Fintech companies, like PayPal, are using machine learning to prevent fraud. It's only a matter of time until AI consumes the finance industry.
Money and technology have a key thing in common: people like both of them fast and easy. That’s why Fintech innovations make it easier for consumers to connect with financial services. The demand for fast and frictionless transactions has led to an increase in real-time access to new credit, mobile wallets, and instant delivery of goods. Heck, Amazon is even bypassing in-store checkouts all together with Amazon Go.
But the improved speed and convenience have led to an increase in fraud. Online fraud rose by 33% in 2016, and 74 percent of businesses were victims of payment fraud. Every $100 of fraud cost businesses $240, according to LexisNexis’ True Cost of Fraud report.
The massive number of transactions coupled with sophisticated criminal tactics has many financial institutions turning to artificial intelligence to fight fraud. According to a PwC report, 30 percent of large financial institutions and 46 percent of the largest Fintech companies are investing in AI.
With millions of transactions taking place every day, AI helps evaluate behavior patterns and other data points to validate transactions in real-time. AI in fraud prevention is nothing new, as companies such as PayPal have been using the technology for years. But it’s becoming more prevalent, and will only increase in the future. Let’s take a look at how the technology is improving fraud prevention.
The Financial Industry’s Problem with Fraud
The financial industry has unique cybersecurity challenges. Not only is it a target because money is directly involved, but the number of transactions along with potential targets—consumers, merchants, and financial institutions—makes security more complex and important.
Fraud tactics range from personal identity theft to malware, to phishing emails, to fictitious accounts. Financial systems are interconnected, linking buyers, sellers, service providers, and banks, and criminals attack the weakest link to infiltrate systems and make fraudulent purchases or claims.
Fraudsters’ attacks have evolved, using distributed networks, big data, and the dark web to detect vulnerabilities. They also mimic good customer behavior to game the system. The problem with traditional protection methods is they are slow-learning and can’t keep up with changing fraud patterns, and they have trouble moving at the speed of transactions, which happens in seconds.
The AI Approach
Organizations need powerful solutions that react in real-time and can learn patterns quickly to recognize fraudulent behavior, which is what AI can do. It can analyze more data without making a tradeoff in latency.
Most organizations use rules and reputation lists as the primary fraud protection method. A rule is an encoded statement used to detect fraudulent behavior. For example, an institution may have a rule that states, if a customer adds five cards to an account in less than 24 hours, flag the account. Similarly, reputation lists are a list of specific IPs, devices, and other characteristics that are considered fraudulent.
However, cybercriminals can easily circumvent these methods by making hundreds of accounts to learn rules and reputation lists and work around them.
Machine learning, which is a subset of AI, uses AI computer systems to autonomously learn, predict, and act without preset rules. It learns from data instead of encoded rulesets and reputation lists. It looks at all features of accounts and transactions instead of just a few features established in rules and lists.
There are primarily two types of machine learning: supervised machine learning and unsupervised machine learning.
In supervised machine learning, a model is fed historical data classified as either fraudulent or non-fraudulent. The data is used to train the machine learning algorithm, which will then be able to recognize new accounts or activity as fraudulent based on previous incidences.
Unsupervised machine learning doesn’t make use of classified data, instead it processes large amounts of data and recognizes anomalous behavior. It’s adept at finding new attack patterns and identifying all the accounts associated with an attack.
Both models have their shortcomings. Supervised machine learning requires human input and only finds fraud similar to previous attacks. Unsupervised machine learning has cracked fewer domain problems and is not as effective at stopping individual fraudsters with low-volume attacks.
However, using the different models together can help organizations both combat existing fraud patterns and spot new methods. PayPal, for example, uses multiple types of machine learning, and as a result, has a revenue fraud rate of just 0.32 percent, while the industry average is 1.32 percent.
Putting It All Together
Overall, machine learning helps make real-time decisions in the fast-paced financial environment, process data faster, continuously learn from new transactions, and make more accurate decisions, which reduces false positives (when a consumer transaction is mistakenly identified as fraudulent).
More companies are turning to the emerging technology, including MasterCard which recently announced its AI security platform. Challenges still remain, as unsupervised machine learning is difficult to build in-house. But as the Fintech sector expands and the market continues to demand seamless and speedy transactions, AI will play a large role in preventing fraud.