Artificial intelligence: a winning strategy for payments

The race is on to reduce fraud and continue improving payment flows. Artificial intelligence (AI) offers a winning strategy, says Chalapathy Neti, head, AI and machine learning platform, Swift.

AI offers a winning strategy in the race to reduce fraud and continue improving payment flows

AI is out of the lab and already well on its way to delivering smarter tech solutions in our daily lives. Just look at the way Amazon and Netflix use machine learning algorithms to continually serve us fresh content and products based on our previous behaviours. We get a better, more personalised experience while they strengthen their business models.

AI has the potential to deliver improvements in finance too, with huge opportunities in strengthening how we analyse and process transactions. As regulation becomes more complex and the fight against fraud continues, AI might be a key to unlocking a more efficient and secure global payments system.

The promise of AI

When sending cross-border transactions, financial institutions must ensure they comply with the different jurisdictional requirements that exist around the world. This can be complicated, and on top of this, it’s essential that they protect their customers against fraud and avoid processing errors that cause delays. As payment speeds get faster and faster, and fraudsters constantly evolve their tactics, the race is on for financial institutions to stay one step ahead.

So, what if errors in payment messages were automatically corrected without the need for manual repairs? Or fraudulent transactions could be spotted upfront as part of a payment’s pre-validation process? It’s a promise that AI and machine learning can deliver, and soon.

How can AI do it better?

Since the start of the digital era, data has been among the most valuable of business assets. And growth in computer power means we can now efficiently process very large amounts of it, with rule-based systems having taken centre stage over the last few decades.

This was well illustrated in 1997, when IBM challenged grandmaster and former world champion, Garry Kasparov, to a game of chess. The catch? IBM wasn’t betting on a talented member of its own team. Instead, it was their Deep Blue computer that would be sitting across the board. In tasks like these, AI and machine learning algorithms come into a league of their own, with the computer able to accurately analyse millions of different positions and potential outcomes. With the best human chess players only able to plan roughly seven moves ahead, Kasparov’s back was against the ropes – eventually losing three and drawing two of the six games played.

Today’s AI systems don’t just process data efficiently, they also use algorithms to identify data patterns and refine searches. These algorithms learn from the repeated use of large data sets – constantly refining their outputs to group, classify and identify abnormalities, and make predictions based on an array of inputs. And there’s no shortage of data within the financial ecosystem, meaning that these algorithms can be trained and improved without the need for a limiting set of rules.

As well as anomaly identification and transaction processing, AI models are also being used to provide operational intelligence. This can help to predict and pre-empt systems outages, analyse spending and improve budgeting. They’re even used to refine recruitment processes by predicting which candidates are most likely to accept a job offer.

The opportunity to collaborate

We’ve seen that AI and machine learning technology is valuable to individual institutions, with many larger ones already working on their own innovative projects. But what if the financial industry could collaborate to realise the full potential of AI?

Pooling institutions’ relevant data (in full compliance with legal and privacy requirements) could help discover shared intelligence about developing risks or uncover coordinated attacks across organisations and geographies. And as the data sets get bigger, so too do the potential insights, offering the chance to better safeguard customers and improve efficiency for institutions of all sizes.

Collaborative transaction monitoring initiatives are already underway in several countries. For example, in the Netherlands, a group of five banks have been piloting a project – Transaction Monitoring Netherlands (TMNL) – and consulting on the legal and privacy aspects of data sharing for transaction monitoring.

The Monetary Authority of Singapore (MAS) has announced that in 2023 it will launch a digital platform and regulatory framework, called Cosmic. This will allow financial institutions to securely share relevant data to combat money laundering, terrorism financing and proliferation financing. MAS is initially working with six commercial banks with other initiatives underway in Australia, the UK and the US.

At Swift, we’re playing our part too. As the financial industry’s neutral and trusted infrastructure provider, we carry an average of 45 million financial messages a day on behalf of over 11,000 institutions worldwide. This traffic generates detailed data which, thanks to state-of-the-art privacy preserving technologies, can be used to train machine learning models and generate real solutions for the entire industry.

A foundational model for anomaly detection

Today, we’re innovating with our community to co-develop an AI model that can detect anomalies in very large data sets. Once built, this foundational model will then be refined and tailored for specific industry needs – like distinguishing between false positives and alerts that really do need investigating or identifying and repairing certain errors more consistently than current systems allow.

This work is just the beginning of a journey to develop these technologies and realise their true potential – providing solutions to the operational and information challenges financial institutions face. It’s a strong promise for the future of finance, and one that’s stronger still when collaboration is on the table.