How can AI keep smaller institutions ahead of financial fraud?

Stuart Tarmy, Head of Industry Vertical Solutions, FinTech and Global Partnerships at Aerospike, outlines the role AI can play for institutions of all sizes.

Failing to respect the abilities of any enemy is a mistake, and financial institutions have long understood that financial fraudsters are their enemy. And they are smart in who and how they attack. Just like petty criminals, they look for the easiest targets that will give them the desired return. Today, small and midsize banks are in the crosshairs and are seeing rising fraud levels across payments and increasingly check fraud. This is a particularly dangerous situation, because unlike the larger banks, the smaller banks often don’t have the financial resources or capital reserves to easily absorb the losses.

Like the wider cybercrime community, financial fraudsters have always looked to exploit the latest technology. For example, in the late 1800’s counterfeiting was so prevalent in the USA that it was estimated that one third of the currency in circulation was fraudulent. From counterfeit coins and bills to today’s more modern credit, debit, real-time, buy-now-pay-later and embedded payments methods, the battle has only grown and become more sophisticated and difficult to fight. This has been exasperated by the increasing movement to real-time payments as the clearing/settlement window has shrunk from T+3 days to almost instantaneous.

Today’s frontier is Artificial Intelligence (AI), and many large financial organisations have already been investing and seeing higher growth, profitability, and lower risks as a result of it. Many think of AI in these financial institutions as being used for market analysis, robo- or algorithmic trading, and customer360 and personalisation. But AI is also being used across many other banking departments in areas such as fraud mitigation, compliance and risk management.

The first question any financial institution needs to ask itself is “where can AI provide the most value to me?” In many of the conversations I have, the answer is often lowering fraud risks.

Fraud can happen at any point in the relationship journey that a financial organisation has with its customers, but one of the most challenging areas is the onboarding of new customers and identity management. Whether opening a bank or brokerage account or submitting a mortgage application, it is essential to ensure the individual is who they claim to be, that they do not have a fraudulent past, and that their financial affairs do not link them with any money laundering or other illegal activities. This is typically a costly and slow process that has relied heavily on human involvement. Today, AI is being used to validate supporting documents and detect manipulated images, something that has become increasingly difficult as more applications are completed online and fraudsters have become more adept at creating synthetic identities and image manipulation.

Getting onboarding right helps minimise fraudulent activity and risk, but also helps ensure compliance with federal and state banking regulations. Beyond identity management, it is critical to also detect fraud at the transaction level itself, where speed and accuracy are critical to ensuring a winning customer experience. In today’s world, a transaction can come from any number of sources or devices, anywhere in the world, and occur across multiple borders, in different currencies, and across many intermediaries. In processing a payment transaction, making a decision on whether to allow the transaction to execute or stop it as potentially fraudulent should ideally happen in a round trip of less than 30ms. Distributed, cloud-based solutions are helping achieve this by putting the decision-making closer to the origin of the transaction, rather than in a siloed ‘ACME Bank’ data centre on another continent. These cloud-based solutions also lower the infrastructure and transaction costs for institutions and allow for scaling during peak usage or seasons.

Moving from rules to AI

For a long time, many institutions have relied on rule-based systems to detect fraud, but this approach is somewhat ‘OG’ (old guard) and has difficulty keeping pace with today’s threats. Once rules are defined, they remain fixed until operators decide they need to be improved or tightened because fraud losses are widening, or a new threat is identified. Rule based systems are, in general, slow to identify problems because they are not designed to recognise patterns on their own. This can lead to significant fraud and risk exposure, especially for the smaller financial organisations which are in the relatively worst position to absorb these losses.

More modern AI, on the other hand, is constantly learning from the data and transactions to which it is exposed. This means it not only reacts to change far faster than any human can, but it refines its decision-making process and can make it faster than humans can blink. The most advanced organisations are already moving beyond conventional AI to more advanced neural networks. Neural nets are a programming technique which was designed to closely simulate the way neurons in the human brain work and allow an AI model to self-learn, improving predictive modelling and decision-making.

Get your head in the clouds

Data is a critical key to successful AI applications. Significant amounts of data are required for the training set which is used to develop the AI model that your financial institution will use for a given task. The AI algorithms will identify relationships, patterns, and the risks associated with them, which informs how decisions are made. The first step in developing a usable AI solution is knowing that the data set on which you plan to train is strong. Most often, the model will use a mixture of your own historical data as well as relevant third-party data.

The data platform chosen to build an AI model, as well as run it in a live production environment, is critical – particularly if you want to use it for a task such as real-time fraud detection. A graph database is particularly good for building fraud solutions. For example, a cloud-based, distributed graph database will allow an institution to store and query massive amounts of data with exceptional performance and reliability without the infrastructure costs of owning and running regional data centres. They are also highly scalable so, as your data sets and AI needs grow, your graph database can do the same with no negative impact on performance.

Paypal has used its data platform to implement real-time AI fraud prevention for digital payments. In doing so, it reduced its daily risk exposure from $5 million per day down to a more manageable $0.17 million per day (based on $300 million in payments per day). Paypal also reduced its false positives by 30x and while significantly cutting the size of its server infrastructure.

Baby-steps

The availability of cloud-based databases and AI solutions means that financial institutions of all sizes can begin to explore the benefits of AI to reduce fraud, improve efficiency and reduce risk, and improve the customer experience. We see many customers starting small, learning how and where to apply AI solutions on a few use cases and, as they see successes, quickly identify more applications to modernise. There has never been a better time for institutions large and small to start their own AI journey and, if they don’t, they run the risk of getting left behind.