Watch all the Transform 2020 sessions on-demand right here.


When it comes to practical uses for artificial intelligence and machine learning, the financial sector has been leading the way with projects that demonstrate the potential of these emerging technologies. Among the companies seeing a big return on their AI investments is Visa.

Melissa McSherry, a senior vice president and global head of data for Visa, said the company prevents $25 billion in annual fraud thanks to the AI it developed. The path Visa took to get here offers lessons to other companies weighing how and when to launch their automation projects.

“We have definitely taken a use case approach to AI,” McSherry said. “We don’t deploy AI for the sake of AI. We deploy it because it’s the most effective way to solve a problem.”

McSherry made her remarks in an interview with Lori Sherer, a partner with Bain & Company, during VentureBeat’s Transform 2020 conference.

The company’s primary use of AI involves its Visa Advanced Authorization platform. McSherry explained that the VAA scores every transaction that goes across the network and rates each one based on the likelihood that it’s fraudulent. By improving the separation between good and bad transactions, the system allows more transactions to be approved more quickly. “With 3.5 billion cards and 210 billion transactions a year, it is really worth it to everyone to make those cards work better and for more transactions to go through,” McSherry said.

The current system represents an evolution of a fraud detection service originally deployed in 1993. Today, the system uses recurrent neural networks along with gradient boosted trees. McSherry said having a defined use case — fraud detection — has allowed Visa to remain focused on how AI and ML can help improve services.

“I think it helps that we started with the first use case a long time ago,” McSherry said. “There’s no substitute for experience, and I think we have a fair amount of experience at this point on how to build and deploy these models. And so the first lesson is just at a certain point, you have to pick a use case and you just have to start.”

How these tools get implemented is also critical, McSherry said. In Visa’s case, the company introduces new AI and ML tools in a layered fashion in systems outside of its main transaction processing network to avoid increasing latency.

“You obviously aren’t going to be dropping a self-updating deep learning model into a mainframe that we use for transaction processing,” she said. “Our latency requirements are such that we need to be able to score these models in milliseconds, because we do score every transaction that goes into the network in real time based on the characteristics of that transaction.”

That said, the decision to implement AI tools directly into a system or adjacent likely depends on how much a business depends on speed. “I think the general idea is that you have flexibility in how you implement it, and you don’t necessarily need to implement every capability in every system,” McSherry said.

She noted that Visa has seen a 20-30% lift in model performance when it has applied advanced AI techniques versus more traditional ML techniques like gradient boosted trees. In some cases, the company has experienced more than 100% lift.

Going forward, McSherry is optimistic about AI’s impact on the financial sector, and on Visa’s business. The company is increasingly helping banks understand which products work and which ones don’t, for instance.

“When we do that faster, it speeds up their product development cycle so that they’re able to put better products in front of consumers faster,” McSherry said. “The idea of using AI to speed up and make those insights more precise is something we’re investing in and we’re very excited about.”