AI in Fraud
An Increasingly Complex Problem
Pedro Branco, Closer Consulting
The quotation “money makes the world go round” was first used, most probably, in the musical “Cabaret”, which was written in the 1960s. The play was a melancholy one. In one of the songs, the female lead actress expresses her wish for love, and the male protagonist replies with this line in the song.
Alan Greenspan, a former president of the Federal Reserve System, once said that “We´d always thought that if you wanted to cripple the US economy, you´d take out the payments systems. Banks would be forced to fall back on inefficient physical transfers of money. The level of economic activity across the country could drop like a rock”.
In fact, and if we think dispassionately about any type of process, whether it's a medical consultation, a contract, a lawsuit, or any other relationship between entities, we realize that it culminates, in most cases, with a financial transaction.
So most likely, the most important system we have today, worldwide, is the payments system. Trust between countries, institutions, companies, and people depends on the trust that the payment system induces in all players.
In a post-covid world, in which we've been facing the increasing adoption of digital payments, electronic transactions, the dematerialization of money, and the rising use of apps and marketplaces, what is the greatest threat to payment systems and consequently to trust in general?
Most probably fraud.
Last year only, online shopping grew 100% compared to the pre-pandemic period, while online payments fraud losses are expected to exceed 206 billion dollars over the next five years. This enormous shift to digital provides an all-new playground for fraudsters and an incremental risk to financial and payments systems.
Fraud detection systems tend to rely on examining millions of transactions that exceed some preset rules or patterns, identifying specific types of behaviors, or predicting transaction scoring based on past transactions behavior. However, while the volume of transactions increases, fraud has become more complex as fraudsters become increasingly sophisticated, making the role of banks and payment processors much more complicated. While in the past, the most common type of fraud was credit card cloning or card skimming, the implementation of EMV chips and strong authentication (multi-factor authentication) forced fraudsters to change and to become increasingly experts in the payments’ business, bringing more (and faster) innovation into financial crime. Nowadays, malware, social engineering, e-commerce fraud, and impersonation scam are the new risks.
The question is how to deal with this expanding threatening effect while millions of transactions are processed per day?
AI is the answer.
The massive use of Machine Learning techniques applied to anomaly detection and patterns of attack, in the identification and fraudulent transactions blocking - capable of ensuring a memory in the face of past attempts -, are essential to provide the levels of protection and responsiveness required. Especially when it comes to analyzing millions of transactions every second, decision-making is measured in milliseconds: a virtually impossible task without using Machine Learning algorithms, capable of processing and analyzing the huge volumes of data generated every second.
When it comes to applying Machine Learning to fraud detection and blocking, data is the secret. The greater the amount of data available to analyze, ensuring adequate balancing between genuine and fraudulent transactions, the greater the ability to develop and train models with greater accuracy.
By leveraging the volumes of data generated for each transaction, it is now possible to ensure that we have ML models for fraud detection and blocking trained to deliver past fraud memory. These types of features that look back at the past are important in cases where fraudsters reuse past attack patterns. So that’s the second major challenge: use the right feature engineering. And in fraud, that’s a bigger challenge because it requires a vast domain and knowledge of the data and business rules.
Data Science and Machine Learning are not an exact science. They require data, a lot of data, and it's pre-processing. They need a deep process of feature engineering. Training and exhaustive tuning. To generate actionable and explainable insights and to be monitored continuously, evaluating the performance of the models. But most of all, they need time, a lot of time.
And of course, promising talents, data specialists, with deep knowledge of techniques and models, but also the business, able to translate results into insights with effective impact on the operations. And in the case of fraud in payments, the impact is in millions and millions of euros, but above all in confidence.
So, AI is not a choice, but an essential tool in corporate strategy. The question is no longer whether a company should opt-out of AI in support of its business. The real question is why it has not yet been adopted. In the context of fraud, banks, insurance companies, payment processors and market regulators can no longer live without the use of AI in their operations.