From rolling out OpenAI’s GPT-4 to employees to preventing $27 billion in fraud, Visa exec Rajat Taneja opens up about the payments giant’s AI playbook

• Visa has more than 300 live AI models that span risk management, cybersecurity, and operations.
• The payments giant has been using AI for 30 years, but ramped up adoption nine years ago.
• Visa is licensing ChatGPT-4 for all employees to be using generative AI by the end of the year.

It’s been three decades since Visa added artificial intelligence to its technology stack, but by the end of the year, there won’t be a job or position that doesn’t use the technology.

Visa relies on AI for mission-critical tasks such as payment processing, fraud detection, and network security. It has over 300 AI models that are live and running. In the last ten years, the payments giant has spent more than $3 billion on AI, and one of its tools for real-time payment monitoring has helped prevent $27 billion in fraud.

According to Rajat Taneja, Visa’s president of technology, the company is now licensing generative AI software from OpenAI and will spend hundreds of millions of dollars on AI each year, hiring thousands over the next two years.

“AI is going to be a huge part of how we grow, but it’ll also be part and parcel of everybody’s work,” Taneja said. “That’s going to be a big part of how we think about the future, where it’s assisting you in almost everything you do, whether it’s marketing, finance, client services, or coding.”

Visa is going all-in on generative AI for its employees

Visa has licensed a private version of GPT-4 that runs on Microsoft Azure and incorporates internal Visa data. Since February, the payments company has made the tool and underlying data available to 13,000 employees. It plans to roll it out to the rest of its workforce by the end of the year.

Taneja recently used it to revamp a job posting for an AI role, rewriting the job description, drafting interview questions based on the position, and then asking the AI for the best answers.

According to Taneja, Visa continues to experiment with similar AI models, such as Meta’s AI model Llama 2 and those from the French-American company Hugging Face.

Visa’s engineers are using AI to write code, and while it isn’t perfect, more AI-generated code is being used each week, with approximately half of the code being approved in some cases. Engineers will direct and supervise models and use them as a “assistant coder,” someone who never forgets anything, is trained on the latest technologies, and learns more every day, according to Taneja.

Not only will AI change the way Visa employees work, but it will also force the company to expand its workforce. Visa’s data and AI platforms and services are built by over 2,500 technologists, but the company plans to hire 2,000 more in the coming years to carry out its AI roadmap.

How Visa is using AI to scale risk management and cybersecurity

Despite the fact that Visa began using AI in 1993, it wasn’t until 2014 that the company rebuilt its core data platform and began applying AI across the business and to maintain operations.

Visa processes up to 76,000 transactions per second. Visa must connect several parties in the blink of an eye to authorize transactions while managing risk.The importance of speed and security cannot be overstated. And AI is required for this to happen in real time, as consumers have come to expect, according to Taneja.

The payments company uses AI to ensure that its systems are operational, such as detecting if a cable or fiber has been severed along its 24 million-mile telecommunications network.

Visa introduced AI models that consume billions of data points every day in order to train, learn patterns, and identify any cybersecurity or payment-security threats. It employs artificial intelligence to continuously improve its technology stack, with Visa making a thousand software changes and updates on some days.

It protects cardholders from criminals using AI to sign into their accounts by learning how a specific customer types in their user ID and password. Taneja explained that biometric behavioral analytics will cause AI to increase authentication if the models believe someone other than the customer is entering the credentials, even if the ID and password are correct and the IP address and device are recognized.

According to Taneja, AI has been critical for Visa in providing reassurance and reliability to clients and customers. “If someone is buying a subway ticket, paying for groceries, buying medicine, or having a cup of coffee, then if there is a problem going on, Visa’s AI systems will continue to work in a way that prevents fraud, catches counterfeits, and catches any ATM withdrawal attacks,” he added.