How AWS Is Working With Wall Street Giants Like JPMorgan on AI

• Amazon Web Services provides data and computing technology to other companies.
• It has become a key part of Wall Street technology, with many firms relying on it.
• Here’s how it’s helping finance firms and fintechs with generative AI efforts.

Generative AI has upped the ante in the public cloud war for Wall Street’s wallet share.

Amazon Web Services, Microsoft Azure, and Google Cloud, the cloud-computing arms of the respective Big Tech companies, have been duking it out to be the financial industry’s go-to cloud provider. Strategies to win that business extended beyond just providing technology — many are now focusing on helping clients roll out AI.

To keep its edge, Amazon said it plans to deploy around $105 billion in its business, much of which will go toward AWS and its AI efforts. For a look at how that money may touch Wall Street, Business Insider spoke with John Kain, the head of financial services market development at AWS. Kain, who worked at JPMorgan Chase and Nasdaq before switching industries, outlined the future direction of AWS’ Wall Street business through four different clients: two large international banks, a hedge fund, and a fintech.

“The tasks are getting more complex, things are becoming much more agentic in nature, much more tailored for individual use cases to get real price performance benefits,” Kain said, referring to the industry’s AI maturity.

Much of the work behind the scenes has been around cutting down on the number of hallucinations, a common issue where inaccurate responses are presented as fact.

One example is with AWS Bedrock, a service that helps customers build generative AI applications and models. Last year, AWS introduced a feature called Bedrock Guardrails that looks at the responses coming out of large language models, and then uses another large language model to check whether that answer actually was a good response. In some cases, this approach detected about 75% of hallucinations, Kain said.

Other AWS efforts, like automated reasoning, have tried to use mathematical proofs to prove that the information from generative AI models is factually correct.

Here’s how AWS is helping Wall Street firms use generative AI.

JPMorgan Chase

AWS focus area: Security and scale

When Lori Beer, JPMorgan’s global chief information officer, took the stage at AWS re:Invent in December, she charted the bank’s embrace of the cloud, which started in 2017. In 2020, JPM had 100 apps in the cloud, and it doubled that number the following year. It also built its UK consumer bank from the ground up on AWS.

JPMorgan now has thousands of applications running on AWS that fully take advantage of generative AI technologies, Kain said.

JPMorgan’s internal data and AI platform relies on AWS SageMaker, a tool for creating and training machine-learning models, with more than 5,000 employees using the cloud-based tool every month, Beer said. The bank has simplified the process of developing new models, from experimentation to deploying them live, she added.

“This platform is empowering us to build the next wave of AI applications at the firm,” she said.

Security, governance, and compliance measures in the cloud were key to JPMorgan’s uptake, according to Kain.

JPMorgan, which processes $10 trillion in payments daily and counts 82 million customers in the US, is considered the world’s most systemically important bank, according to the Financial Stability Board, which identifies global systemically important financial institutions.

“Organizations like JPMorgan, they have a unique scale and complexity that comes with being a globally regulated organization in multiple lines of business and we get to learn alongside of them,” Kain said, adding that the bank’s business compliance and security perspective helped drive AWS’ roadmap forward.

Bridgewater

AWS focus area: coordinating specialized models for investment research

About two years ago, Bridgewater pulled together a group of investors, data scientists, and technologists to rethink how the hedge fund understood markets and economies with AI and machine learning first. Thus, AIA Labs was born. Short for artificial investment associate, the division within the investment firm sought to recreate “everything that we do via machine-learning techniques,” Bridgewater’s co-chief investment officer Greg Jensen previously told BI.

“It started on a notepad, graduated to Excel, and is now running on EKS and various other AWS services,” AIA Labs CTO Aaron Linsky said at re:Invent in December.

Early on, the capabilities on the generative AI side were mostly limited to asking a straightforward question, getting the AI to figure out how to write code to exract that data from Bridgewater’s system, and produce a response, Kain said.

“That was great, it was saving hours and hours of time,” he said, and made it so investment analysts didn’t have to bug developers to get the data.

Now, Bridgewater’s AI platform can take a complex investing strategy and analyze it.

“They showed how they were able to take that complex investment question, break it into multiple steps, and have each of those steps go out to a particular agent,” Kain said. For example, one agent may check how interest rates affect overall returns, another could double-check the financials, and a third may summarize the risk profile.

“We found today that limiting the breadth of responsibilities for any given agent is really important,” Linsky said.

“We’re on the path towards full agentic workflows,” Linsky said, adding, “we’re definitely not replacing our investment associates with the capabilities right now, but it is helping to speed along their process.”

MUFG

AWS focus area: turning multiple data sets into new sales ideas

Mitsubishi UFJ Financial Group, which offers everything from investment banking to treasury management and trade finance, is using generative AI to give its corporate salespeople a leg up. An AI platform that suggests sales ideas has led to a 30% conversion rate, Tetsuo Horigane, head of quant innovation at MUFG, said at re:Invent.

MUFG began developing generative AI applications in 2023 after launching an in-house AI/ML team two years prior, Horigane said. MUFG also has about 2,000 employees serving roughly 1 million corporate clients,

The bank’s salespeople typically read through hundreds or thousands of pages of documents to understand a given client’s situation and what financial product is most relevant to them. But now the AI platform combines multiple data sets, like the client’s transaction history, previous sales conversions to understand what they’re in the market for, their financial filings, and public information like news, Kain said.

This process of drafting a sales pitch, which could take several hours or days, could now be done in a matter of minutes, Horigane said.

AWS focus area: using call center AI to influence strategy and experience

AI in call centers is nothing novel, but for Rocket Mortgage, it’s leading executives to think about whole new strategies and experiences.

The fintech integrated AWS’s generative AI technology into its call centers to lessen the load for its thousands of call center employees who field calls, emails, and webchats every day.

But the idea is not just to have one AI assistant, but “an entire network of agents,” Dan Vasquez, VP of AI strategy at Rocket Mortgage, said at re:Invent. They help with transcribing, picking out key pieces of information mid-call, and providing insights and data about the call after the fact, he said.

Generative AI has helped save some 40,000 hours annually for call center employees and enabled 70% of client support to be fully self-served, Vasquez said.

But now, Rocket Mortgage is using those 10 petabytes of data to figure out “what should we do next?”

That’s the real benefit, Kain said, for the fintech to start asking big questions like “why are my customers calling me” and “what are my most common problems,” and use that intel to rethink its online platforms, streamline workflows, and improve the overall customer experience.