Artificial intelligence has gone from an emerging technology to a foundational component of modern fraud prevention. As digital commerce volumes continue to grow and fraudsters increasingly leverage generative AI, machine learning, and automated attack tools, organizations are relying more heavily on AI fraud detection systems to identify suspicious activity in real time.
The challenge remains the same as it was several years ago: stopping fraud without creating friction for legitimate customers. However, the scale and complexity of today’s payments ecosystem make that balancing act even more important. As consumers increasingly expect instant, seamless transactions across cards, mobile wallets, and digital channels, AI fraud detection has become one of the most critical technologies supporting the global payments infrastructure.
By using artificial intelligence (AI), Visa Inc. helped issuers prevent an estimated $25 billion in annual fraud, the company announced on June 17. The company accomplished this using Visa Advanced Authorization (VAA), a comprehensive risk management tool that monitors transaction authorization on the Visa global network, VisaNet, in real time.
VAA evaluates every single transaction on VisaNet and helps issuers swiftly identify emerging fraud trends and patterns, allowing the issuers to respond promptly to instances of fraud, while approving legitimate transactions.
“One of the toughest challenges in payments is separating good transactions made by cardholders from bad ones attempted by fraudsters without adding friction to the process,” said Melissa McSherry, senior vice president and global head of Data Products and Solutions at Visa.
Speed is Key
The speed with which Visa can evaluate a transaction is crucial.
If the process is too slow (if there’s too much friction) and a payment is falsely declined, the affected cardholder is likely to just use a secondary payment card to complete the transaction, potentially a card issued by a competitor. In fact, 51 percent of cardholders who experienced a false decline simply used another card, according to a study.
Therefore, Visa Advanced Authorization is strikingly quick, with each transaction being assessed in about one millisecond. In that millisecond, the AI searches for indicators of fraud — looking for activities and patterns common in fraudulent transactions. Put another way, Visa’s technology allows financial institutions to approve legitimate purchases, and prevent fraudulent ones, at nearly the speed of light.
How It Works
Visa Advanced Authorization starts the moment a transaction is initiated by a merchant. As the hundreds of pieces of data from the transaction are sent over VisaNet, an artificial intelligence model analyzes the data for more than 500 unique risk attributes. These attributes can be thought of as clues that fraud may have occurred.
For example, the AI will look at what type of transaction it is, whether it’s being made in a store or online, with a contactless card or with a chip card. The AI will also determine whether the account associated with the card has been used at that store before. Even the time of day or the amount of money involved is considered by the algorithm. Advanced Authorization is robust enough that it can identify good transactions even when they are made by a new or infrequent shopper, which further helps reduce the rates of false declines.
After completing this analysis, the Advanced Authorization system will then generate a score which reflects the likelihood that the transaction is fraudulent. The scores range from 1 to 99, with 1 being the least risky and 99 the most risky.
Visa will then send the score to the accountholder’s financial institution, and the institution makes the determination of approving or rejecting the transaction. All this occurs in the blink of an eye.
The Size of the Problem
While each transaction can be assessed in a short amount of time, the amount of transactions in need of assessment have been skyrocketing. Over the past two decades, Visa’s transaction volume has increased by more than 1,000 percent; VisaNet processed more than 127 billion transactions in 2018 alone.
With billions of transactions being processed each year, stopping fraud is a major challenge. In fact, 55 percent of retailers cited fraud as their top payments-related challenge, according to a survey conducted by the National Retail Federation and Forrester Research.
Despite the scope of the problem, Visa’s AI has been largely successful. Even as the volume of transactions proliferated by 1,000 percent, the global fraud rate has declined by two-thirds, to less than 0.1 percent. This drop is made possible because VAA is widely used; more than 8,000 issuers in 129 countries are currently using the technology.
As more and more transactions go through VisaNet, and are subjected to Advanced Authorization’s algorithm, the model actually improves.
“One underappreciated aspect of supervised machine learning is that the model’s accuracy is increased as additional training data becomes available,” said Tim Sloane, VP of Payments Innovation at Mercator Advisory Group. “Given the scale of the Visa network, it almost certainly collects more transactions than its competitors.” In a way, the size of the problem actually helps create a possible solution.
However, many challenges remain. “The key issue for all networks isn’t just its ability to develop great machine learning models, it’s also the ability to manage and enhance the transactional data into effective training data,” cautioned Sloane.
To transform transactional data into training data, it must be tagged as either a good transaction or a fraudulent one, so the machine learning model can be trained and improved. “While that may sound easy to accomplish, it typically isn’t because the fraud is often not detected until days or weeks later,” explained Sloane.
How Visa Got Here
Part of Visa’s success in combatting fraud stems from the fact that the company has been using AI for a while now.
“Visa was the first payment network to apply neural network-based AI in 1993 to analyze the riskiness of transactions in real time, and the impact on fraud was immediate,” said McSherry. Prior to using cutting-edge technology, fraud detection was analog and consequently cumbersome.
For every transaction, for example, a cashier would have to search through a voluminous book of stolen cardholder account numbers to confirm that the card was not stolen. Another method consisted of the cashier dialing up a call center representative to verbally authorize the card. In either case, the process was slow.
In the years since 1993, Visa has been improving upon its fraud detection services. By incorporating biometric data and mobile location confirmation into the suite of fraud detection tools, Visa continues to innovate and improve the fraud prevention space.
However, Visa is not alone in using AI to combat fraud.
“Machine learning has greatly enhanced the ability to detect fraud and all of the major payment networks are applying this technology through a combination of internal R&D as well as through investments and acquisitions,” said Sloane.
The payments industry has changed significantly since AI fraud detection first became mainstream, but the underlying objective remains unchanged. Financial institutions, merchants, and payment networks must continue finding ways to distinguish legitimate transactions from fraudulent activity in milliseconds, all while maintaining a positive customer experience.
As fraud tactics evolve to incorporate generative AI, synthetic identities, and increasingly sophisticated social engineering schemes, the role of machine learning and real-time analytics will only grow. Organizations that can effectively combine transaction data, behavioral insights, and advanced ai fraud detection capabilities will be best positioned to protect customers while minimizing false declines and preserving trust in digital payments.
