Is Generative AI More Friend Than Foe?

GenAI carries risks, some not yet well-known. But its potential to transform payments is too compelling to ignore.

Generative AI has been described as a Gold Rush for businesses looking to cash in on the better user experiences and operating efficiencies it is said to create. It has also been described as a minefield because of how rapidly the technology is evolving and the security risks associated with the vast amounts of data it accesses.

Despite its pros and cons, GenAI remains one of the hottest topics in payments because of the potential for this form of artificial intelligence to transform the business. Top use cases include customer service, merchant underwriting, personalization and targeted marketing, and fraud detection.

At its core, GenAI consists of deep-learning, artificial intelligence-based models that generate high-quality text, images, videos, or other data in response to prompts. While generative models have been used for decades to analyze numerical data, the addition of deep-learning capabilities makes it possible to extend GenAI models to images, speech, and other complex data.

Through deep learning, these models learn the patterns and structure of the inputted data, then generate new data that has similar characteristics to the data the model was trained on. As a result, the models can write engaging text or create photorealistic images with just a few prompts.

When it comes to payments, GenAI is generally viewed as a transformative technology that will propel innovation in payment processes and the user experience.

“[GenAI] lays the groundwork for enabling the creation of more secure, efficient, and bespoke payment solutions, which will ultimately prove pivotal in both accelerating innovation and creating a more competitive landscape for the payments industry,” says Netanel Kabala, chief data and analytics officer for Montreal-based processor Nuvei Corp.

Because of GenAI’s ability to make humans more productive— and to generate personalized recommendations and answers to questions—the technology is being widely adopted in customer- service centers. The technology can be used, for example, to help representatives sift through reams of data in an instant to find the correct answer to a customer’s question or create chatbots that can be as conversational as a human.

Discovering AI

In April, Discover Financial Services, which recently agreed to be acquired by Capital One Financial Corp. for $35.3 billion, announced the rollout of Google’s Vertex AI app across its call center. The rollout will provide Discover’s nearly 10,000 customer-service agents with access to intelligent document summarization and real-time natural-language search assistance during live interactions, Discover says.

Google’s Vertex AI app analyzes and summarizes complex policies and procedures to provide customer-service representatives with information to answer customers’ questions.

The impetus for adding GenAI to its call center grew out of Discover’s need to provide agents with tools to quickly and correctly respond to more complex queries, the card company says.

One advantage of GenAI for customer-service reps is that it can read hundreds of thousands of words per minute to find the right answer, payment experts say. By comparison, humans can read a maximum of 200 to 300 words per minute.

“Our customer agents’ jobs have become significantly more complex in the last five to 10 years,” says Szabolcs Paldy, senior vice president of operations for Discover. “With the emergence of digital apps, customers can get responses to easy questions, like checking their balance, from the app. Because of this, our customer-service agents are faced with more difficult questions and [are] having to sort through lengthy procedure documents to find the right answers.”

Deploying Vertex AI will help Discover’s service agents by distilling complicated answers into simple paragraphs they can share with customers quickly, versus putting them on hold for long periods while searching for an answer, says Paldy.

Since Discover began testing the technology in January, its service agents have reduced call-handling times, and improved policy and procedure search times, by as much as 70%. Plans are also in the works to use the technology to help with phone-call transcription, categorization of incoming requests by topic, and customer-sentiment analysis.

“We also use this technology to better understand customer sentiment in general and about our different products and features,” Paldy says. “We believe GenAI can help our agents in ways that previously weren’t possible.”

‘Scams And Deep Fakes’

While GenAI can significantly improve call-center productivity as well as the customer-service experience, it is being put to use in other customer-facing situations.

For example, Colleen AI, a Daytona Beach, Fla.-based software developer has added AI-based voice technology to its payment-collections app for multifamily property-management companies. The addition of AI-based voice technology will enable property-management companies to leverage GenAI to initiate personalized outreach to renters across all communication channels, such as a text, phone, and email.

In the case of a phone call, Colleen AI’s voice application answers customer questions and can accept rent payments. The app replaces chatbots, which Colleen AI chief executive Itamar Roth describes as “limited in scope by transactional, robotic responses that are not personalized,” while freeing service reps from backlogs of one-off calls.

For marketing and personalization, GenAI can be used to identify the characteristics of individuals who respond to specific marketing messages, whether it be a merchant pitching a consumer or a merchant getting pitched by a payment-technology provider, says Darryl Cumming, director of product management at payments-technology provider NMI LLC.

“GenAI can also be used in marketing to determine what pitches generate the most leads and have the highest conversion rates,” says Cumming. “That information can be used to refine our sales pitches and tailor them to the persona of the recipient.”

In addition to credit card companies and payment-technology providers, processors are making use of GenAI. Swipesum Inc., a St. Louis-based processor and merchant-consulting firm, has created a GenAI-based app called Staitments, which analyzes merchants’ monthly processing statements to determine if they are paying too much in interchange, what additional discounts may be available, and guage what they are charged in fees compared to other processors.

“There is a lot of detailed information in processor statements that merchants can use to save money on their payment-acceptance costs, but it is not always easy to identify or understand, and can be spread out across multiple pages,” says Michael Seaman, founder and chief executive at Swipesum. “With Staiments, merchants can do in minutes what used to take hours or weeks to do manually.”

Because of its ability to spot trends over reams of data, GenAI is also making its mark as a fraud-detection tool. Criminals are co-opting GenAI to develop sophisticated phishing scams, as well as so-called deep fakes that deceive humans or fraud-detection systems by using the technology to mimic legitimate user behavior, including biometric characteristics, which makes fraudulent activities harder to detect.

GenAI can combat this threat by loading known examples of phishing campaigns and deep fakes into the model and asking it to spot trends within the data.

“When data around scams and deep fakes are pushed through a GenAI model, it can lead to information that helps build models that identify these scams,” says Rajesh Iyler, global head, machine learning and generative AI, for financial services at technology consultancy Capgemini. “By spotting trends, rules can be put into place to … correct vulnerabilities within fraud-detection models.”

Some processors are using GenAI to underwrite a prospective new merchant. NMI, for example, uses the technology to determine when it needs to scrutinize a prospective merchant and when it is OK to auto-approve a merchant, Cumming says.

Key to using GenAI to detect fraud scams created with AI models is constantly updating the model on techniques criminals are using. “Continuously updating the models with fresh data and adjusting them in response to evolving attacker techniques is crucial for maintaining their effectiveness over time,” says Cynthia Printer, director, financial crime compliance and payments, at LexisNexis Risk Solutions.

‘Hallucinations’

While early adopters of GenAI in the payments business are seeing promising results, experts note the industry is just starting to realize the technology’s potential.

That can be a double-edged sword, payments experts warn. While technology can open the door to innovation in payments, users need to be mindful the technology is only as a good as the data it is trained on.

“Generative AI can predict what information should be provided and mimic human functions based on training, but the technology can hallucinate,” says Siva Narendra, cofounder and chief executive at Tyfone Inc., a Portland, Ore.-based provider of digital-banking solutions.

A “hallucination” is an incorrect or misleading result generated by GenAI model. Hallucinations can be caused by insufficient training data, incorrect assumptions made by the model, or biases in the data used to train the model.

When hallucinations occur, consumers or businesses interacting with GenAI models can get led down the wrong path. “That’s worse than saying, ‘I don’t know the answer,’” says Cliff Gray, principal at Chicago-based Gray Consulting Ventures LLC. “It’s not so much about how to implement GenAI, but how to implement it correctly.”

GenAI users also need to keep in mind that because its models connect to critical back-end databases, such as customer relationship management and enterprise resources planning apps, those connections must be secure to prevent opening a back door that makes the data vulnerable to criminal exploitation.

“While AI brings numerous benefits to the payments industry, it also comes with risks. Among these are the potential for algorithmic bias, a lack of transparency, data-privacy concerns, and the ability to impersonate others, which can facilitate scams,” says Brenton Harder, head of enterprise automation at Fiserv Inc.

Lastly, GenAI users are advised to keep in mind that advances in the technology are occurring at breakneck speed, which means today’s trends can be obsolete tomorrow.

“We’ve gone through three generations of GenAI in two years, which makes staying on top of the technology tough,” Narendra says. That said, he adds, the technology is “too good to ignore.”