How a fast-growing payment processor uses AI to prevent payment fraud

Online payment processor Adyen uses AI and machine learning to reduce online payment fraud by accessing customer data.

Adyen is a listed Dutch company that is assists business in accepting payments. It started off focusing on online payment systems but also moved to mobile and point-of-sale systems. Currently, Adyen assists merchants with accepting payments from consumers across the world using various payment methods: credit cards, debit cards, bank transfer, real time-time transfers and other methods. Ayen’s services include acting as a payment gateway (capturing and transferring the payment data from customer to merchants bank[1]), payment service provider (connecting merchants to the broader financial system to allow payments[2]) and offers risk management tools[3].

The rise of digital payments has also led to a sizable increase in consumer fraud. E-commerce retail transactions have been growing at a rate of 12% CAGR for the last five years, with overall retail spending increasing by 4% [4]. The rise of digital payments has also led to credit card fraud increasing at staggering amounts – from 2012 to 2013, fraud in the US increased by 35% to $7Bn. The increase in fraud is due to the rise of “Card Not Present” transactions. Fraud has increased as merchants do not have customer’s swiping a card on a POS terminal in a brick-and-mortar shop. Instead, merchants have to trust customer entering card information on a payment page, leading to hackers and fraudsters intercepting the details or sending incorrect information to the merchant.

Payment processors such as Adyen act as the middlemen between the customer and merchants and ensure that transactions are completed securely. Tasks include validating customers credit card details (and encrypting card details, enabling data to be processed securely), accessing the information on whether a customer account has enough funds for a given transaction.

Understanding the level of risk involved in online businesses has led many traditional merchant acquirers and payment processors to decline suspicious transactions that can be thought to be fraud. This, however, leads to companies losing out on revenue – a 2019 report by Merchant Risk Council estimated that merchants on average decline 2.5% due to fraud, and most of these transactions were actually due to fraud. This leads to businesses losing out millions in revenue. The key to a better fraud management system would then require tools that better fit online companies and the risks they face. Due to the number of many indicators that could signal fraud – Adyen and other payment processors have started to use AI and machine learning to predict online payment fraud.

Fraud detection was Adyen’s first use of AI. It started by building its algorithms internally, allowing the company to use many data points that payment processors have access to understand better the likeliness of transactions being fraudulent. Adyen also uses AI when reviewing a transaction that has been flagged as possible fraud. Adyen’s use of automated review has reduced the time of transaction reviews by over 30%[5]. Adyen has also built a new product, “RevenueProtect”, that allows merchants to have a global view of possible fraudulent transactions and customize activities that may lead to higher fraud cases. The merchant can then quickly build intelligence rules to identify fraudsters. Adyen’s risk engine will use machine learning to lean and optimize risk checks in real-time – without contacting them.


Case study:

Adyen upgraded OLX (one of the largest online marketplaces in the world) payment infrastructure. After only eight weeks, they led to a 2.6% increase in authorized transactions (previously, those transactions would’ve incorrectly be attributed to fraud[6]). Adyen developed a payment processing tool that would use machine learning technology alongside other intelligent data usages. Adyen started by analyzing the payment information base and building an algorithm that would understand each step of the payment cycle on OLX’s system. The algorithm was updated to use location, email, average ticket size, card data, shopping cart products etc. The algorithm was also fine-tuned to bring in a new set of rules and was trained against the old anti-fraud tool to check how effective it was. Initially, 10% of sales would go to the new anti-fraud payment profit. Each week after more transactions were processed, the new tool became better and eventually, up to 50% of all transactions were moving to the new displaced the older anti-fraud tool. Ligia Pires, the Trust and Safety Manager of OLX Brazil, commented that Adyen allows OLX to approve more valid financial transactions and still have a robust anti-fraud system blocking illegitimate transactions.




[4] Harvard Business School Case 8-516-027 “Apple Pay” by Sunil Gupta, Shelle Santana, Margaret L Rodriguez




Zebra Medical Vision: transforming patient care through AI

Leave a comment