Consumers demanding faster, more convenient payments led to the rapid adoption of digital wallets and contactless payments. The pace of adoption of real-time payments has only accelerated, forcing banks and merchants to quickly add new payment types, such as QR code payments, mobile wallets and payment codes, to meet changing expectations.
While financial institutions are working hard to fulfill evolving needs for new services, simply responding to demand by launching a new product is not enough. These new ways to pay are often launched quickly to meet demand without having implemented robust fraud prevention. By answering consumers’ needs without considering fraud implications, financial institutions are creating new ways for criminals to exploit weaker controls. The challenge becomes less about innovation and more about protecting these new payment types and channels.
In Q1 of 2021 alone, with lockdowns across many countries, global digital revenue grew by 58 percent from Q1 2020. Machine learning has often been seen as the key driver of any successful fraud or AML management strategy, but it is not without its limitations. Traditional machine learning is based on predictive analytics, and it is impossible to predict external factors such as unprecedented changes in consumer behavior. Today’s payments can come from multiple sources and be made through almost any device. This generates a wealth of data that helps to surface new fraud patterns. Once these fraud patterns emerge and financial institutions create new services to meet consumer demands, machine learning models begin to degrade. Financial institutions are forced to play catch up, spending time and money re-training and re-deploying the models.
To combat this, ACI is introducing Fraud Scoring Services, a fully managed subscription service in which machine learning models are managed with our patented incremental learning technology, alongside a financial institution’s pre-existing fraud prevention software.
What is incremental learning?
Simply put, incremental learning enables traditional machine learning models to continuously and autonomously adapt to fraud patterns as new data becomes available. With incremental learning, financial institutions, banks, intermediaries and merchants do not have to worry about retraining their machine learning models. Instead, with incremental learning technology, a model can refresh itself within two hours, enabling it to stay more effective for far longer.
How does incremental learning work?
Much like learning to play the piano — our incremental learning approach does not need to re-learn every composition that it knows already to gain knowledge about a new piece of “music”. It is only the new composition that needs to be learned. Traditional re-training of a model would mean it would have to learn all pieces of music again, including the new one.
According to our tests, this process allows the model to adapt to change quickly and efficiently, meet emerging consumer demands and prevent up to 35 percent more fraud.
Advancing machine learning with Fraud Scoring Services
With today’s huge volumes of transactions and disparate channels, the challenge of combating fraud is harder than ever. Being able to instantly react to new fraud signals across all payment types offers a massive advantage to financial institutions. While traditional models have to consume a huge amount of historical data, incremental learning consistently monitors performance as it focuses on streaming data, eliminating the need for less reliable historical data. In our tests, incremental learning model performance was maintained or even improved over the course of 13 months, compared with just three months for traditional models. Financial crime evolves faster than the time it takes for machine learning to re-learn models. With Fraud Scoring Services, financial institutions would have more operational efficiency, saving thousands of U.S. dollars on time taken to retrain models.
Beyond preventing fraud more effectively, institutions benefit from a reduction in fraud-related losses and poor customer experiences, allowing banks to maintain a level of trust with their customers. If a sudden increase in postage scams emerges, for example, financial institutions can be confident that their fraud management solution is supported by machine learning models that are already on the case, and customers will be protected.