How big U.S. bank BNY manages armies of AI agents

The financial services industry is one of the most regulated sectors. It also manages a lot of data. Realize the need to be cautious, financial companies own Slowly add generated AI and AI agent services.

The industry is no stranger to automation. But using the word “proxy” has been muted. Understandably, many people in the industry have accepted it A very cautious position on generating AIespecially without a regulatory framework. But now, Banks like JP Morgan and Bank of America debuts AI-powered assistant.

The most cutting-edge banks are BNY. The financial services company founded by Alexander Hamilton is updating its AI tool Eliza (named after Hamilton’s wife) and developing it as a multi-agent resource. The bank believes that AI agents provide valuable help to their sales representatives while attracting customers more.

A multi-mechanical approach

Sarthak Pattanaik, head of BNY’s AI Center, told VentureBeat in an interview that the bank is beginning to figure out how to connect many of its units so that their information can be easily accessed.

BNY has created a lead recommendation agent for its various teams. But this does more. In fact, it uses a multi-agent architecture to help its sales team make appropriate recommendations to customers.

“We have an agent that has it all [the sales team] Know[s] About our customers,” Patanak said. “We have another agent talking about products, all the products that the bank owns… from liquidity to collateral, to payments, the Ministry of Finance and so on. Ultimately…we are working hard to solve customer needs through the features we have and the product features we have. ”

Pattanaik added that its agents have reduced the number of people many customer-facing employees have to talk to to determine good advice for customers. So, “rather than talking to salespeople with 10 different product managers, 10 different customers, 10 different segments, and all of this is done through that agent.”

The agent asks its sales team to answer very specific questions customers may encounter. For example, if a customer wants to launch a credit card in the country, does the bank support a foreign currency like the Malaysian ringgit?

How they built it

Multi-agent recommendation feature debuted in BNY’s Eliza tool.

About 13 agents “negotiate with each other” to find a good product recommendation based on the marketing department. Pattanaik explains that these proxies range from functional proxies such as client proxies to segmented proxies involving structured and unstructured data. Many agents in Eliza have a “sense of reasoning.”

Banks understand their agency ecosystem is Incomplete agent. As Pattanaik pointed out, “The full proxy version will automatically generate PowerPoints that we can give to the client, but that’s not what we’re going to do.”

Patanak says banks turn Microsoft’s Autogen Bring its AI agent to life.

“We started with Autogen because it’s open source,” he said. “We’re usually a builder company; no matter where we can use open source, we can do it.”

Pattanaik said Autogen provides banks with a set of sturdy guardrails that can be used to base the reactions of many agents and make them more certain. The bank also investigated Langchain to structure the system.

BNY has established a framework around the proxy system that provides agents with a blueprint for responding to requests. To this end, the company’s AI engineers work closely with other banking departments. Pattanaik highlighted that BNY has been building mission-critical platforms for years and has expanded its products such as its clearance and mortgage platforms. This in-depth knowledge base is key to helping AI engineers responsible for the agent platform provide the agent with the expertise they need.

“The hallucination characteristics are always helpful compared to just having AI engineers drive an engine,” Patanak said. “Our AI engineers work very closely with full-stack engineers who build mission-critical systems to help us solve the problem. It’s about components and therefore can be reused.”

For example, building enablement agents in this way can develop it through different businesses of BNY. It acts as a microservice for “continue learning, rationality and action”.

Expand Eliza

As the proxy footprint expands, BNY plans to further upgrade its flagship AI tool Eliza. BNY released the tool in 2024, although it has been in development since 2023. Eliza allows BNY employees to access the marketplace of AI applications, obtain approved data sets and seek insights.

Pattanaik said Eliza has provided a blueprint for how BNY can move forward with AI agents and provide users with more advanced, intelligent services. But the bank doesn’t want to stagnate and hopes Eliza’s next iteration will be smarter.

“What we built with Eliza 1.0 is a representative and also a learning aspect of things,” Patanak said. “Using 2.0, we’re going to improve the process and ask, how do we build a great agency? If you think about an agency, it’s about something that can learn and be rational, and there’s some action for it at some point in time, it’s a break, it’s not a break, etc. This is the direction we’re going to go in when we build 2.0, because before we become completely autonomous, there’s a lot of things to build based on risk guardrails, interpretability, transparency, connections, etc.”