As bankers continue to arm themselves against fraudulent actors, they have a powerful - yet underutilized - tool at their disposal: machine learning data analytics. Machine learning is a powerful resource able to strengthen a banks fraud prevention capability. This resource, when utilized correctly, gives banks a greater ability to view all transactions in real-time and seek and capture fraudulent items while allowing the vast majority of legitimate transactions to process uninterrupted. Thankfully, the overwhelming majority of attempted payment transactions arelegitimate, but for the few that aren’t, leveraging quality data can help isolate the culprits and identify risk.
Keeping the Customer Happy
Bankers walk a tight rope between ensuring legitimate customer transactions are rarely, if ever, interrupted, while simultaneously detecting and preventing fraudulent transactions. It would be impossible for bankers to interrogate every transaction before allowing it to pass, after all, The Association for Financial Professionals’ 2018 Fraud and Control Survey revealed that 78% of responding companies experienced check fraud in the prior year - the highest rate on record.
The reality is that most declined transactions are in fact legitimate. Detection systems are great for narrowing down a pool of suspicious transactions, but not so much in pinpointing the individual fraudulent items with total precision. Casting too wide a net isn’t necessarily a good thing as it leads to the capture of more false positives, resulting in customer inconvenience and dissatisfaction.
Fortunately, automated fraud systems enable the bank to apply machine learning to analyze large batches of data, identifying a subset of suspicious events and routing them for manual review by a specialist and a final pay/no pay decision. In turn, the volume of fraud cases requiring manual vetting decreases drastically, saving the bank time and cutting cost.
The Dreaded False Positive
Ideally, banking software could narrowly detect and capture all fraudulent transactions with 100% precision. However, in reality, attempting such precision carries a strong likelihood of alienating some of a bank’s best and most loyal customers. As an industry practice, legacy detection systems have typically stuck with flagging roughly 200 suspects for each actual fraud. However, having such a wide pool of items to review significantly adds cost for the labor that must be applied to the research, and fuels frustration in customers whose transactions are held up. Worse yet, fraud analysts who discover nothing wrong with 99.5% of the cases they investigate, inevitably find their eyes glazing over and can eventually wind up overlooking one of the bad actors attempting a true fraudulent transaction.
By leveraging more robust data and stronger algorithms, modern fraud detection systems can reduce this costly 200:1 ratio to a more cost effective and customer friendly 60:1. Not only does this generate significant cost savings, it addresses a customer pain point - not only by minimizing the number of exception items, but also resolving more of them in real time.
Fraud professionals must also concern themselves with false negatives. These actual fraudulent payments can slip through the system undetected, generating a loss. To solve this challenge, machine learning based systems must be in place to further limit these occurrences.
The Future of Pinpointing Fraud
Machine learning based systems have the ability to consider more factors in the transaction’s ecosystem, looking for correlations in channel usage, payment sequencing, account type and geo-location, among other factors, while discarding irrelevant factors. It may also help banks distinguish between first and third party fraud, a key consideration in determining next steps. Another critical factor is a workflow that gets all appropriate data into analysts’ hands for real-time decisions.
Of course, any process in which 59 of 60 readings are false positives leaves room for improvement. First things first, there are significant benefits ready to be realized today, as well as room for future gains and/or ways to free up bandwidth to combat the fraudsters’ next advance. As bankers understand the importance of automated fraud detection systems, banks become better prepared to detect and prevent fraudulent transactions, especially in high exposure channels such as teller lines, ATMs and remote capture deposit.
- Machine Learning and AI ‘Crucial’ to Fighting Fraud, Research Shows
- UBS, Comerica and Mastercard: The Biggest Banking Moves in February
- Authorities Shed New Light on Wells Fargo Fake Accounts Scandal
- Details Behind Morgan Stanley’s Decision to Acquire E*Trade in $13bn Deal
- Ally Pushes into Credit Card Market with $2.65B CardWorks Deal