In 1993, Visa became the first payments network to apply neural networks to calculate, in real-time, the ‘riskiness’ of a transaction.
With every one of the billions of transactions since – with 127 billion transactions between merchants and financial institutions handled by Visa last year alone – the AI system has been getting better.
Its ability to spot a dodgy payment is now so good, Visa says, it is preventing US$25 billion in fraud each year.
Visa says the AI doesn’t know an individual’s name or what has been bought, but it does consider previous activities and purchasing patterns to build a profile of previous transaction data and “learn what your typical purchase behavior looks like”.
Behaviours from all Visa users is grouped into clusters – such as ‘buys petrol after work’ or ‘uses phone to pay for train’ – which helps Visa spot whether a transaction departs from norms enough to be flagged as having a high risk of fraud.
“This delivers faster and deeper insights through previously unknown correlations,” Visa said.
Risk scores, based on more than 500 risk attributes, are shared with the account holder’s financial institution, where the decision is made to either approve or decline the transaction, or flag the transactions for follow up.
As well as being good at spotting dodgy transactions, the AI has also improved the confidence financial institutions can have in identifying good transactions, even when made by new or infrequent shoppers, which reduces the likelihood of false card declines.
The AI is served via Visa’s ‘Visa Advanced Authorisation’ offering, which is used by more than 8,000 financial institutions in 129 countries. Risk scores are served back to banks in about one millisecond, the company added.
“One of the toughest challenges in payments is separating good transactions made by account holders from bad ones attempted by fraudsters without adding friction to the process,” said Melissa McSherry, Visa’s global head of data, risk and identity products and solutions.
“The impact on fraud [from Visa’s AI] was immediate. By striking the right balance between human expertise and technology innovation, we continue to evolve our capabilities as new AI breakthroughs expand the realm of what’s possible,” McSherry said.
The US$25 billion figure is the first time Visa has revealed the value of the fraud it prevents per year. The rate of payment fraud is at a ‘historic low’ of less than 0.1 per cent the company said.
Visa told CIO Australia it expects that rate to half by 2025, with industry collaboration.
Fraud is the top payment-related challenge faced by retailers, cited by 55 per cent of merchants in a survey by the National Retail Federation and Forrester.
Visa this week announced its intent to join the Libra Association, a group of organisations governing Facebook’s newly announced cryptocurrency called Libra.
Along with Mastercard and 26 other organisations, the association will govern the new digital coin which is set to launch in the first half of 2020.
Libra is built on a blockchain and its value will be aligned with, and backed by, a reserve of low volatility financial assets. It is hoped the cryptocurrency will power transactions between established companies and consumers, as well as the estimated 1.7 billion unbanked consumers globally.
“We see an opportunity to be a voice at the table and provide the expertise for our clients and the payment ecosystem. We will seek to bring our expertise in security, interoperability and acceptance, along with our focus on privacy and trust,” Visa said in a statement yesterday.