How AI Helps Visa ‘Step In’ (Smartly) When Issuers Go Offline

Getting transactions done — particularly card transactions — requires a seamless flow of card-level data between cardholders, card networks, acquirers and issuers.

When part of that ecosystem goes offline, faces a service interruption or technical difficulties, there needs to be a “back up” to ensure that commerce proceeds without a hitch.

Wednesday (Aug. 26), Visa announced Visa Smarter Stand-In Processing (Smarter STIP), which leverages artificial intelligence (AI) to help Visa act as a “backup” to approve or decline transactions on an issuer’s behalf.

As Gourab Basu, vice president of global network products at Visa told PYMNTS, using high tech and dynamic analytics to help financial institutions better manage authorizations keeps end-users happy even when the unexpected happens and issuers face occasional disruptions.

“In some rare instances, the ability for the issuer to provide decisions is not there because they are experiencing some kind of outage,” Basu told PYMNTS.

With “smarter” STIP, explained Basu, leveraging a combination of past transaction history and deep learning, Visa can make informed decisions about whether transactions should be approved or not.

The company said Tuesday that Smarter STIP is the first of a suite of new innovations featuring AI delivered through VisaNet.

At a high level, service disruptions and downtime affect everyone in the ecosystem, and an optimal cardholder experience is vital for issuers and issuer processors.

Drill down to the granular level, and Smarter STIP uses real-time AI to analyze account-level data (Visa has billions of historical records in place to train its model) and past transactions which leads to better transaction decisions.

As Basu noted, the “smarter” STIP builds on the stand-in processing capability that has long been part of Visa’s network services and traces its genesis back to pilot programs stretching back about a year.

In a typical transaction, a cardholder presents a merchant with a card and the card data, such as account number and chip, is read at the point of sale (or checkout).

The merchant transmits the data to the acquirer, who transmits an authorization request to Visa. Visa sends the authorization request to the issuer for review. The issuer approves or denies the transaction to Visa.

But if an issuer goes offline, that flow of information is interrupted.

Visa Steps In   

At that point, Visa can step in, Basu noted, and under traditional STIP, approving or declining a transaction is based on pre-configured, static parameters set by issuers.

By way of example, those parameters can be set with, say a $500 limit for cards in a particular portfolio.

“That’s how STIP works today,” said Basu, “but any two cardholders, even with the same banking relationship, the same issuer, even the same card program, will have different preferences and transaction behaviors.”

Against that backdrop, smarter STIP can take into account those behavioral patterns — the daily transactions at the coffee shop, for the $3 purchases, hypothetically speaking — that would allow Visa to closely mirror what an issuer would have done for that specific cardholder for that specific transaction.

It’s a decision that takes milliseconds through the ecosystem of acquirers and issuers connected to VisaNet.

“Instead of using ‘one size fits all’ parameters, we can customize account level decision data based on how the issuer has been responding,” he said.

“We’re doing this by using transactional data that we already had … The ‘new’ part that makes this smarter is the incorporation of the AI model,” said Basu, “as the transaction goes through VisaNet.”

Decisioning is made through real time, intelligent analysis across factors that include not just card numbers, but on card type, whether transactions are face to face and a host of other data points.

He said the model has been refined over the past 18 months by data scientists in the Visa Research team.

He said that in terms of accuracy, the model and decisioning mirrors what the issuer “would have done” 95 percent of the time.

The fact that the model uses “deep learning” means that the accuracy of those decisions will improve over time as it is applied.

Looking ahead, Basu said that smarter STIP will be available beginning in October.

“The ultimate goal is that the consumer experience of using a Visa card is preserved,” even amid those rare issuer outages, he told PYMNTS.