Fraud and financial crime are pressing issues for governments, banks and consumers around the world right now. The sophisticated tools that financial criminals can deploy today have allowed crime to flourish.
At Featurespace, we believe fighting financial crime is a question of technology. The banks that invest in tools like machine learning for fraud detection today will gain themselves the upper hand on fraudsters.
Featurespace Subject Matter Expert, Paul Evans, explores how fraud detection in banking works today, what the role of technology is and how banks can protect themselves against the next generations of financial criminals.
What is fraud detection in banking?
Fraud detection in banking describes the tools and processes that banks use to monitor transactions and payments for suspicious activity. When a transaction or pattern of behavior throws up a red flag, the bank’s fraud team can intervene.
Banks have developed their fraud detection capabilities for years. Today, we see that most banks rely on previous-generation machine learning tools that are programmed to spot certain types of activities (e.g., large, aberrant transactions to unknown accounts).
As we will see in a moment, that approach to fraud detection has its limits.
Types of fraud in the banking sector
Fraudsters have a variety of financial scams they run on businesses and individual consumers. These are some of the most common types of fraud we see in the banking sector:
- Payment fraud. Payment fraud describes a variety of scams, including account takeovers (when someone gets access to a victim’s passwords and payment details to make fraudulent purchases) and authorized push payment fraud (when someone deploys social engineering to convince a victim to send them money).
- Card fraud. Card fraud happens when a fraudster either steals a person’s credit or debit card to make fraudulent purchases, or gets access to their credit card data and makes fraudulent card-not-present purchases.
- Merchant acquiring fraud. Merchant acquiring fraud describes the types of scams that target merchant acquirers, the intermediaries that connect the banking and commerce sectors. These scams include things like transaction laundering (when scammers use an acquirer’s gateway to process fraudulent transactions) and chargeback fraud (when scammers request repayment for a purchase they claim they never received).
How do banks detect fraud?
Banks monitor transactions and comb through transaction data to identify patterns of behavior or suspicious activity that might indicate fraud has taken place or is about to take place. Ideally, the bank’s fraud detection system will detect fraud before money can leave a customer’s account.
That fraud detection system is what monitors transactions for any unauthorized activity or access to sensitive data. Transaction monitoring relies on specific tools, techniques and strategies for detecting fraud, which include:
- Mining data with artificial intelligence to spot patterns in transaction data.
- Modeling customer behaviors with machine learning.
- Assessing the level of risk with statistical techniques like regression analyses and probability distributions.
The system then flags suspicious events for manual review.
How Featurespace can help with bank fraud detection
Here is where we return to the question of rules-based and simple machine learning (ML) tools.
As useful as rules and models have been in the past for banks, it’s not a technology that can keep pace with the sophistication of today’s financial criminals. As soon as the detection system gets updated with new rules and/or machine learning models, the criminals change their behavior to seek for a new way to bypass them. That’s how they remain a step ahead of older technologies.
Featurespace developed Adaptive Behavioral Analytics technology to solve this problem. Adaptive Behavioral Analytics studies the behaviors of bank customers to learn what kinds of transactions are normal and what should get flagged as abnormal. With Automated Deep Behavioral Networks, the next generation of fraud prevention, it offers an enhanced layer of protection from scams, account takeover, card and payments fraud.
The algorithms enable continuous learning, so they remain responsive to changes in how scammers operate as well as to changes in how customers behave. This allows the model to spot more instances of banking fraud while reducing the false positives that can frustrate customers.
The deployment was live within 92 days, and provided an immediate advancement to Contis’ existing fraud detection tools. Now, Contis can detect authorized push payment fraud in real time, which creates an important layer of fraud protection for its banking and payments clients.
Get in touch