What machine learning can teach us about marketplace fraud

Online marketplaces have reached ubiquity and, in turn, so has online marketplace fraud. The same frictionless, automated and remote experience that makes e-commerce and peer-to-peer marketplace sites so popular also makes them vulnerable to fraud that is very difficult to detect with the naked eye.

Enter: machine learning. Using a powerful combination of massive amounts of global data, sophisticated algorithms and intelligent analytics to parse that data into meaningful insights as well as human input from experienced fraud managers, we can quickly and accurately find those needles of fraud in marketplace haystacks.

By analyzing millions of transactions for marketplaces on four continents over the past year, we have uncovered some common attributes of devices used by fraudsters to conduct online fraud. These examples include:
? cleared browser cookies and referrer histories
? use of Windows machines, and
? surprisingly not using “private mode” or “Incognito mode” when visiting the marketplace

In addition to looking at fraudsters’ device trends, we also look at the behaviors and schemes employed by fraudsters in online marketplaces. Let’s take a closer look at two of the most common types of online marketplace fraud that we encounter:

1) Fake Profile Fraud - A fraudulent seller copies the profile of a legitimate seller in order to fool victims into buying something they’ll never receive or donating to fake charitable fundraisers.
2) Fake Buyer and Seller Closed Loop Account Fraud - Another kind of fraud involves multiple fake buyer and seller accounts created by the fraudster. The fake sellers pay the fake buyers for nonexistent items or services using stolen credit cards.

In both types of fraud above, it is often the marketplace that will pay the price for chargebacks. But there is another, perhaps greater, cost to this fraud: eroding consumer trust in using digital payments and online marketplaces overall.

How then can marketplaces and payment providers fight back against marketplace fraud? The answer may lie in automated fraud prevention. Machine learning can help marketplaces identify which transactions, profiles and accounts are most likely to be fraudulent.

Fraud Analysts can also help by continuously feeding fine-tuned rules and feedback into the fraud detection engines. There are some telltale signs to watch for like:

? Strange email addresses: It can be difficult to create many email addresses tied to fake accounts, so fraudsters sometimes resort to auto-generated or gibberish addresses like asdfjkasdf@gmail.com or emails that don’t relate to the username like bakerywhale122345@hotmail.com for a user named Tom Johnson.
? Multiple names, same profile photo: Often people in online marketplaces submit a photo to better connect with prospective buyers or sellers. So when a copycat con artist gets lazy and reuses the same photo for multiple accounts you can nab them. However, fraudsters have picked up on this and tend to spread photos across multiple marketplaces.
? Tip: Save an image, and then upload said images to Google Images to see if any photos match the one in question.

Online marketplaces represent a fast-growing and popular segment of e-commerce and have quickly become a huge source of digital payments transactions. Their rapid rise in popularity and global proliferation have led to fertile ground for fraudsters. Working with available technology and data, each of us has a role to play in stamping out the scourge of marketplace fraud. 

About Rahul Pangam

Rahul Pangam is co-founder and CEO of Simility. Prior to Simility, Rahul was a Director at Google where he led a global team of 200 that reduced ads fraud by 90%. Rahul is a fraud detection industry veteran having spent 6+ years at Google building teams responsible for fighting fraud and abuse on Google’s ads, local and social products. Prior to Google, Rahul was a Lead Engineer at General Electric (GE) working on GE’s smart grid software products. Rahul holds an MBA from the University of Michigan and a Master of Science degree in Electrical Engineering from Clemson University.

About Simility

Simility is a fraud prevention provider that combines machine learning with human analysis in a cloud platform that protects clients from sophisticated types of fraud. The founding team has a combined 27 years of experience fighting fraud at Google. Founded in 2014, Simility analyzes millions of transactions per week for customers on four continents as part of a limited beta release, and is backed by Accel Partners and Trinity Ventures.