American Express credits AI with improving customer experiences

The next time you use an American Express travel counselor to book a flight or get advice about a travel itinerary, he or she may use the power of generative AI to supercharge the response.

The credit card company has embraced gen AI and machine learning (ML) as it moves to embed automation technologies across the customer support lifecycle. The company has used traditional AI technologies for about two decades, and in recent years, it has rolled out two major customer-service-focused automation initiatives.

The first project, launched in 2022, uses ML to measure customer sentiment on telephone calls with support agents. American Express now measures customer sentiment on calls with about 12,900 support agents in the US, Canada, Australia, New Zealand, and Mexico.

The second project, which began with pilots in late 2023, uses gen AI to assist the company’s travel counselors as they interact with customers.

The Travel Counselor Assist project enables employees to tap into gen AI to prepare before speaking to a customer, then create itineraries, search for deals, find hotels, and create customer-specific recommendations for restaurants, sites of interest, and other vacation activities. The gen AI tool also summarizes the interactions between travel counselors and customers.

American Express extended Travel Counselor Assist to all its 5,000 counselors across 19 markets in October 2024.

Taylor Swift tickets

The Travel Counselor Assist can help support agents deliver recommendations on a wide variety of topics and handle all kinds of requests, says Anthony Devane, American Express EVP and head of global strategy and enablement.

“You can get incredibility complex inquires,” he says. “I’m calling up about getting a Taylor Swift or Harry Styles ticket. You also get inquiries saying, ‘I want to go to the Maldives, and I want to go via Dubai. I want to go to a restaurant that specializes in vegan food.’”

The gen AI recommendations also help travel counselors find recommendations or tickets in real-time, leading to shorter phone calls for customers and agents, Devane says. Before the gen AI tool, travel counselors sometimes had to place the customer on hold to conduct research or initiate a second call after looking into the customer request.

“Most inquiries take a significant amount of time to answer,” he says. “We have used gen AI to create agendas and itineraries, which we’re immediately able to relate to the customer on the phone. Rather than maybe wait an hour for a call back, you can real-time solve the itinerary, and it respects the customer’s time.”

American Express is seeing an increase in travel bookings as gen AI has decreased the amount of customer time needed to create their travel plans, Devane says.

Measuring empathy

The ML-based customer sentiment project also focuses on the customer experience, he adds, with American Express using the tool to improve customer support. The company has increased its Net Promotor Score since rolling out the ML sentiment technology, he says.

“We handle many millions of phone calls, and we have machine learning in the background, which gives every single call feedback to the customer care professionals who handle those calls,” he says. “It enables us to give targeted feedback to our colleague base to say this is how the card member felt in that call, because it can measure empathy and really help us understand how we construct our value and product proposition.”

While both projects have yielded beneficial results, an important first step was to get employees on board, Devane says. American Express has run employee focus groups to talk about its investment in automation technologies, and the company encourages its customer support agents to experiment with AI, he adds.

“I’ve always had the same idea — bring people on board the journey versus expect them to arrive at a destination,” he says. “We have to earn the right to win hearts and minds of both our customers and our colleagues, and our colleagues are essential in driving, my opinion, the best service in the world.”

The company has about 100 automation ideas in its IT test kitchens right now, many from employees, and they are active in the discussions about whether to implement them, Devane says.

“We implement some of the ideas, and some we don’t because the hallucination rates are too high, or the source code is hard to crack,” he adds. “We don’t walk away from it, but we put it on pause because maybe these [AI’s] intelligence will increase in the next three months or the next six months, and we always go back to iterate.”

Testing the automation technologies before full launch is also a major focus, Devane says. In addition, the company always has human oversight of automated actions or recommendations.

“Whether it will be literally booking a ferry ticket to go from the UK to France versus booking a $1 million around-the-world cruise for six months, it’s important to the customer,” he adds. “You have to test and iterate. Before we deploy any technology, we have a sandbox environment where we use redacted card member data, and we work with the models to understand the accuracy.”

Flexibility focused

Beyond the need for employee buy-in and rigorous testing, American Express designed its gen AI solution with the ability to use multiple large language models (LLMs), says Gary Kensey, EVP and CIO for the Global Services Group and corporate technology at the company.

With LLM capabilities changing rapidly and models leapfrogging one another, the company wanted to avoid being locked into one AI tool, Kensey says.

“It starts with an enablement layer that provides what you can think of as common recipes, like starter code and frameworks that are standardized as much as we can, that we empower engineers who build things to ensure consistency across applications,” he says. “Along with this is an orchestration layer that connects with various models, enabling us to kind of swap in and out as needed, depending upon what the business need might be.”

The ability to quickly adapt is a key feature of the company’s automation tools, he adds.

“Like with any technology build, given the world is evolving and changing so fast around us, we want to make sure that we are building a foundation in a way that’s as flexible and adaptable as possible,” Kensey says. “What one model might do for us today may not be fit for purpose worldwide six months from now for potentially the same use case. It’s really about that adaptability and flexibility and not cornering ourselves to have it be a significant lift for us to adapt.”

Kensey also advises CIOs to take an active role when companies are deciding what automation technologies to deploy. In some companies, business or product leaders may be leading the charge for AI, but CIOs need to have a major voice, he says.

“If you’re going to get AI right, it’s not about developing a strategy and handing it off to technology and saying, ‘Ok, now tell me where we could leverage AI,’” he says. “It’s about technologists being at the table, hearing how they’re thinking about either the problem experience or capability they want to innovate on, and having a perspective day one on that the journey for AI.”

Early reviews of the American Express approach to AI and automation are positive. The initiatives make sense and address common pain points in corporate travel, says Arun Chandrasekaran, a Gartner analyst focused on AI.

Chatbots and natural-language BI aren’t new on their own, but embedding them seamlessly into mission-critical travel workflows at scale is a meaningful step forward, he adds.

“For travelers, a natural-language virtual agent reduces friction by complex requests quickly and intuitively,” Chandrasekaran says. “For managers, conversational analytics makes insights about spend and travel patterns far more accessible than static dashboards.”

See also:

• Walmart looks to cash in on agentic AI
• UPS transforms air cargo operations with data, AI
• Accenture reimagines IT operations with agentic AI