Catching big fraud with small data
With fraud prevention, many banks emphasize how their systems leverage “big data” to find patterns of unusual activity by using data analytics to process through millions of transactions and pinpoint fraudulent activity. While big data has many advantages in fraud detection, it has also alienated some banks and other financial services institutions that think they are too small to be part of this data revolution. They should think again.
If big data describes how companies leverage massive volumes of data, then the industry needs a term to describe the alternative – let’s call it “small data.” How do banks leverage smaller volumes of data, particularly in areas where the data is new and not well defined? One example would be the emerging patterns of fraud on Apple Pay, which launched late last year. Some banks report rates of fraud in excess of 600 basis points, or 60 times the level of traditional credit card fraud.
Here are five ways banks can unlock the power of their own data – no matter the size – to stop big fraud:
Participate in a fraud consortium. Pooling data through a consortium has been an effective fraud prevention approach proven to reduce fraud in card portfolios and mortgage originations. Whether it’s check fraud, Automated Clearing House (ACH) fraud, wire transfer fraud or anti-money laundering (AML) risk, banks can pool their limited data to share fraud and risk patterns with each other. For example, in the mortgage industry, individual lenders didn’t have robust databases of known fraudulent loans to help them prevent origination fraud. By contributing data to a mortgage fraud consortium, a pooled database of 65 million loans was created for the benefit of each contributing lender. Highly effective predictive fraud models were then built based on the data these originators amassed and shared.
Leverage scorecards as an analytical technique. Scorecards are a powerful, simple technique built using sparse or limited data. Scorecards are made up of scenarios that are typically risky. An analyst assigns weight to each scenario based on a combination of limited data analysis and fraud expertise allowing the data to be used together to assess the risk. This may be a relatively simple analytical technique but can create big results.
Deploy a rules-based system. In the absence of any data, the next best thing is leveraging the experience of people on the front lines. Banks that have no data but a lot of experience fighting fraud can operationalize that knowledge through fraud rules. Leveraging knowledge of experts can create business rules that prevent huge amounts of fraud.
Leverage analytical science techniques. Small data can present challenges to analytic scientists; however, a variety of analytical techniques can be used to improve results. Fraud typically reveals itself as an anomaly to normal behavior and trained models will often make the offending transactions stand out like a sore thumb. Scientists can use creative techniques, including proxy tags, such as returned checks or reversed wire transfers, when robust fraud data is not available.
Link analysis. Sometimes flat data just doesn’t reveal any particularly strong insight. In those cases, the bank needs to look at the data in a new way. For example link analysis, whereby data is joined to other sources of internal data, can reveal a “network fraud” where fraudsters use common information such as phone numbers or IP addresses. By looking at small data through a different lens, a company can sometimes reveal hidden fraud.
Big data is inherent in banking origination strategic goals, but let’s not dismiss the power of small data. Big insights can be gleaned from the smallest sources of data. Whether it’s pooling data with other lenders or using techniques that unlock that data’s value in a productive way, small data is the next big thing.