Most people have probably been exposed to fraud in some form – whether it was through a suspicious transaction made on their account or the bank calling to verify a purchase made. Traditional banks can take months to review genuine fraud cases whilst several forms are filed, various parties conferred with and evidence gathered.
Modern-day consumers want immediate resolutions but given that so many transactions are done online, investigating suspicious transactions require an incredible amount of resources for banks, which can be costly. In order to efficiently combat fraud, financial institutions are turning to artificial intelligence (AI) and its applications such as machine learning.
In this article, we will look at the ways in which AI is being used to support fraud prevention in banking and finance.
What is AI?
When it comes to fraud, AI is all about training machines to understand the patterns and behaviours that could point towards a suspicious activity. Machine learning takes vast amounts of existing data e.g. previous transactions and analyses the patterns. When a new transaction takes place, the machine can quickly ascertain whether it looks like an anomaly and send an alert to the bank.
In some cases, this does lead to genuine transactions being flagged as suspicious purely because they are anomalies (known as a false positive). However, more powerful algorithms are finding ways to combat those as well.
Fraudulent Transactions
Companies such as Datavisor provide specialist software that can track fraudulent transactions. They claim that the software can detect up to 30% more frauds with as much as 90% more accuracy. The alerts are sent in real-time meaning they can sometimes be stopped before even being finalised.
At the start of 2020, start-ups bank Monzo started freezing customer accounts after their system picked up signs of fraud that turned out to be genuine transactions. It is important that banks deploy strategies that can put alerts through triage before making decisions to avoid such false positives as best as possible. For example, Accenture proposes a method like below.
False Insurance Claims
Insurance fraud costs up to $80 billion per year across all lines and the US saw a loss of over $30 million in 2019 alone. Machine learning models are now being deployed that can help insurers find false claims quickly.
The Turkish insurer AK Sigorta say that their predictive analytics tools can work out whether a claim should be investigated within about 8 seconds, negating the need for extensive human review. The solution has helped them to improve their detection of fraudulent claims by 66%.
Security
Banks and financial service institutions are now investing in more advanced technology to mitigate the risk of fraud. Many have already deployed biometric features like fingerprint and retina scans into their apps to avoid personal information getting into the wrong hands. RBS, Wells Fargo and Barclays have all implemented biometrics into their solutions. Barclays recently upgraded their offering in partnership with Hitachi.
Data
The key to accurate detection with a machine learning process like this is data. If banks can collect and integrate data from disparate sources into one, they form a much more rounded view of the customer. For example, transaction data can be combined with online behaviour, social media data, credit agency data and other open sources. The more data points banks can hold on a customer, the better the accuracy of crime detection.
When looking into activity like money laundering, this is especially important to get a full 360 view of the customer before making a decision.
Machine learning models are designed to self-learn. This means that over time, they can start to establish their own rules and contexts around data without the need for human intervention. Learning through data reduces the need for an investigation, creating cost and efficiency savings for the banks.
Over time, the model will learn which alerts are low, medium, or high risk. This means it can make accurate decisions as to which ones need to be investigated, which can be hibernated for a later time and whether any can be formally closed off.
Behaviours
Companies like Ravelin are using a series of neural networks which are algorithms that act like the human brain. Each neural network in their model focuses on a different aspect of human behaviour. For example, one looks at how long they take to place orders on a website, one analyses the addresses they use to make purchase and another checks how long it takes them to enter card details. Each of these metrics could flag as a sign of fraud.
The behavioural layers are pulled into a single model for financial institutions to use. Rules will be trained and tested until they are correct for the user. For example, Barclays may want a different threshold for fraud than Bank of America based on the transactions they see.
Summary
The threat of fraud will always be a problem for banks. The digital world is creating new channels and giving cyber criminals the opportunity to attack banks from all angles. As much as banks create new ways to fight fraud, criminals will discover new ways to try and infiltrate the systems. This means there will always need to be a hybrid way of working between humans and machines to find the best possible solutions.
AI has gone a long way towards reducing fraud, the costs associated with it and the need for manual resource time. There is still some way to go before eradicating fraud but solutions have come a long way in a very short period of time.
Disclaimer: The author of this text, Robin Trehan, has an undergraduate degree in Economics, Masters in international business and finance, and MBA in electronic business. Trehan is Senior VP at Deltec International deltecbankstag.wpengine.com. The views, thoughts, and opinions expressed in this text are solely the views of the author, and not necessarily reflecting the views of Deltec International Group, its subsidiaries, and/or employees.
