Eric Lefebvre, chief technology officer at Sovos, said generative artificial intelligence (AI) will transform the ways in which consumers and businesses interact with financial institutions and merchants — and will prove essential in beating fraudsters at their own game.
Broadly speaking, he said, “the space has moved really quickly since ChatGPT came out just over a year ago.”
But the use of AI in payments will be more evolution than revolution, he said. Harnessing AI’s potential depends on tapping into large datasets.
Against that backdrop, payments throw off reams of data.
“For every transaction, every payment card transaction and ACH” taken as a whole, “there are just terabytes of data produced every single day in the normal course of commerce,” he said.
For forward-thinking companies tied to fund flows, the evolution from analytics and correlation in static models has moved to machine learning, and now, generative AI.
The progression and significant improvements in technologies have allowed AI solution providers to essentially commoditize AI itself. Every public cloud provider has some form of a large language model that is featured in their online stores, where clients can embrace those models in minutes, start training those models and see value in relatively short timeframes, he said.
Asked by PYMNTS where the low-hanging fruit might lie in the payments arena and realizing AI’s potential there, Lefebvre posited that the best results will be seen in areas where there’s already knowledge about several metrics: approval rates, fraud rates and even how consumers view their online interactions with companies and providers.
He said fraud is a particular greenfield opportunity. Fraudsters have been using AI to create a variety of attack vectors (impersonating legitimate individuals and entities), so the goal of banks and merchants is to make sure that the transaction itself is valid, and that someone sending or receiving money is who they say they are.
As payments skew ever faster and move toward real-time status, there’s an “AI arms race” in battling the bad actors, which means that companies must play offense at the same time they are playing defense.
In doing so, “we’re mining this data to create rules from that data,” as the transactions are viewed in real time, enabling enterprises to be proactive about stopping fraudsters in their tracks, rather than reactive, he said.
There’s also potential to use AI more fully in speeding up and improving the delivery of various financial services products themselves. Loan origination can be brought fully into the digital age as financial institutions rely less on paper-based bank statements and scraping databases such as LexisNexis in the bid to validate applications and their documentation, he said. Transaction authorization and approval rates get a lift as large language models can examine recurring payment status and CVV/card data to make sure that declined transactions are “re-authorized.”
On the consumer-facing side of the equation, he said, AI is finding a wide berth for fine-tuning chatbot interactions so that they are more conversational, intuitive to use and can help decrease calls that are dialed into call centers, thus lowering bank and merchant operating costs.
Looking at the months and years ahead, Lefebvre said we’ll see more attention paid to regulation and standardization of AI rules and processes. Thus far, the technology has outpaced the regulation. But as new iterations of Payment Card Industry Data Security Standard compliance mandates come to the forefront, they’ll have to address the use of AI. The initial regulatory frameworks that emerge over the near term will be fragmented but will eventually become cohesive across geographies.
“We’re at the tip of the spear in using this technology,” he said.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.