Three Macro Trends in GenAI for Financial Services

Generative artificial intelligence (GenAI) and agentic models are revolutionizing financial services, transforming user experience, fraud detection, payments innovation and compliance processes, according to John Kain, head of financial services market development at Amazon Web Services (AWS).

“AI has been such an integral part of how the industry has modernized over the last decade. It really is in every part of the financial services value chain,” the former JPMorgan executive said in an interview with PYMNTS. “What’s dramatically changed in the last two years is the impact of generative AI on all those processes.”

Kain sees three overarching trends across banking, capital markets, insurance and payments:

• End-to-end user experience: AI enables seamless onboarding, product recommendations, servicing and back-end processing. It is also powering hyper-personalized experiences, enabling firms to tailor offers, advice and communications in real-time based on user data and behavior.
• Data modernization: Legacy systems are being upgraded to unlock core data, enabling more personalized services and better analytics.
• Embedded financial services: Firms are embedding financial tools into broader value chains, meeting users at any channel, whether in eCommerce, social platforms or mobile apps.

In the payments space, Kain highlighted the global rise of real-time payment systems, from India’s UPI (Unified Payments Interface) to Brazil’s Pix, which offer instant settlements and low fees. In the U.S., banks are now leading the charge, according to a PYMNTS Intelligence report in December.

“These changes are transforming customer expectations,” Kain said, adding that innovations like buy now, pay later and stablecoin rails are creating new challenges for infrastructure and fraud prevention.

Machine learning is being deployed to improve fraud detection — especially for instant payments, where there’s little room for error. AWS is also seeing adoption of techniques like distributed model training for fraud and clean room environments to share fraud data across institutions.

Read more: More Than Half of US Companies Use Real-Time Payments

Kain said that AI agents — AI systems that research, interact with other agents and complete tasks for the user — are “definitely the way the industry is going.” One common use case is agents that access internal information to serve customers more effectively.

Kain cited the case of Remitly, a money transfer company serving transactions in 18 languages. For 95% of the remittances, it goes straight through without issue.

But for the remaining 5%, there could be delays due to customer identification and other wrinkles, which erodes trust, according to Remitly. The company uses GenAI to help resolve cases by things like finding the right internal information to help the customer.

Other use cases include:

• Call center automation: Real-time transcription, sentiment detection, and routing to digital channels.
• Compliance: Accelerates anti-money laundering (AML) investigations of alerts, in one case by 80% to 90% using LLMs.
• Personalization: Banks using AI can tailor marketing of not just products for the customer, but also optimal ways to pay.

When it comes to accuracy of responses and hallucinations, which are critical for financial services institutions, Kain said that models have improved on this front based on retrieval-augmented generation (RAG) and other techniques as well as having a knowledge base of trusted information. Also, AWS’ GenAI platform, Bedrock, has guardrails that use GenAI to detect hallucinations with a 75% success rate.

However, “there is still a heavy human aspect” when it comes to important financial decisions, Kain said.

Acknowledging concerns over the cost of GenAI, Kain detailed AWS’ efforts to lower expenses through custom chips (Inferentia and Trainium), model distillation, and flexible model comparison tools. Smaller, distilled models tailored to specific tasks lower costs while preserving high performance.

Kain also said that clients can choose models such as Nova, AWS’ own family of large language models (LLMs), which is 75% less expensive than comparable models from competitors when it was unveiled last December.

Also available on AWS is DeepSeek R1, from the Chinese AI startup that roiled Silicon Valley and Wall Street for its economic model. Kain said companies can use DeepSeek without concerns about their data being sent overseas. Since it’s hosted on AWS, the data stays with the client.

“We let customers test different models and price points so they can optimize performance and cost,” he said.

For example, Nasdaq was using an LLM for AML tasks but found the cost too high. It turns out they were needlessly using a very large model both for summarization and to write the final report that goes to a regulator or their internal audit teams.

Nasdaq discovered that by using a much smaller language model to summarize the facts but use the large model to write the report, they can still maintain the same quality and lower the cost, Kain said. “By mixing the two, they were able to find something that made economic sense.”

More customers will find their own optimal approach. “I think you’re just going to continue to see much more flexibility,” Kain said.

The post Three Macro Trends in GenAI for Financial Services appeared first on PYMNTS.com.