Typically, the decision to approve or reject transactions is made by the bank that issued the card. But banking networks can fail due to natural disasters, faulty software, or other reasons. Visa said that starting in October, its stand-alone, autonomous system, capable of self-verifying transactions, will be available to banks that subscribe to the service. “This technology is an important first step in helping to reduce the impact of disruptions, ” said Rajat Taneja, President of Visa Technology.

The new service reflects the growing use of AI in banking. Research firm International Data estimates that banks will spend $ 7.1 billion on artificial intelligence technologies in 2020, and in 2024, annual spending will rise to $ 14.5 billion on initiatives such as fraud analysis and investigation.

The new Smarter Stand-In Processing (SSTIP) service uses deep learning and was created by data scientists and software engineers from Visa: the company has three patents for this technology, two of which are pending. According to Mr. Taneji, various outages in banking networks affect several million credit and debit card transactions each year, often leading to unjustified rejections of transactions. In this case, cardholders sometimes have to call their bank for help. Stores and banks can lose income if transactions are not completed due to interbank network failures.

“These problems affect business, but our key goal was to make the consumer environment more friendly,” said Mr. Tanezha.

The service will be triggered automatically when the Visa network detects that there is no access to the infrastructure of a particular bank. According to Carolina Barcenas, senior vice president and head of Visa Research, the older version of STIP, which used a simpler rule-based model, performed worse and was not very accurate.

SSTIP doesn’t use hard-and-fast rules – the new deep learning model is more advanced, trained to sift through billions of cardholder activity data points to figure out relationships on its own. For example, it can automatically know that a particular cardholder transaction was normal and must be approved based on the user’s previous payment and transfer history, as well as the merchant’s location in relation to the cardholder, the time of purchases, and so on.

“The model detects many of these relationships and creates the kind of intelligence that analysts have had to create in the past. The new version of the technology works better and more accurately than the previous one, ”said Ms. Barsenas. During the tests, she said, the artificial intelligence model based on deep learning was 95% consistent with the decisions of banks to approve or reject transfers. According to Visa, this technology has more than doubled the accuracy of the old method. Both versions will continue to coexist, but the more advanced version will become available to customers as an additional service.