Amazon.com calls on SAS for fraud detection

In operation since July 1995, Amazon.com today offers "Earth's Biggest Selection" of nearly everything. Amazon.com has 35 million customers and lists millions of unique items in categories such as electronics, kitchen and houseware, books, music, DVDs, videos, photography equipment, toys, software, computer and video games, tools and hardware, outdoor living and wireless products.

Amazon.com operates four international Web sites: www.amazon.fr, www.amazon.co.uk, www.amazon.de and www.amazon.co.jp. It also operates the Internet Movie Database (www.imdb.com), the Web's comprehensive and authoritative source of information on more than 250 000 movies and entertainment titles and 1 million cast and crew members dating from the birth of film in 1891 to the present.

An early commitment to minimising the risk of fraud led Amazon.com to choose SAS as the foundation of its fraud detection system. According to Jaya Kolhatkar, Amazon.com's director of fraud detection: "Our implementation of SAS has greatly reduced the cases of fraud on our site - and in the first six months of SAS implementation our fraud rates were cut by 50%."

Kolhatkar explains why Amazon.com chose SAS for fraud detection: "SAS has the best toolkit for the kind of analysis we wanted to do as part of our fraud detection process. No other vendor offers as comprehensive a range of tools - decision trees, neural networks and so on. It saves resources to have all of the functionality you need in one tool rather than having to prepare the data differently for different niche tools. SAS can also access all of the data that we need to analyse. Finally, SAS has been very proactive in responding to our requests."

For Amazon.com, as for all retailers that sell goods remotely by Internet, phone or mail order, credit card fraud is the most prevalent type of fraudulent activity. Detecting and preventing this activity becomes a priority because retailers -- rather than the cardholder's bank - bear the financial responsibility for this type of fraud. Perpetrators targeting online retailers use the same techniques that they might apply to traditional retailers. However, argues Kolhatkar, the fact that they leave a data audit trail with every click makes it somewhat easier to catch them. Some of the typical "cons" include:

* Dumpster-diving - Perpetrators look through trash cans to find credit card receipts and use the information from them.

* Double-swipes - In a store, the card is swiped twice. Data from the second swipe is used for fraudulent purchases.

* Creation of new credit card numbers - Perpetrators determine which banks, usually smaller ones, don't match credit card numbers immediately, meaning that the retailer will not be able to authorise the purchase with the bank before shipping the goods. The fraud perpetrator creates credit card numbers that are similar to the bank's credit card numbers, knowing that when they use the card to purchase goods fraudulently, the purchase is unlikely to be authorised immediately by the bank.

"Fraudsters generally follow similar patterns of behaviour," says Kolhatkar. "That makes it easier to detect fraud because you can look for corresponding patterns in transaction and customer data. For example, fraudsters tend to purchase goods that they can dispose of easily on the grey market, such as electronic goods. Obviously, they do not have the goods shipped to the same address that is used for billing purposes so an order not shipped to the billing address might be an indication. They'll also tend to use the fastest possible shipping method. Clearly, any one or even combination of these features does not mean that fraud has definitely taken place but, combined with other indicators, these would be the kinds of pointers that we would follow up on."

Amazon.com uses SAS to analyse the behavioural patterns of fraudsters and build predictive scores that indicate the likelihood of fraudulent behaviour having occurred. "We run these scores against the customer database," says Kolhatkar. "We then use SAS to prioritise the results. Obviously, we have to investigate a case of potential fraud very thoroughly before beginning legal action, so we prioritise the results of running the fraud scores and begin with the highest priority cases. We also do all of our reporting on fraud - which cases we are pursuing and their status - in SAS."