Agentic AI is becoming commonplace in financial services as tech providers emerge with end-to-end agentic frameworks.

More than 25% of institutions are deploying agentic AI frameworks, with financial institutions warming up to the tech, according to data shared with Bank Automation News by Craig Le Clair, vice president and principal analyst at research firm Forrester.

The global market for agentic AI in financial services is projected to grow from $2.1 billion in 2024 to $80.9 billion in 2034, according to a Feb. 13 report by research firm Market.us.

In the past six to nine months, there has been a major shift from basic automation to AI-driven agents that orchestrate entire processes, with more than 200 startups flooding the space, joining tech giants such as UiPath, he said.

One example is AI-driven financial services platform Hapax, which announced its Agentic Workflows tool this week. The solution uses customizable AI agents to manage certain processes for FIs, including document analysis.

The fintech will add to its agentic AI offerings on May 30, when it rolls out a voice assistant that builds on the solution, Chief Marketing Officer Kevin Green told BAN.

Hapax’s product launch is one of multiple agentic tools announced recently:

Customized AI

Hapax’s Agentic Workflows is a domain-specific, customizable tool designed to help FIs securely implement AI-driven automation within their existing systems, Green said.

Hapax began developing the system about eight months ago after consulting with more than a dozen banks to identify operational inefficiencies, Green said.

“Processes that used to take days now take minutes and just require a human to validate,” he said.

Key features of Agentic Workflows include:

• Single-tenant large language models to give FIs full ownership and control;
• A no-code workflow builder; and
• Integration with Hapax’s Intelligence Core, a knowledge graph launched in 2024 with Hapax’s platform that links internal documents, policies and real-time regulatory updates.

The voice assistant will convert the capabilities of Agentic Workflows into a phone call format, Green said. Both tools are available for free to FIs already using the Hapax platform, he said.

The $1 billion Velocity Credit Union is in an early stage of using Agentic Workflows from Hapax’s platform, Green said.

Other FIs using Hapax’s platform include:

• $1.5 billion American Bank of Commerce; and
• $500 million Capra Bank.

Keeping LLMs in check

In addition to streamlining document management, the Intelligence Core capability prevents Agentic Workflows and the succeeding voice assistant from being too agreeable or producing outdated documents, he said.

Hapax specifically built Intelligence Core to inform the FIs when information is insufficient for generating an accurate outcome, Green said.

“What that’s telling the bank is, ‘Hey, you probably haven’t uploaded a policy or a procedure.’ You haven’t informed the Intelligence Core that this is important,” he said, adding that early testing and adoption of Agentic Workflows suggest a 99% accuracy rate, compared with the 60% to 70% accuracy of generalized tools such as ChatGPT.

The single-tenant, self-contained architecture of Agentic Workflows also provides FIs with a sense of security, Green said.

“Everything from the front end down to the database and the LLM is under their control,” he said, adding that they can integrate with their core systems without fearing data leakage that can occur using a large provider such as OpenAI or Microsoft.

“The LLM is contained — no data leakage, no hallucinations,” he said.

Optimizing ROI

Domain-specific AI models can offer lower cost and higher reliability than general LLMs by using filtered, finance-specific data via a semantic layer or knowledge graph — reducing hallucinations and improving trust, albeit at the cost of broader flexibility, Forrester’s Le Clair said.

Fintechs that can develop domain-specific, yet amendable agentic solutions could have an advantage in financial services, said Le Clair, who authored “Random acts of automation: How to fight back when automation threatens your work, your life, and everything you do.”

Generic AI agents lack full access to financial institutions, which lowers their ceilings, lifespans and, ultimately, their return on investment, he said.

In contrast, domain-specific, customizable AI agents give financial institutions complete control, enabling FIs to scale the tech and tailor it to their specific operations, he continued.

Hapax’s Green said, “There’s the saying: An LLM built for everything is an LLM built for nothing, because it doesn’t have the specialization. It doesn’t have that inherent knowledge of how things work.”

While IBM and Oracle’s watsonx Orchestrate on Oracle cloud solution is general purpose, IBM helps FIs build and scale AI agents to create an integrated agentic ecosystem aligned with their priorities and operating models through its IBM Consulting arm, an IBM spokesperson told BAN.

Banks that work with IBM include:

• $3 trillion HSBC;
• $2.5 trillion Bank of America;
• $940.7 billion Natwest; and
• $536 billion Truist.

What’s next in agentic AI

Today, most agentic AI use involves basic augmentation like summarization, said Forrester’s Le Clair. However, many institutions plan to move toward higher autonomy in AI, with the emergence of “worker agents” as a broadly adopted tool expected by 2027, Le Clair said.

Hapax, for one, is planning to roll out new AI tools approximately every two weeks through July as FIs increasingly demand flexible AI options, Green said.

“Our belief is that AI is going to be that next transformation in banking, and we don’t want financial institutions to make decisions on the technology they’re going to use, not realizing that six to 12 months down the road, the choices they make may not scale based on how they operate,” he said.