Digital payments – securing the gateways

Rupesh Vashist | Associate Director | mail me |	Dean Friedman | Partner | mail me |
| KPMG Southern Africa |

Payment gateways play a critical role in transaction processing. The transaction messages are processed through them while they broker the approval of the transactions from the source (generally customer or merchant) to operators (generally issuer).

Traditional fraud detection methods operate with predefined dimensions. Fraud administrators set fraud rules based on specific parameters, such as transaction time, volume, velocity and merchant risk profiles, however, these rules are inherently limited, as they exclude dimensions not explicitly defined in the fraud logic.

Identifying fraud

With the advent of payment methods and the increase in the channels, fraudsters keep trying to learn newer ways to exploit the vulnerabilities in the system. AI serves as a great tool to identify fraud when the volume is high and there are multiple dimensions in the fraud. AI is particularly effective against certain types of fraud where traditional rule-based methods fall short.

Artificial Intelligence (AI) and Machine Learning (ML) enhance fraud detection by identifying and analysing anomalies across all possible dimensions in real time to prevent frauds or offline for early detection of frauds. Unlike static rule-based approaches, through ML, the defined models dynamically learn transaction patterns and adapts to emerging fraud trends. For instance, a traditional fraud detection system might flag high-velocity transactions as fraudulent based on the defined rule. However, AI can refine this assessment by factoring in additional context, such as whether the transaction occurs during a peak shopping period or a promotional event, where higher velocity may be expected and is therefore legitimate.

For example, AI can help to identify unusual login behaviour or anomalies using telemetry in case of fraudsters taking over customer accounts which is a common fraud type in digital payments. In cases of synthetic identity fraud, AI is very effective in identifying anomalies in identity documents by comparing them with captive, open-source and subscribed data. We have seen some very effective use cases of AI in money laundering, where ML was effective in detecting unusual transaction flows and layering techniques, identifying shell merchants and hidden links between accounts, and flagging suspicious transaction behaviours that human analysts might miss.

Management systems

Fraud management systems have been leveraging various forms of AI and ML techniques for quite some time. Practitioners of fraud management, including vendors, service providers and personnel in Financial Service Providers (FSPs), consistently confirm that AI-driven fraud detection capabilities far surpass those of traditional analytics-based approaches. AI-based routines not only detect fraud in real-time but also provide early warning indicators by identifying subtle patterns and anomalies across vast transactional datasets.

By continuously learning from transaction behaviours, AI not only improves fraud detection accuracy but also reduces false positives. This ensures that human operators focus on genuine threats rather than spending resources on reviewing transactions that pose no real risk. The result is a more adaptive, efficient and intelligent fraud prevention system that strengthens payment gateway security.

While it is impossible to quantify the exact losses prevented due to AI-driven fraud detection, tangible evidence exists in the form of increased detection and reduced chargeback rates across industries reducing losses and chargeback cases. Payment schemes such as Mastercard and Visa are increasingly using AI for fraud detection and minimising disputes between stakeholders.

Following are some notable instances:

• AI-powered fraud detection models used by major payment networks like Visa and Mastercard have reduced fraudulent transactions. According to a press-release on the Visa website, Visa reports that its AI-powered risk scoring system helps prevent approximately $25 billion in annual fraud losses.^[1]
• Mastercard is using AI in its SafetyNet alert to identify suspicious transactions and merchants. According to a press release on the Mastercard website, Mastercard is using generative AI to…