Mastercard adopts new gen AI tech to onboard customers | PaymentsSource

Mastercard is accelerating its generative AI strategy as the technology makes inroads in the financial services industry, providing a source of new business for the card brand that does not rely on card fees.

The payment company this week added a gen AI-powered digital assistant to its customer service mix, following earlier gen AI-related moves that focused on security. Mastercard is using Databricks’ data intelligence platform to train the gen AI engine that supports the card brand’s onboarding assistant. This is designed to automate routine tasks and generate responses to customers with less human supervision.

A new nugget for Mastercard’s digital assistant is retrieval augmented generation technology — an advanced AI tool that uses information tailored to a specific task from a dedicated organization. RAG uses Mastercard’s existing onboarding documentation to locate the accurate information needed to respond to user prompts.

This is designed to improve the performance of “older” gen AI that uses less granular data. “There are a number of steps when a new product is being onboarded,” said Greg Ulrich, chief AI and data officer at Mastercard. “This tech can get answers to consumer questions faster than we could otherwise.”

The market stakes

Gen AI accesses data and user prompts to produce original content and has led companies in financial services and other industries to explore what “human tasks” can be replicated. Since being introduced in late 2022, gen AI is starting to demonstrate results. Financial services companies are saving between 5% and 8% in productivity from using gen AI, according to Capgemini. 

Mastercard’s gen AI strategy includes a May partnership with Salesforce to support payment dispute resolution. 

Mastercard in May also deployed gen AI to detect compromised cards faster. The real-time decisioning product helps banks produce risk scores and approve transactions. Gen AI scans additional data points to predict whether a transaction is likely to be genuine, building Mastercard’s ability to analyze account, purchase, merchant and device information in real time. Called ID Pro, it has improved fraud detection rates by an average of 20% since May, according to Mastercard. 

The card brand will next determine how gen AI can be used beyond onboarding and security. “If a consumer has a question about a product, for example, it can be helpful,” Ulrich said. “How can we make our systems smarter and more intelligent, and make better recommendations to our customers?”

The AI assistant announcement came in close proximity to a non-AI related release at Mastercard designed to minimize manual steps in payments. Mastercard is combining biometric authentication with tokenization, or the replacement of card account numbers with a temporary value that makes the card useless if stolen. By combining biometrics with tokens, the card brand hopes to eliminate the need for manual card entry for e-commerce purchases by 2030.

By using AI and other technology, Mastercard hopes to compete with other payment companies that are also using gen AI to streamline work patterns.

Visa’s gen AI strategy includes a security component, using the technology to alert banks about threats, particularly those that use bots or other forms of machine learning. PayPal under new CEO Alex Chriss has invested in gen AI to support less navigation and faster checkout and shopping tools for merchants and consumers. Klarna has deployed gen AI to most of its workforce and has changed its recruitment strategy based on AI’s impact on manual work. 

“Gen AI can integrate data sources more rapidly and readily,” Ulrich said. “And it can create new content to answer questions.” 

Still challenging

Gen AI is evolving from general-purpose Large Language Models into more specialized models that are trained to understand context and perform a few tasks very well, according to Celent. This includes new AI designed to operate with little or no human supervision, or “agentic AI.”

“The Mastercard announcement is very much consistent with some of our findings of where the ‘puck is going’ in terms of investments into Gen AI by payments companies,” said Zil Bareisis, a senior analyst at Celent. Celent’s research, which was conducted with AWS, cautions against inflated expectations for these cutting-edge versions of AI. Payment companies must consider what they can do with GenAI in the next three years, while preparing for the time past 2028, according to Celent. 

“With AI, finding the right fit is more important than moving fast,” said Christopher Miller, lead analyst for emerging payments at Javelin Strategy & Research. The overlapping regulatory frameworks and the need for payments to be exact and correct make the most obvious short-term use cases for gen AI on the fringes of payments, rather than playing a direct role in processing. 

“This means the well-defined use cases such as customer service and fraud reduction, well out of the direct line of payment,” Miller said. “Mastercard appears to have found a good fit with this product, staying on the periphery of the payment flow.”

However, because AI technology will continue to change rapidly and find new delivery models, the “copilot” or “human review” framework that is widespread in late 2024 won’t age well, Miller said.Mastercard is focused on its own onboarding as an initial use. One of the challenges of using gen AI or related technology for onboarding is selling it to clients for their own use, Miller said. 

“A key question for other companies considering either developing or using a tool like this is whether or not they have access to the data necessary to train models, and if they have the technical talent to assemble multiple AI technologies into a stack that delivers successful products.”