Questions? Feedback? powered by Olark live chat software

Simility exposed fraudsters in India who ordered products online, promised cash on delivery, then returned more than 90% of their orders.

If Simility can detect fraud as specialized as this,

there’s no limit to what we can do for your company.

A NEW KIND OF MAIL AND DELIVERY FRAUD

What is delivery fraud a.k.a. package scamming? As in many emerging economies, Indian consumers fully embrace online commerce, but without using credit cards. Customers pay in cash when their shipments are delivered.  A large e-commerce company in India had an alarming rate of returns from a new kind of delivery fraud by brick-and-mortar businesses using cash on delivery as a loophole. Traditional fraud detection was useless to catch this scheme since they rely on payment information to detect fraud.

The fraudster businesses ordered hundreds of products from the victim’s website to be delivered on a daily basis.  Meanwhile, if customers came into their store asking for an out-of-stock product, they were told it would be in stock later that day. Then the fraudsters paid the delivery person in cash for the small fraction of products they had pre-sold to customers, while returning the vast majority of unsold products without paying for them at the cost of the e-commerce company, thus completing the delivery fraud cycle.

TRADITIONAL METHODS FAIL

The e-commerce company was well aware this type of fraud was occurring, but was powerless to stop it since its existing fraud detection vendor was reliant on credit card information to identify unusual ordering patterns. Since that didn’t apply, it tried limiting orders to four per day for each shipping address.

EXPLOITING AN ECOMMERCE LOOPHOLE

The fraudsters quickly found a loophole. They wrote scripts to automatically generate new accounts with fake delivery addresses. The addresses created by these scripts were all fake but very close to the fraudster’s real address. Although the addresses were fake, they were real enough to put the delivery driver in the same neighborhood as the fraudster’s business. When the “lost” driver called the business for directions, the fraudster would direct him to the real address a few blocks away.

GRAPHICAL ANALYSIS FINDS THE PATTERN

The e-commerce company knew they had an extremely high return rate but couldn’t detect a pattern, so they contacted Simility analysts. Using Simility’s powerful graphical analyzer on the e-commerce company’s order return data, they correlated the order information with the shipping address.

Interesting patterns emerged. Many accounts that reached the four order per day limit could be grouped by similar addresses. Fake street names were sometimes models of cars or the building address number would change by increments of three. Clearly it was the work of computer scripts written to create bulk orders and evade fraud detection.

DEVICE RECON: IDENTIFYING THE TRUE SOURCE

Simility’s proprietary Device Recon detects when a user accesses a website from the same device even if he tries to hide his identity by creating multiple accounts or altering browser settings. When Simility’s analysts looked at the Device Recon data, it confirmed their suspicions. Many of the bulk orders originated from the same computers even though the fraudsters tried to mask that fact by assuming hundreds of different fake identities.

HUMAN ANALYSIS & THE RULE BUILDER

Using Simility’s Rule Builder, the company quickly defined each discrete fraudulent pattern and added them to their detection model. Accounts that reached the 4 order-limit and email addresses made of random-character strings are two examples that highlighted suspicious behavior.

It could have taken weeks to write new rules manually and test them with traditional fraud detection. Simility’s software gave the e-commerce company the power to outline these fraud patterns in plain English and test their hypotheses in minutes.  Then, using full end-to-end integration, Simility immediately began to detect the new fraudulent activity and block perpetrators from completing orders on the company’s website and prevent package scams from occurring.

SIMILITY’S MACHINE LEARNING: CONTINUAL ADAPTATION

Fraud detection is an arms race. Once a particular pattern is blocked, fraudsters will look for a loophole to circumvent it. Simility’s machine learning engine constantly evolves as the fraudsters’ strategies change, picking up on fraudulent correlations that aren’t discernible by human analysis. Other fraud detection solutions offer a rudimentary form of machine learning, but can’t adapt to fluctuating behavior patterns of large-scale e-commerce environments.

Simility’s platform creates a new machine learning model for every behavioral pattern it detects, completely customizing it for that type of fraud. Decision trees and neural networks are infused into the fraud detection dashboard. The analyst’s experience is supplemented by the machine learning engine working in the background, helping to rapidly and accurately inform all of their decisions.

END TO END FRAUD DETECTION WITH FULL INTEGRATION

Once the e-commerce company brought in Simility to detect this unusual fraud pattern, it deployed Simility to detect other types of fraud. New models are created and implemented as quickly as the company pinpoints undesirable customer behavior it wants to block.

The company even started using Simility to detect good customers and speed them through their checkout process to provide a better user experience. Ultimately, Simility’s platform became an integral part of their business model, allowing the company to grow into new markets and industries without fear of fraud.