ARTIFICIAL INTELLIGENCE
CRACKING OR COPING? HOW COVID RESET AI AI has had to run fast to catch up with dramatically different patterns of behaviour during the pandemic. David Bailey, Santander’s Head of Payments Strategy, and Andy Renshaw, VP for Banking Solutions at anti-fraud software firm Feedzai, consider the short-term and long-term consequences of changes in the way we spend Artificial intelligence (AI) software spots bank account fraud by identifying behaviour changes. A sudden interest in iTunes from an account held by an 81-year-old that has never even featured an online purchase before would be an obvious red flag.
human help to make sense of the bigger picture and intervene when needed. “We use machine learning and AI, particularly in financial crime and fraud, to help us identify patterns of behaviour that we would deem suspicious,” he says. “But it’s always been complementary to the existing tools that we use, and we’ve always had to combine it with some human interaction – human intuition – to guide its direction.” For Andy Renshaw, vice president for banking solutions at fraud software firm Feedzai, changes as the pandemic began were manageable because, overall, the behaviour was still recognisable. Feedzai screens bank accounts
But what happens when seemingly everyone switches from buying clothes and holidays online to face masks and toilet rolls? That was the scenario that faced bank security systems from February onwards as COVID-19 swept across continents. With big increases in home-delivered groceries and a collapse in restaurant and travel costs, almost everyone’s buying patterns changed as countries locked down. A report by the MIT Technology Review in the US examined how the top 10 Amazon search terms were knocked off the chart and replaced by sanitisers and protective wear, We’ve had a reflecting anxiety around more stable the pandemic. And machine learning some AI programs built and AI performance to act upon subtle shifts than we originally were thrown into chaos feared, right from as a result. the start [of David Bailey, Santander’s head of the pandemic] payments strategy, says Andy Renshaw, the crisis underlines that Feedzai AI still needs
“
and debit cards in real time, using hundreds of data points to pinpoint anomalies. The immense volumes of data it handles require several data centres to run concurrently, but Renshaw says the systems held up well. “We’ve had a more stable machine learning and AI performance than we originally feared, right from the start. “The MIT Technology Review asked the question ‘is AI cracking under the pressure of the coronavirus pandemic?’,” he says. “Fundamentally, machine learning models look at individual behaviour and, from our perspective, it involves huge volumes. But, although the coronavirus pandemic is unique in terms of the scale of change, the underlying principles behind it are not unusual. “Machine learning software is already used to spikes on paydays, Black Friday, Cyber Monday, the last shopping day before Christmas, etc. With coronavirus, although customers are behaving in a slightly different way, their transactions are still relatively normal. “They’re still using merchants that they used before, for high-value and low-value payments. Fraudsters, too, are still, in many ways, committing the same types of fraud, and those transactions still look unusual, relative to normal behaviour.
Doesn’t compute: In some areas, massive changes caused by the pandemic confused AI
www.fintech.finance
Issue 17 | TheFintechMagazine
15