Using Big Data to Prevent Ecommerce Fraud
A recent report estimated the fraud cost for online retailers to be $3.5 billion or 0.9 percent of online revenue. To put it in different words, an online retailer loses, on average, $9,000 to fraud for every $1 million in revenue.
This is a significant amount of fraud that, ideally, should be a part of the retailer’s profits instead of being added to the cost of doing business. Unfortunately, the fraud rate increases if the retailer supports mobile commerce or ships orders to international customers.
Every retailer strives for zero fraud losses but that goal has been elusive. With the rise of Big Data in the last couple of years — see “Understanding Big Data For Ecommerce,” our previous article — the goal is now within reach. Big Data can prevent fraud by identifying insights that have not been easily accessible to online retailers in the past. Before presenting how Big Data can help, I’ll first review the key types of fraud that impacts online retailers.
Credit card fraud. This is the most common fraud that impacts online retailers. Since the retailer is unable to see the card physically, thieves can more easily use a fake or stolen credit card. Many retailers use the tools provided by credit card companies like address verification service, which matches the card number with the billing address on the card, or the three-or-four digit code on the card, called the card verification number or card security code, to minimize fraud. But these methods are not foolproof.
Return fraud. This takes many different forms. The most common are returning merchandize after using it or claiming the product was not delivered and selling it through other channels. A few online retailers encourage customers to order more products and return those they don’t want. Some customers assume that every retailer supports this policy, leading to higher return fraud. Online retailers that sell high-value items or have a brick-and-mortar store are more prone to this type of fraud. Even after taking measures like charging restocking fees for returns, getting customer signatures to confirm order delivery, and monitoring customers who are likely to commit fraud, return fraud continues.
Identity fraud. Identity fraud involves stealing a consumer’s personal information: A fraudster logs in, using a customer’s credentials, shops on the site, and ships the goods to a different location. It impacts retailers heavily as, in most cases, it leads to chargebacks and the retailers have to pay the chargeback penalty unless they can prove the transaction was a fraud. The fraud may involve using a customer’s personal information to get new credit cards. Identity fraud could result in larges losses. Having the right rules in place to check ordering patterns of each customer, his frequency of ordering, and his order shipping addresses is important. But those steps do not eliminate it entirely.
Preventing Fraud with Big Data
The use of Big Data can combat fraud, in three main ways.
Analyze all the data. In the past, retailers used a sample or a subset of their data for fraud analysis. It took too much time and money to use the complete data set. With Big Data, not only can all the data be analyzed for fraud, but new data sources can also be introduced. Analyzing the full data set leads to several benefits: (a) Reviewing all transactions based on the defined fraud rules; (b) Identifying new fraud patterns that get added to the growing list of fraud rules; (c) Minimizing false positives to avoid losing revenue and turning away customers.
For example, to minimize return fraud, big data can determine if the product was actually delivered by analyzing data streams from social networks, conducting image analyses, and partnering with third parties like eBay and Craigslist to access their listings. All this data can be aggregated and analyzed using Big Data tools like Hadoop.
Detect fraud in real-time. Live transactions are combined with data from other sources, like existing data warehouses, to detect fraud in real-time. This can prevent credit card fraud where the transaction is screened against a set of pre-defined fraud rules as part of the credit card authorization. This includes combining site data with data from customer’s social feed, the geo data from customer’s smartphone apps, purchase history, and web logs. Since all this is done in real-time, fraudulent authorizations are declined. Moreover, Big Data solutions enable analyzing historical transactions in the previous weeks, months, or years to identify new fraud patterns, which can be automatically added to the set of fraud rules and used as part of the real-time authorization process. Visa recently reported identifying $2 billion in potential annual incremental fraud opportunities by using a robust fraud management system that looks at 500 different aspects of a transaction to detect and prevent fraud.
Another way real-time analysis prevents fraud is by processing streaming data from sensors attached to high-value items to transmit their location. This minimizes return fraud, as the retailer now knows exactly when the item was delivered to the customer.
Use visual analytics. Other Big Data tools offer the capability to visually analyze data and derive insights, even though the data may be coming from different sources. Retailers can use these tools to identify regions, products, and customers that have a higher fraud rate based on historical analysis. This identifies the areas where time and money should be invested to minimize fraud. Visualization also reduces manual efforts to reviewing every order. The reports can graphically depict the probability of fraud for each order transaction and connect to email or SMS alerts for escalation, as needed.