The Role of AI in Fraud Prevention: Strengthening Payment Security in 2024

Humans can’t match computers when it comes to analysing data, which is why AI and machine learning are so valuable. They can identify and act on patterns far too complex for us to detect.

So far, machine learning has only been lightly used to tackle payment card fraud, but the need is clear. Fraud in Europe alone totals €1.8 billion a year, and prevention is getting more complicated as e-commerce grows, cross-border payments increase, and new digital payment methods emerge.

Let’s talk about leveraging AI to meet the global challenge of fraud.

Modern fraud environment

Payment card fraud is becoming more common and usually happens in one of two ways. Fraudsters either have the physical card for in-person transactions or just the card details, which they use for online purchases (CNP transactions). With the rise of e-commerce, CNP fraud has skyrocketed and now accounts for nearly 80% of all card fraud in Europe (according to the European Fraud Report).

Traditional fraud prevention relies on rules set by humans to flag suspicious transactions. This works to an extent, but it’s time-consuming, expensive, and results in many false positives, which frustrate both customers and merchants. Meanwhile, criminals use techniques like skimming, phishing, and data breaches to steal card details. These details are then sold on the dark web, which makes it harder to track how they were stolen or used.

E-commerce merchants face extra challenges when working with multiple payment providers. If one provider is breached, some cards may be compromised, but it’s difficult to pinpoint which ones. Blocking cards as a precaution is another avenue to false declines, which harms customer trust and loyalty.

Fraud’s impact on consumers, businesses, and beyond

Fraud affects not only finances but also people’s lives, especially the vulnerable. It worsens inequality, causes emotional and physical trauma, and leads to lost opportunities.

• Impact on consumers — Fraud leaves victims with financial and emotional damage. For vulnerable groups, it can deepen their struggles and increase their vulnerability.
• Business impact — Fraud creates unfair competition and drives legitimate businesses out of the market. It also increases operational costs for companies, including expenses for fraud detection, investigations, and program adjustments.
• Financial burden — The financial impact of fraud extends beyond immediate losses, with hidden costs tied to fraud detection, response efforts, and potential lost business opportunities.
• Reputation damage — When fraud is mishandled, it erodes customer trust and can severely damage a business’s reputation.
• Environmental and security risks — Fraud can lead to environmental harm, costly clean-up efforts, and may even fund criminal activity or terrorism.

Need for precision in fraud prevention

In the fight against fraud, businesses are trying out a new solution — AI. Its ability to detect suspicious activity with precision makes it a powerful tool that identifies fraudulent transactions more accurately while minimising false positives.

Effective fraud prevention depends on clear rules that help differentiate between real and fake transactions. These rules reduce the impact of fraud but also need to be fine-tuned regularly to avoid blocking legitimate transactions. The goal is to stay ahead of fraudsters, who constantly adapt their tactics.

A strong AI fraud prevention strategy uses several types of rules: velocity checks to flag suspiciously fast transactions, geographic checks to spot inconsistencies in location and limits on high transaction amounts. AI also helps by analysing user behaviour to predict normal patterns and identify any unusual activity. Verifying contact information, such as emails and phone numbers, is another key step to preventing fraud. To stay effective, these rules must be updated and refined over time.

Turning weaknesses into strengths

Fraudsters face several challenges in trying to exploit systems, and AI detection tools are designed to exploit these limitations.

Identifying common points of purchase

Fraud happens in two key stages: first, the payment card details are stolen, and second, those details are used for fraudulent transactions. This creates a limitation for fraudsters — they can’t commit fraud without having access to the card information. This opens up a critical opportunity for fraud prevention — predicting which cards are most likely to be compromised, and that allows businesses to stop fraud before it happens.

Traditionally, fraud teams try to identify the common point of purchase (CPP), the merchant where compromised cards were likely used. Once identified, businesses can either block all cards linked to that merchant or take the risk of letting some transactions go through. However, this approach mistakenly flags legitimate merchants that have high transaction volumes or overlapping customer behaviour.

AI improves this process by going through vast amounts of transaction data, not just a small subset of compromised cards. It identifies CPPs more accurately by considering multiple factors, like the merchant, time period, and transaction history. This reduces false positives and improves fraud detection, i.e., it creates a clearer overview of the risk. AI also continuously learns from historical data, so it adapts to new fraud patterns and improves prevention strategies.

Understanding merchant behaviour

Not all merchants or ATMs are good places to commit fraud. There are only certain locations where fraudulent transactions are likely to happen. Even though it’s simple to calculate the fraud risk of a particular merchant by looking at the number of fraudulent transactions compared to total transactions, this alone isn’t enough. However, when combined with other data, this information becomes much more useful.

Merchants provide a lot of useful information, much of which is easy to track. These merchant characteristics provide clues about whether a transaction is legitimate or suspicious. For example, AI can look at the average number of transactions a merchant processes each day or the usual size of those transactions. Informed by the range of amounts spent, it can identify patterns — e.g., seeing where 25% of purchases fall below or above a certain value.

Other important factors to track include how consistent a merchant’s transaction volume is. If there are sudden spikes in payments or big fluctuations, that could be a warning sign. On the other hand, steady transaction patterns indicate a trustworthy merchant. The length of time a merchant has been accepting payments is also a key factor. New merchants with little history are more likely to be fraudulent. That’s because some fraudsters set up fake businesses quickly to exploit cardholder data before getting caught.

Tracking cardholder activity

Fraudsters face another big challenge — they don’t know how a legitimate cardholder typically behaves when it comes to spending. They might try to make fraudulent transactions at various merchants, but they can’t predict which ones will look suspicious based on the cardholder’s past purchases. AI makes a real difference here by using the cardholder’s transaction history to assess the likelihood of fraud.

In simple terms, AI personalises fraud detection. By analysing a cardholder’s typical spending habits — like which merchants they usually buy from, how often they make purchases, and the amount they spend — AI can figure out if a new transaction fits with their usual behaviour. For example, a big purchase in a foreign currency might be normal for a business cardholder, but it could raise red flags for someone using a personal card.

AI goes even further by checking if the cardholder has bought from a particular merchant before. It can also predict whether they’re likely to make a purchase at a new merchant based on their history and the spending patterns of other similar cardholders. This is similar to how recommendation engines on sites like Netflix or Amazon work. Except in this case, AI is figuring out whether a transaction makes sense in the context of the cardholder’s past actions.

Detecting time-based anomalies

Fraudsters avoid detection by making smaller transactions quickly because large ones are easier to spot. They try to withdraw as much as they can before the cardholder notices anything unusual. To counter this, fraud detection systems track recent activity and compare it to normal behaviour in real-time.

Most fraud detection systems update their features regularly, but not all track changes instantly. By focusing on recent behaviour — such as the number of transactions made in the last few minutes or the total amount spent in a short time frame — systems can spot suspicious activity quickly. This is known as time trains, a way of capturing short-term behaviour patterns that might show something unusual.

For example, if a cardholder usually makes small, low-risk purchases but suddenly starts buying from high-risk merchants, that change stands out. Similarly, if a series of small purchases are made in a short period, it could be a sign of fraud.

Balancing security and customer experience

AI-driven fraud prevention tools sometimes create a bit of friction in the customer experience, especially when security measures like multi-factor authentication or extra identity verifications are involved. These additional steps feel like a hassle, particularly when customers are used to quick, streamlined transactions.

For instance, a customer might need to enter a one-time password or answer security questions after they’ve already submitted payment details, which slows things down. Another issue is that AI tools need to analyse transaction data or behaviour patterns, which causes minor delays in approvals.

In the end, though, the benefits of AI or any other advanced security system are clear. It might take a little longer or feel more involved, but adding some friction in transactions will ultimately make customers feel safer and more confident.

Smarter fraud prevention with optimised thresholds

The main goal of fraud prevention systems is to reduce false positives while stopping as much fraud as possible. The focus should be on the total value of fraud prevented rather than just the number of flagged transactions. Many systems use rules with set monetary limits — if a transaction looks suspicious but is under a certain amount, it won’t be blocked. The issue is that fraudsters can figure out these limits and adjust their actions accordingly.

That’s why fraud score systems are more effective. Instead of relying on fixed rules, a fraud score looks at both the likelihood of fraud and the transaction amount. For example, a transaction with a 42% chance of being fraudulent may have a different rejection threshold than one with a 38% or 46% chance. This makes the system more accurate and flexible.

This approach also makes it harder for fraudsters to predict or bypass the system because the thresholds depend on the fraud score. It also lets businesses set different rules for fraud prevention and alert generation, which helps human agents focus on the most suspicious transactions. And if the alert volume gets too high, the system can be easily adjusted to match the team’s capacity.

Conclusion

AI has incredible potential, but there are still a few challenges holding back wider adoption. The most important factor for many businesses is how easy it is to set up and use an AI system. Other top priorities include real-time scoring, automated feature creation, and making models easier to understand. Even with the current obstacles, the future of AI looks promising.

Financial institutions are actively developing new solutions to simplify their processes and improve customer experiences, which shows that the journey to fully embrace AI in payments is well underway.