Harnessing the Power of AI to Transact
September 3, 2020
Visa Inc. released news of its newest capability: Visa Smarter Stand-in Processing (Smarter STIP). The AI (artificial intelligence) platform offers financial institutions a powerful tool to “manage transaction authorizations when service disruption occurs.”
Debuting this October, the technology boasts the latest in deep learning, using billions of records at its discretion to make transaction approvals that factor behavioral patterns before a system falls offline.
The sheer number of records generated through VisaNet processing network gives Visa Inc. a keen position to leverage AI as a service: “Visa pioneered the use of AI and neural networks to prevent fraud…[and] Smarter STIP builds on that track record.”
Among the host of reasons for service disruption, routine maintenance and power outages can get in the way of transaction processing. If a stand-in processing system is not in place, the ramifications are wide-reaching: lost revenue, poor cardholder experiences (resulting in flooded customer service lines), poor public perception, and increased scrutiny from regulators.
Smarter STIP is built on Visa’s existing digital infrastructure, allowing AI to analyze transactions at the cardholder level based on unique insights on a specific user’s spending patterns. Rather than “static rules applied across an entire card portfolio,” Smarter STIP makes transaction decisions in tune with an issuer’s decision-making process with an accuracy rate close to 95%.
Long Term Goals
Visa Inc. has invested heavily in optimizing core infrastructure, including “…scalable, high-performing, GPU-based platform designed to support the rapid deployment of deep learning capabilities.” It is a transformative process for the company:
And the technology’s adoption is well underway in the industry. Zurich-based UBS Group AG and Melbourne-based Latitude Financial Services have taken Smarter STIP in stride. What makes the technology so appealing? The model relies on:
- Multiple recurrent neural network layers with millions of parameters.
- Billions of historical records to train the model.
- Continuous learning from real time transaction and outage data, allowing the service to adapt and improve as an Issuers’ behavior changes over time.