How Stripe is using GPT-4 to fight fraud

OpenAI’s newly released AI GPT-4 is lighting up the tech world as people poke around in the program, with various creative uses blossoming like a desert after the rain.

Irish/American financial services company Stripe is one of OpenAI’s first collaborators on GPT-4. They pulled together a team of 100 workers from across the organization to drop what they were doing and instead think of how GPT-4 could help create new features or optimize old ones. (The collab works both ways: Stripe says the fintech company is also going to be helping OpenAI as it commercializes its AIs.)

“Our mission was to identify products and workflows across Stripe that could be accelerated by large language models and to really understand where LLMs work well today and where they still struggle,” Eugene Mann, the project lead of Stripe’s Applied Machine Learning team, said in an OpenAI post.

“But just having access to GPT-4 enabled us to realize, ‘Oh, there are all these problems that could be solved with GPT surprisingly well.’ ”

Stripe began incorporating OpenAI products even before GPT-4 — last year, its predecessor, GPT-3, was integrated into their customer service program Stripe Support, to help agents fix problems faster. 

This time, Stripe employees brainstormed a list of 50 possible applications for GPT-4, landing on 15 that were strong enough to already be put into practice. They fell into three main buckets: better understanding the businesses of their customers; providing support for questions about documentation; and monitoring for fraud.

When a customer comes to Stripe with a problem, agents need to have a strong understanding of how exactly their business is using Stripe’s software, which is primarily used for payment processing. This can require hours of legwork to ascertain — and some clients are more difficult than others.

“Many businesses, such as nightclubs, keep their websites sparse and mysterious so it can take a great deal of searching and clicking to understand what’s going on,” Mann said. GPT-4 can now scan clients’ websites and crank out summaries that come out more accurate than an employee’s, according to Mann.

In a similar vein, GPT-4 can also be used to act as a “virtual assistant” for clients facing technical or troubleshooting issues. The AI accepts the user’s question, combs through documentation to find answers, and relays them back, summarizing a solution.

In their final listed application, Stripe has begun to use GPT-4 to analyze the syntax of posts in Stripe community forums on platforms like Discord. 

“Like the introduction of email, smartphones, or videoconferencing, GPT-4 has the potential to fundamentally rewire—and improve—how businesses run.”

Eugene Mann

While these forums can be used to help crowdsource solutions to particularly niche or vexing tech troubles, they can also fall prey to bad actors. By posing as community members, these actors can try to pry critical information from the community — the ol’ wolf-in-sheep’s-clothing trick. Using syntax analysis, GPT-4 can flag posts for Stripe’s fraud team, who can then follow up on the lead.

While the features are still relatively new, Stripe seems all-in on the AI.

“Like the introduction of email, smartphones, or videoconferencing, GPT-4 has the potential to fundamentally rewire—and improve—how businesses run,” Mann said in a statement for Stripe.

We’d love to hear from you! If you have a comment about this article or if you have a tip for a future Freethink story, please email us at [email protected].