Visa Inc. said Wednesday it has developed a more advanced artificial intelligence system that can approve or decline credit and debit transactions on behalf of banks whose own networks are down.
The decision to approve or deny a transaction typically is made by the bank. But bank networks can crash because of natural disasters, buggy software or other reasons. Visa said its backup system will be available to banks who sign up for the service starting in October.
The technology is “an incredible first step in helping us reduce the impact of an outage,” said Rajat Taneja, president of technology for Visa. The financial services company is the largest U.S. card network, as measured both by the number of cards in circulation and by transactions.
The new service reflects the growing use of AI in banking. Banks are expected to spend $7.1 billion on AI in 2020, growing to $14.5 billion by 2024, on initiatives such as fraud analysis and investigation, according to market research firm International Data Corp.
The service, Smarter Stand-In Processing, uses a branch of AI called deep learning that roughly mimics neurons in the human brain and is an underlying technology powering self-driving cars, voice-enabled digital assistants and facial recognition.
It was built with Visa’s in-house data scientists and software engineers and the company has three patents related to the technology, two of which are pending.
Network disruptions and outages affect several million credit and debit card transactions annually, often causing transactions to be unnecessarily declined, Mr. Taneja said. When that happens, cardholders may have to call their bank for assistance. Merchants and banks could lose revenue if sales aren't completed.
“There’s a business impact because of the transactions flowing through, but our motivating driver was consumer experience,” Mr. Taneja said.
Smarter STIP kicks in automatically if Visa’s network detects that the bank’s network is offline or unavailable.
The older version of STIP uses a rules-based machine learning model as the backup method to manage transactions for banks in the event of a network disruption. In this approach, Visa’s product team and the financial institution define the rules for the model to be able to determine whether a particular transaction should be approved.
“Although it was customized for different users, it was still not very precise,” said Carolina Barcenas, senior vice president and head of Visa Research.
Technologists don’t define rules for the Smarter STIP AI model. The new deep-learning model is more advanced because it is trained to sift through billions of data points of cardholder activity to define correlations on its own. For example, it could automatically learn that a particular cardholder transaction was normal and should be approved based on historical data about that person, such as the location of the merchant in relation to the cardholder and the time of day they are shopping.
“The model finds a lot of those relationships, and it’s creating that intelligence that in the past, the analysts had to do,” Ms. Barcenas said. “It also outperforms.”
In tests, the deep-learning AI model was 95% accurate in mimicking the bank’s decision on whether to approve or decline a transaction, she said. The technology more than doubled the accuracy of the old method, according to the company. The two versions will continue to exist but the more advanced version will be available as a premium service for clients.
UBS Card Center AG, a subsidiary of Swiss bank UBS Group AG, has been working with Visa since 2018 to design and test the technology and is expected to be one of its first users.
Disruptions on UBS Card Center’s network sometimes occur because of telecommunications failures or natural disasters, said Anindya Mukherjee, executive director and head of strategic initiatives at UBS Card Center. Visa’s technology is an innovative stopgap measure for approving transactions during such outages, he said.
“Each of these has the potential to disrupt our business,” he said. “Even the most robust systems actually benefit from redundant layers of backup technology.”
Write to Sara Castellanos at sara.castellanos@wsj.com