Fraud continues to be a problem for consumers and the banking industry. According to Javelin Strategy & Research in 2017, 16.7m people were victims of identity fraud and they lost an estimated $16.8bn.
The Consumer Sentinel Network, maintained by the Federal Trade Commission (FTC), found that of the 2.7m identity theft and fraud reports received in 2017, 1.1m were fraud-related, costing consumers almost $905m. The median loss in these cases was $429.
Imposter scams were the most reported type of fraud and accounted for $328m in losses. The group also reported that credit card fraud was the most reported incident to the Consumer Sentinel Network, with 133,000 reports. Bank fraud accounted for 6.4% of the identity theft frauds.
Identity theft and fraud complaints, 2014-2017. Source: Federal Trade Commission, Consumer Sentinel Network.
The problem is not just the volume of incidents. Fraud schemes are constantly evolving and organizations need to keep up with these changes. For the banking industry, this can be a challenging task. Not only are financial institutions a preferred target, they also absorb the losses in terms of revenue and consumer confidence when a fraud schemes goes undetected.
For example, in a recent case in Canada, a man publicly blamed a bank when his father fell prey to a Malaysia-based romance fraud scam that drained him of his $732,000 in life savings. Even though bank staff had asked questions about the wire transfers and probably followed policies, the bank is perceived as a negligent party in the eye of the public.
Types of bank fraud and role of technology
There are many bank fraud schemes including card not present (CNP), counterfeit, lost/stolen card, account takeover, internal and check kiting (bust out) fraud. Technology can go a long way in reducing some of these fraud types.
For example, the credit card industry saw a reduction of counterfeit fraud with the introduction of chip-enabled cards. However, account takeover and identification fraud, where accounts in someone else’s name is opened and used to buy goods or take out loans, continues to defraud billions from unsuspecting consumers.
The key is to implement new technology to combat fraud continuously. Financial institutions are increasingly turning to AI-driven models to augment their rules-based systems to transform their fraud prevention programs. This includes implementing advanced anomaly-detection models and transaction monitoring, with machine learning and robotic process automation.
Case study: Using anomaly detection for check kiting
Check kiting is where a scammer takes advantage of the float to withdraw non-existent funds in a bank account. There are a number of possible indicators for kiting including a large number of check deposits, accounts with large proportion of uncleared cash by the paying bank and deposits through multiple bank branches.
It is a multi-million dollar problem that can be difficult to detect especially with accounts with normal check writing and depositing activities. However, financial institutions of any size can use their data and anomaly-detection models to find transactions that may be indicative of check kiting.
The first step is to create the “ground truth”. This is where we create a baseline, or the model that establishes what is expected, or normal behaviour for an institution’s client base.
To create the ground truth, the system takes in all the supporting transactions and starts grouping them by characteristics. Any anomalies are isolated for review – are they “normal”, are these cases of “Not Sufficient Funds (NSF)” or are they “suspicious or provide indications of fraud”?
The process does take several iterations but once you have identified the valid and NSF transactions, the system can quickly help fraud departments identify the transactions that are potential cases of check kiting.
AI-models create “expected behaviour” profiles and identify entities/transactions that are suspicious and require further investigation. Source: Alessa
The next step is to investigate these incidents and to classify whether there is an indication of fraud or not. The concluded information is then fed back into the model and is used to identify future suspicious transactions.
This feedback model ensures that the technology is adapting with changes in consumer behaviour, the addition of new products and services and new fraud schemes. It also helps to reduce the number of false positives that can weigh down fraud investigators.
Finding cases of stolen card data
Theft of personal, credit and debit card data is frequent and pervasive. The impact is not only felt by individuals but also the final institutions that have to absorb added risks and costs.
To give an idea of the magnitude of the problem, here are just a few of the incidents where personal data, including credit and debit card data, ended up in the wrong hands:
The worst corporate hacks of all time. Source: Bloomberg
AI-based methodologies are also useful to detect out-of-pattern purchases that may be indication of stolen card data.
There are the usual red flags, like an individual is based in Washington but purchases are being made in the United Kingdom, but there are also many hidden patterns that are not easily to catch with the human eye. Technology can comb through the millions of transactions to identify those hidden patterns and isolate those incidents that require further attention.
Recently, CaseWare RCM worked with a customer who wanted to use their transaction data to identify anomalies in their customers’ purchasing patterns. To start the process, we took a snapshot of their user base and used algorithms to cluster individuals by their usage pattern.
From there, subsequent transactions are compared to the usage pattern and assigned a risk score. If they fall very far out of the user range, an alert is created for further investigation.
In one case, the model identified an instance where a fraudster was using the credit card in a different time zone to make illegal purchases.
The process of taking millions of transactions and identifying suspicious activities. Source: Alessa
Technology is only part of the solution for reducing bank fraud. Evolving policies and procedures have to be in place to reduce the risk. Here are some strategies to address some of the various types of fraud.
Preventing bust out fraud
- Implement a five business day clearing period
- Identify check payments versus electronic payments and frequency of deposits by customers
Preventing account takeover fraud
- Alert when there is an account level change (authorized user, address, phone number, card request)
- Alert when there are changes in spending pattern
Preventing identity fraud
- Verify identification
- Screen new customers against sanctions lists
- Review consumer alerts
- Review discrepancies between information from bureau and application
- Validate KYC (know your customer) information
- Call to verify employer
Preventing lost/stolen fraud
- Detect spending pattern changes (low to high volume)
- Look at small monetary value transactions that could be indication of credit card testing
- Look for aggressive transactions at unattended merchants (gas stations, or PayPass)
Preventing internal fraud
- Prohibit emails from being sent outside the bank with card or customer information
- Review activities of top performers to look for irregular practices
- Follow-up with inconsistencies in employee resumes
Preventing bank fraud is deliberate activity that requires continuous update of technology, policies and procedures. On the technology side, AI-based models are no longer something just for tier 1 institutions and allow smaller organizations to enhance their existing rules-based internal controls and identify previously undetected and evolving fraud schemes.