” model.
GOAT, which has both e-commerce and physical retail locations, has 400 employees and 60 of them are engineers with 7 data scientists. We caught up with GOAT CTO Andy Shin to talk about machine learning, cloud computing and how to manage a team to focus on hitting key performance indicators. Here are some of the key takeaways.
The authenticity issue. Shin said GOAT, which employs 400 people, specializes in the “ship to verify model.” Shin said:
Nike is the world’s most counterfeited sneaker. If you go online, there’s a coin flip’s chance that the shoes you’re looking at is fake. And what our service provides is trust and authenticity. Sellers list sneakers for sale. Buyers find that sneaker at the right price and they purchase it.
And the experience that the buyer has is just like any other e-commerce website. They have no idea who’s selling that sneaker to them. But we’re in the middle. And what we provide is safety. And when that shoe is purchased, the seller ships it to us to verify. And we verify that product the same day using our sophisticated machine learning and computer vision, and our industry knowledge. And release the funds to the seller. And then we ship that product to the buyer to purchase.
Using computer vision and data to verify authenticity. Shin said GOAT uses computer vision and a bevy of Amazon Web Services tools to verify authenticity. Shin explained:
We use a lot of different heuristics and data points to verify whether the shoe is authentic or not. We see the most sneakers in the world. And we are the leader in data collection around sneakers. So every shoe that comes in, we’re collecting all the data points for it. So we actually, we know what fakes look like. And we know what real shoes look like. Every single sneaker has multiple data points that we focus on.
In a nutshell, GOAT is taking in physical attributes of shoes and replicate them digitally. Things like the suppleness of the sole, hardness of certain rubbers and seams can tip off fakes.
Building vs. buying machine learning and algorithms. Shin said GOAT’s machine learning and algorithms are proprietary. AWS is primarily used for compute infrastructure to run those algorithms. “We also leverage machine learning in our business. So we control the whole process from the moment that a buyer wants a product, to the moment they receive it. We control the whole supply chain,” said Shin. Machine learning is used throughout the process to optimize warehouse management so GOAT knows the most efficient way to distribute shoes.
What the cloud brings? Shin said what AWS and the cloud has enabled GOAT to do is focus on the machine learning behind distribution, authenticity and customer experience. GOAT would have had to build its own data centers and computing infrastructure and invest in the staff to run it. “What cloud computing really allows us to do is focus on building value,” said Shin.
Physical retailing and machine learning. GOAT has physical retail stores called Flight Club and machine learning also plays a role in the experience. Shin said retailers need to innovate with experiences that can’t be replicated online. GOAT uses machine learning to figure out how to merchandise sneakers. Using sensors, GOAT can track what sneakers are being picked up, dwell time and how a consumer works through a buying decision. Shin explained:
Flight Club is much like a Foot Locker, in that there’s a wall of shoes. Except we have 5,000 shoes in a retail store in New York City or Los Angeles. That’s a lot of stuff to merchandise, right? So how do we know what to put at eye-level? We leverage machine learning. And we know for example how long someone’s picked up the shoe. We know how long what they’ve done with it, and how long. And when they put it down. We also know which products have been requested. Whether or not we have the size for it. How long it took for that customer to get that product and try it on. So we leverage all these data points in order to create a great experience for our customers.
That data is collected and utilized for a weekly merchandising refresh.
Managing a team. A year ago, GOAT had 20 engineers. Today, the company has 60. AWS has enabled GOAT to focus on its core KPIs since it doesn’t have to do infrastructure. Now that the engineering team has tripled in size, Shin has retooled the structure via “pods.” Shin said:
These pods are essentially KPI driven. They are key performance indicator driven pods that focus on certain aspects of our business. I’ll give you an example. So our customers have high expectations for when they get a product. And so we have a pod focused on time of delivery. How do we get people product faster, and close the gap between when they purchase and when they actually receive the product?
So that’s a KPI driven pod that focuses on delivery. And what we really like to do is provide autonomy for these teams. We give them a KPI, and it’s really up to these teams to figure out how they’re going to reach this milestone, or to reach this metric. And what AWS has allowed us to do is to move to micro-services away from our monolithic application.
Shin added that the pods need to stay small and nimble. GOAT has 10 KPI-focused pods with 6 engineers each. These pods will evolve over time to focus on both the buy and sell side of a transaction.