Rajat Taneja has a delicate balancing act in his role as president of technology at Visa.
He is responsible for the security and resilience of 639 million transactions each day, money that moves between more than 150 million merchants and 4.6 billion active Visa cards while keeping pace with fast-evolving technologies like generative artificial intelligence.
“That infrastructure has to run in a rock solid manner,” says Taneja, who has worked at Visa for more than 11 years, after serving as chief technology officer at video game publisher Electronic Arts and a 15-year career in research and development leadership roles at Microsoft.
When he initially joined Visa in 2013 as a EVP of technology, the company was beginning to see an evolution from simpler forms of AI to more deep learning, neural networks that can learn from data including images and text.
Taneja and Visa then spent a few years—and $3.5 billion in total—rebuilding the company’s data platform from scratch, including rebuilding and rearchitecting systems to support more modern technologies like digital and mobile commerce, as well as tokenization, which replaces the 16-digit number found on a card with a digital “token” that Visa says reduces the risk of a data breach. The refreshed data platform also helped position Visa for the generative AI boom.
Visa’s researchers had been exploring generative AI but it was after the ChatGPT breakthrough in late 2022 that Taneja and his team created rules to protect the company’s data, avoid any copyright infringement, and ensure the models Visa uses are compliant with regulations. The company also gave nearly all its workforce access to a secure, internal version of ChatGPT, and it has licenses with Microsoft Copilot for software developers and the sales team.
“We give unfettered access, but with the guardrails and the controls that we built at the beginning,” says Taneja.
Today, Visa has embedded more than 100 products with AI and generative AI technologies. Fraud prevention is a big area of focus for Visa’s generative AI applications, especially because bad actors are also using generative AI for phishing attacks, helping scammers craft emails, texts, or calls to extract personal banking information from their victims.
Some newer generative AI tools include Visa Account Attack Intelligence, which uses the technology to identify and score attacks before a merchant is compromised, and Visa Protect for Account to Account Payments, which provides a risk score to assist financial institutions and automatically block bad transactions deemed suspicious.
Story Continues
Visa says generative AI, along with other forms of AI like deep learning detection models, has helped it prevent around $40 billion in fraud attempts annually.
As is the case with fraud prevention, most of Visa’s generative AI applications are behind the scenes, including assisting with cybersecurity and software development. As for some consumer-facing applications of generative AI technologies, Taneja says they are mostly still in development and not yet ready for deployment.
Taneja, who uses generative AI to craft songs to thank his 12,000-strong team or celebrate his father’s birthday, also aims to inspire excitement about the technology at Visa. A company hackathon he organized saw thousands of participants and more than 2,300 ideas for AI across departments varying from IT to finance to marketing. A second, three-week hackathon is currently underway. He also hosts regular office hours that are free to anyone on his team to demo new ideas, with the only parameter being that they must show him the code.
Visa works with a wide variety of AI LLMs, including OpenAI, Anthropic’s Claude, Google, and open-source language models from IBM, Meta, and Mistral AI. Taneja says his team aims to match the best large language models for the solutions for software development, building risk models, and agentic workflows to perform workplace tasks autonomously.
Taneja says he’s very open minded on the AI providers Visa works with, wanting to avoid vendor lock in. He prioritizes models that he feels are transparent, that put effort into fairness, and that give Visa the control to fine tune the models for the company’s needs. Hallucination rates are also an important factor in how he distinguishes between models.
“It’s only been two years now,” says Taneja of generative AI’s boom. “It is a toddler in terms of the life cycle of what these models will be.”
John Kell
Send thoughts or suggestions to CIO Intelligence here.
This story was originally featured on Fortune.com
