Financial fraud occurs every day, accumulating in trillions of dollars in losses each year. In some segments, fraud has reached the highest levels on record. It affects more organizations than ever across all industries – including finance, retail, healthcare, government and utilities. A survey by PwC revealed that 49% of businesses experienced fraud over the previous two years.
What about the other 51% of organizations? Did they avoid falling victim to fraud? Or did they just not know it was occurring?
Why Fraud Is Difficult to Detect
Legacy fraud management platforms generate too many false positive alerts – more than can be investigated. This situation allows malicious activities to go undetected. And often, these platforms produce evidence of activity only after fraud has taken place.
A shortcoming of these fraud platforms is that data fed into their analytics engines are siloed and lack context. This prevents organizations from accurately assessing risk across the entire environment.
For example, suppose an organization suspects that its accounts payable department is making fraudulent payments to bogus vendors. If the company focuses exclusively on its payments data to detect suspicious transactions, it will miss the opportunity to dig into the behaviors of employees authorized to make payments. By analyzing this behavior, the company can determine whether a malicious insider or an external hacker (who stole an employee’s legitimate credentials) created a fake account to which they may be sending unsanctioned payments.
Another flaw of legacy fraud platforms is their reliance on rules to make a judgment on the legitimacy of transactions. The problem is that rules in these signature-based security tools are centered on already known threats. They don’t handle unknown or unpredictable threats that occur for the first time.
Here is a use case that illustrates the limitations of rule sets. A fund manager for a wealth management firm exploits the rule about investing a maximum of $100,000 daily in high-risk stocks. He learned he can skirt the rule and avoid detection by investing $99,999 each day. Technically, this isn’t fraudulent. But, it’s still risky behavior that management would presumably want to know about. A rules-based tool would not pick up on this.
A third problem with conventional fraud detection tools is that they cannot correlate activities from different channels. For example, banking transactions take place on mobile devices, on computers, in person with a credit or debit card, at ATMs and via face-to-face interactions at local branches. A skilled cyber criminal can create fraudulent transactions on one system that will not be correlated with activities on other systems in instances where the bank’s fraud platform is unable to link data in incompatible systems.
Fraud Analytics for Fraud Prevention
Fortunately, recent advances in technologies from big data to machine learning have coalesced in new fraud detection solutions. These solutions can detect anomalous behaviors in real time. From there they can provide accurate risk assessments so that mitigations can be instantly triggered.
The main factors required for machine learning-based fraud detection include:
Big data store: The first requirement is an architecture that can scale to millions of data points over time. A big data system should support large and varied data sets (both structured and unstructured) and enable your data analytics to uncover hidden patterns, unknown correlations and trends.
Data sources: The processing engine should be able to ingest data from all available sources. This includes online and offline, regardless of format. More data sources will result in better correlations and insights.
Data linkage: The data must link to a specific identity. That identity could be an accountant, a sales rep, a partner and so on. Likewise, the identity could be an entity, such as a point-of-sale device or a laptop. Linkage is essential to the creation of a baseline of behavior for each identity so that new activities can be compared to the baseline to look for anomalies.
A machine learning model: Once a big data store, data sources and data linkage are established, you need to set up artificial intelligence (AI) and models that analyze data feeds in real-time, establish baselines and automatically risk score activity. This process uses sophisticated algorithms to look for patterns, adjust risk scores and make better future decisions based on analyzed data.
Gurucul Fraud Analytics
At Gurucul we designed our fraud detection solution to align with all key components listed above. Gurucul Fraud Analytics relies on machine learning with open analytics to create customer specific fraud models, in addition to a large out-of-the-box models library.
Gurucul Fraud Analytics ingests large volumes of data generated by users and entities across multiple channels, both on-premise and in the cloud. It identifies potentially fraudulent behavior that spans time, place, devices, access and transaction actions.
The Gurucul Fraud Analytics risk engine continuously scores this activity against historical behavior, peer groups, and third-party intelligence to generate risk prioritized alerts. The risk score can automate remediation responses by enforcing information security policies or making real-time decisions to prevent fraud before it occurs.
Criminals and hackers already use advanced technologies, including machine learning, to harvest information and perform fraud at machine speed. To keep pace with attackers, organizations need to enhance legacy rules-based fraud detection and prevention. It’s time to embrace new approaches that use data science to process multidimensional sources of information in ways humans cannot.