AmEx is experimenting cautiously with generative AI for fintech

Execs at AmEx Digital Labs, the company’s innovation arm, say they likely won’t launch their own LLM and will instead rely on integrating existing offerings

The wave of new business products and services launching with generative artificial intelligence ( generative AI) features isn’t slowing down, but many established companies are taking a measured approach when evaluating how to use and deploy the technology. 

Among those exploring AI for business is American Express (AmEx), the financial services, credit card and remote concierge giant. Founded in the United States in 1850 originally as a physical goods delivery company, AmEx has transformed itself into a global fintech leader, steadily embracing new technologies throughout its nearly two centuries of existence.

Now, confronted with the recent rollout and drumbeat of hype around OpenAI’s ChatGPT, Google’s Bard, Anthropic’s Claude and similar generative AI products built atop large language models, AmEx sees opportunity to use these technologies to better improve its customer experience across its credit cards and bank offerings for businesses and individuals.

“It may not be a standalone product this year, but we’ll be looking at direct-to-consumer or small-business services as part of our experimentation, and see what works and what doesn’t,” said Luke Gebb, executive vice president of American Express Digital Labs. AmEx Digital Labs is a division of AmEx formed in 2017 with the ethos of a startup, rapidly evaluating and testing new technologies and using them to incubate new products that make their way to the rest of the company.

AmEx Digital Labs was working with AI before it was trending

In fact, AmEx Digital Labs has already been at the forefront of using AI for financial services. This team of 100 technology experts scattered around the globe was responsible for integrating Mezi, an AI digital assistant that makes customer travel booking recommendations, into an AmEx mobile app pilot program. AmEx later acquired the company behind Mezi and wove its technology throughout some of its services. 

“We launch 20 to 25 pilots a year,” said Laura Grant, VP of product development for emerging platforms and AI at AmEx Digital Labs. “We look at technology from the perspective of, ‘How is this going to impact our customers and make their lives better?’”

The way AmEx Digital Labs works is that it first builds and prototypes products within its own small unit, before rolling it out to other parts of the company, ultimately turning over control to whatever team within AmEx is best equipped to lead that particular digital offering.

For example, AmEx Digital Labs created the company’s Rewards Checking account, a high-yield (1% annual percentage yield) consumer banking offering wherein customers earn rewards points for every dollar spent with their debit cards, then turn those points back into direct deposits. 

In addition, AmEx labs stood up the company’s “Pay with Bank Transfer” feature, which allows online merchants to include an AmEx-powered button among the payment options they accept. If a customer presses the “Bank Transfer” button, they’re prompted to securely and privately link their checking account at one of many supported banks, and pay directly for the product from it, avoiding credit card transaction fees.

How AmEx plans to use generative AI

AmEx Digital Labs intends to explore the potential of generative AI in enhancing predictive analytics, helping them to better understand customer patterns and behaviors.

In relation to AI applications, AmEx aims to focus on “predicting how our customers are going to perform over time, enabling better financial planning and decision-making,” said Gebb. Another major use case: customer sentiment analysis and customer interactions.

“The way I think about AI is as we think about the customer experience,” said Grant. “What is it going to do to make people’s lives easier and customer experience better?”

AmEx Digital Labs is currently exploring ways LLMs could be used “behind the scenes” to analyze all of the feedback and inquiries customers provide through AmEx’s existing customer service portals, as well as unofficially on social media, and to understand how to deliver appropriate and helpful responses. 

Grant said that when approaching all new technology but especially AI, AmEx Digital Labs seeks to understand how it can help with its “3 Ps,” making a product more personalized to an individual customer, more proactive and more predictive.

As for whether AmEx would seek to develop its own homemade LLM using its financial services data, following in the footsteps of the financial information and media giant Bloomberg launching its BloombergGPT with 50 million parameters a month ago, Gebb was dubious. 

“Our hypothesis at the moment is that we would be better suited using LLMs through partnerships,” Gebb said, but added he did not have any specific ones to announce. “I don’t see us spinning up our own LLM from scratch.” 

Ring-fencing for security 

While AmEx Digital Labs sees a customer-facing generative AI based on an LLM as potentially useful, for now it is focused more on backend implementations.

“It feels smarter initially that a human is looking at and reviewing the output” of an LLM, Grant said.

Furthermore, because of the critical importance of its product offerings — affecting customers’ bank accounts and credit lines — and the associated hefty amount of government regulations in financial services, AmEx Digital Labs also makes sure to approach all of its experimentation with AI very cautiously and securely.

Asked how AmEx might avoid the same fate of the three Samsung employees who reportedly shared sensitive, proprietary information with LLMs, Gebb said: “We are ring-fencing,” referencing the broader security approach that limits software applications and the data they can access.  

“Not every employee can access LLMs through their employee laptop clear — only those who have been cleared and trained up on what they are doing.” 

How AmEx’s AI approach compares to those of other large enterprises

AmEx’s experiments with generative AI are clearly in their earliest phases, but that is in line with many of its peers. 

A recent survey by KPMG found that 65% of 225 executives surveyed believe that generative AI will have a high or extremely high impact on their organization in three to five years, but 60% say they are still one to two years away from implementing their first solutions, and anticipate spending the next six to 12 months increasing their understanding of how generative AI works, evaluating internal capabilities and investing in new tools.

The survey also showed that executive prioritization of generative AI varies significantly by sector. Most of the executives in technology, media, telecommunications, and healthcare and life sciences felt they have appropriately prioritized generative AI, while only 30% in consumer and retail said it was a priority. Respondents in technology, media, telecommunications and financial services said that researching generative AI applications is a high or extremely high priority in the next three to six months.

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.