Identity data is a readily available commodity on underground cybercrime sites and can be used “as is” in fake applications or stitched together with made-up details in synthetic fraud. Some applicants may even use their own personal information in first-party fraud attempts, and in other cases ‘mules’ may be encouraged or coerced into applying on behalf of the scammers.
As one customer, a German-based retail bank recently revealed, traditional rules-based approaches are increasingly underpowered in the face of ever-changing fraud patterns. This is where machine learning-powered and cloud-based solutions are particularly effective.

Hiding in data

A client lender came to us with a challenge: try to find fraud hidden among two years’ worth of anonymised customer applications for auto financing—amounting to roughly two million forms. The client had been using detection rules written by its in-house team. However, in practice it was found that the rules were difficult to maintain as the fraudsters adapted their tactics to stay hidden. The bank was also concerned about false positives (alerting on non-fraudulent applications) creating extra customer friction, as well as false negatives (failing to alert on fraudulent applications).
The rules ultimately created a major resource burden on the bank’s fraud and investigations team, and crucially failed to detect incidents of fraud that came to light further downstream.

Uncovering fraud

We leveraged two key technologies to uncover fraudulent activity amongst the anonymised data:

• Machine learning, a sub-field of AI that is tailor-made for these kinds of problems. By learning patterns of non-fraudulent behaviour it is then able to spot the tell-tale trends that deviate from these patterns, indicating potential fraud.
• Network analysis, which is designed to detect suspicious inter-connections between attributes of different application forms, and therefore expose collusive behaviours between applicants with each other and with the car dealers. For example, multiple applications may be submitted by the same gang under different names but with the same contact details.

There is a wealth of data in car financing application forms that you can mine for insight in this way. From an applicant’s geographic and demographic risk factors, to financial data such as size of loan amount and vehicle purchase price, and more. We also trawled through dealership details to check for recurring/unusual patterns, and the possibility that these businesses are aware of/facilitating fraud themselves.

From raw data to added value

Our proof-of-concept - from raw data to delivering value to the bank - was delivered in just three months. We extracted multiple insights from the data indicating detections of fraudulent activity, including evidence that some applicants and dealers were engaged in collusive behaviour involving shared personal identity details and duplicate vehicle identity numbers.
Our focus wasn’t just on the fraudulent applications, but also at the bigger picture. Application fraud is not only a major cause of financial loss in itself for banks, but also a money laundering predicate offence. After all, what better way to clean the dirty money obtained in financing fraud than selling the car anonymously and pocketing the cash? Thanks to network analysis capabilities, we were able to uncover connections between individuals to better unmask those fraud groups engaged in money laundering.
As a highly regulated crime that costs the global economy trillions, banks have major anti-money laundering (AML) compliance obligations to meet. And the bottom line for our banking customers: lower fraud losses, reduced reputational risk and happy regulators.