Skip to main content

Unlocking the Potential of Generative AI in Payments

Transparent brain with visible neural pathways on a dark background.

Generative artificial intelligence (GAI) has soared to the peak of its hype cycle. While the technology has its detractors, those who are quick to dismiss it as mere hot air risk overlooking its very real potential impact. 

This is the view of Chris Uriarte, Partner with Glenbrook Partners, who spoke at a recent PULSE event. Uriarte is a payments expert and past recipient of the Infoworld CTO 25 Award, which recognizes the 25 most innovative technology executives in the country each year.

A New Generation of AI

Despite the extensive attention GAI has garnered over the past year, Uriarte emphasized that many business leaders remain puzzled by how to effectively harness the technology and navigate the associated risks.

“GAI is fundamentally different from predictive AI and the machine-learning techniques that have been used in payments for a generation.”

Chris Uriarte, Partner, Glenbrook Partners

The first thing to understand about GAI is that it is fundamentally different from predictive AI and the machine-learning techniques that have been used in payments for a generation. Predictive AI – commonly used in fraud mitigation and predictive modeling – focuses on performing specific tasks using structured data. In contrast, GAI thrives on unsupervised learning, using unstructured data to create entirely new data.

Emerging Use Cases

Amazon is using GAI to fill a gap in real-world data for testing its Just Walk Out technology. The problem the company faced is that, because relatively few stores use Just Walk Out in this early stage, it would take years – maybe decades – to see the volume of checkout experiences reach the critical mass needed to mature the technology and prove that it is secure and effective. GAI can fill that gap through synthetic data creation. 

As described in a story in the food-technology publication The Spoon, “The Just Walk Out team used datasets from millions of AI-generated synthetic images and video clips mimicking realistic, and sometimes rare, shopping scenarios, including variations in lighting, store layouts, and crowd sizes.”1

This unique use case for GAI greatly reduced the time needed for Amazon to gain confidence that Just Walk Out could recognize and properly interpret real-world customer experiences.

“This is a big real-world problem, and it’s a great example of how GAI can work alongside predictive AI,” said Uriarte. “If you speak to your engineers who are responsible for testing transaction-processing or fraud-detection systems, generating test data to do that can be an obstacle. You can solve the problem by having GAI create millions of virtual scenarios to feed into predictive AI models, making them more capable and robust.”

Uriarte said companies are similarly using GAI to create and optimize fraud rules without the intervention of a data analyst. Another use case example is the fintech Justt, which uses GAI to automate responses to chargebacks and disputes. 

What’s to Come

Looking ahead, Uriarte said traditional uses of predictive AI are here to stay, and GAI will coexist with and strengthen these existing tools. He predicts that it will take time for companies to build internal capabilities related to the technology, with external service providers leading the charge in the meantime. 

There is a potential downside to GAI. Uriarte predicts that criminal use of GAI tools will lead to significant losses for companies in payments. Regulation is another inevitable development, but what it will entail is undetermined.

What can you do to ensure your organization is positioned to capitalize on these emerging capabilities?

1. Develop use cases.

Become familiar with what’s being brought to market. Ask questions, start small, and keep it simple. 

2. Engage with vendors.

There are lots of companies that are thirsty for clients to trial the GAI technologies they’ve been working on.

3. Assemble your internal team.

You may be surprised that stakeholders from throughout your organization – including data scientists, product managers, marketers, and the fraud-detection team – have an interest in defining a short-term and long-term strategy.

4. Collaborate with legal and compliance.

Recognizing that these tools come with risks, it is essential that the risks are well understood.

PULSE has long used predictive AI in its DebitProtect fraud-detection service. The network and its parent company, Discover Financial Services, also are exploring potential GAI uses via selective experimentation.


1 Amazon Details Usage of Generative AI-Created Synthetic Data to Train Just Walk Out Technology, The Spoon, September 19, 2023.