Online retailer G2A is looking to artificial intelligence to identify fraudulent transactions on its site, using a combination of algorithmic risk assessment and human screening to spot buyers who aren’t who they seem to be.
Maciej Kuc, G2A’s communications director, says the company is able to gather a lot of customer data before they ever provide an email address to the site – plenty of identifying information is available as soon as someone visits G2A.
“Wherever you go to do any kind of [online] commerce, you provide us with a lot of data,” Kuc said. “Where you’re from, how you react, what you do on the website. After a few seconds, we already know if you’re a high risk.”
Feeding detailed user data into tested algorithms gives G2A a pretty decent idea of when fraudulent transactions are likely to happen, and anything that looks suspicious automatically gets flagged for a human staffer to review. The staffer can then verify whether the user’s purchase or browsing history matches what they’re providing on the purchase page, or check for other known indicators of stolen credit card numbers or identity theft, for example.
Now, though, Kuc says G2A are teaching AI to sort through these risk assessments to write algorithms of its own. With AI’s capacity for deep learning and fast iteration, he says it’s possible to make the risk assessment and flagging process faster and more reliable.
“The more data an AI reads, the better it gets,” Kuc said. “And everything happens at the same time.”
AI can pull more data into the risk assessment framework than existing algorithms currently use, and Kuc says it will ask for help when it needs to. “If it has a problem with something, it will look for help. That’s why a human is needed to teach the AI.”
Payment providers such as Visa and Mastercard require that no more than 2% of all transactions on an e-commerce site can be fraudulent, otherwise they will not work with said site. That would obviously cripple the site’s business, so such sites are quick to reject any transaction if it looks even slightly suspicious, and can end up rejecting legitimate orders. These ‘false positives’ can be a major problem in e-commerce today. Kuc says G2A’s goal is to reject fewer than 1% of its transactions on suspicion of fraud (G2A says that instances of confirmed fraud are well below this figure already) and that AI will make that possible.
While that takes care of G2A’s relationship with its customers, what then of its sellers? After all, headlines about G2A have most often stemmed from the fact that it is a key marketplace, trading in the ‘grey’ market of unused Steam keys.
Kuc says that’s a different proposition. While the site has millions of buyers, he says there are something on the order of 400,000 sellers they do business with. That’s still a lot of suppliers, but it’s not the kind of number that requires AI to handle. And so, G2A generally still verifies the sellers who can list on its marketplace ‘by hand,’ so to speak. For that side, Kuc says the site has dedicated staff who look at business documentation submitted by sellers and apply company guidelines on whether to work with them.
“When you’re a seller, you provide us with a lot of data, a lot of proof on how you acquire the keys,” he said. “Where did they come from, where do you come from, what’s your company history, et cetera, et cetera.”
Kuc emphasized that G2A is stringent on security, in part because it has to be.
“We’re not just a marketplace, we’re a marketplace for digital goods,” he said. “If you do that, you need to be very, very good at it.”
G2A has recently added physical goods to its catalog of digital keys, so you can now find gaming-related apparel and accessories there as well.