The data-rich nature of payments made using standards such as Swift’s ISO 20022 messaging system means there is now a wealth of data available for AI to learn from. It comes as little surprise then that, at the highest levels in the payment ecosystem, steps are being taken to integrate AI tools.
Earlier this year, Swift introduced a pair of measures to tackle fraud with AI. The first was to improve its existing Payment Controls service, using historical data to analyse and identify anomalies in live payments traffic data.
In its second development, Swift teamed up with 10 global banks including HSBC, Standard Chartered and BNY to launch an AI pilot in identifying cross-border payment fraud.
AI tools will be used to analyseanonymously shared data from various sources, a development which could be a “game changer for the industry”, according to Swift.
The process will test the use of secure data collaboration, while the anomaly detection model will be able to identify potential fraud patterns in the data. These two measures have been developed following the guidelines laid out in the ISO 42001 standard, the US’s National Institute of Standards and Technology’s AI Risk Management Framework, and the EU AI Act.
On an individual level, many banks are looking at developing their own AI models. Manish Kohli, head of global payments solutions at HSBC, says the bank has found several applicable uses for AI: “AI in predictive analytics identifies future needs based on transaction signals, which can include predicting currency preferences in cross-border payments or anticipating client requirements based on historical behaviour.”
He points to the bank’s FX Prompt, which uses predictive analytics based on historic payment data to predict the currency of a beneficiary account when making international payments.
While there are multiple use cases emerging for practical uses of AI, a combination of approaches to data gathering and assessment is suggested within the payments space, both as a precaution and to yield the most comprehensive results.
Cleber Martins, head of payments intelligence at ACI Worldwide, says there is a need for “correlation between human intelligence, data intelligence, and artificial intelligence”.
Edward Metzger, vice-president, market planning for payments efficiency at LexisNexis Risk Solutions, says getting the data right in international transactions is a pressing issue, with as many as 10 per cent of cross-border transactions failing.
The hope is that the effective use of AI could help to reduce this number. Martins points to the next generation of AI tools being able to assess more than just the details of a transaction, but also who it believes is making the transaction. Using signals such as how the transaction is initiated through to how the device is being held if the transaction is being conducted on a smartphone, AI tools can flag concerns if fraud is detected.
Federated machine learning, where multiple parties can train a model while keeping their proprietary data decentralised, is another system which will enable greater collaboration without comprising customer details or falling foul of rules on holding data domestically.
Complying with regulation there is still a cautionary note when dealing with AI, with concerns around bias and hallucinations within large language models themselves joining worries about data leaks and the progress of new regulation related to the technology.
Adam Gable, senior product director at Temenos, says: “One should be cognisantthat AI is not a silver bullet and leveraging this capability comes from deep understanding of the business problems and infusing these tools and techniques in the right place within the infrastructure and ecosystem.”
“A robust permissions and access security framework is required for AI in banking. This is to ensure that sensitive payment data remains secure by anonymising, encrypting, and tokenising private information, preventing unauthorised access while still enabling valuable data-driven insights,” he added.
Read the full article here
