Wall Street’s AI race: 5 things that McKinsey says will separate the winners from the losers

The bill is coming due for Wall Street banks’ AI investments.

It’s been two years since generative AI captured the attention and dollars of bank leaders. They amassed teams of technologists to experiment with generative AI and run proofs of concepts. Some of those have since scaled to enterprise-wide initiatives used by thousands of employees. Now, leaders are beginning to question when these investments will pay off.

“That is the $20 billion question,” according to Larry Lerner, a partner in McKinsey’s banking practice.

For a handful of firms, Lerner said tangible returns are starting to emerge in the form of current cost savings, future cost avoidance, and incremental revenue. But for many, the reality is “POC purgatory,” Lerner said, referring to proofs-of-concept pitfalls where firms get stuck in the experimentation phase and “become very tepid about really leaning in.” In those cases, the “institution has spent the last two years investing and investing and not seeing anything at all,” Lerner said.

According to an October report from Evident AI, which tracks AI adoption in financial services, only six out of 50 banks disclosed dollar-level cost savings or revenue lifts as a result of their AI investments.

So, what separates the frontrunners from the laggards? According to fresh research from McKinsey, it can come down to a few key decisions around concentrating efforts on a couple of uses, having CEO buy-in, and using generative AI in conjunction with other technologies. Most of all, it’ll involve a mindset shift where AI is viewed and treated as a business opportunity rather than a technological problem, Lerner said.

Lerner outlined what will separate the winners from the losers. He declined to comment on specific companies.

Viewing AI as a business problem, not a tech one

Leadership teams have to recognize that generative AI is a business opportunity, not just a technology play, Lerner said. Because of that, he said business leaders should bear the brunt of the accountability, rather than that responsibility falling solely on tech leaders’ shoulders.

“The institutions that make business leaders accountable for delivering their results will over time tend to do better because there’s a much stronger partnership,” Lerner said.

Concentrating firepower

Generative AI has lead to more value when there are only a handful of use cases, instead of every business unit doing a little bit here and there and seeing what sticks, Lerner said.

“Instead of having 60 use cases across 15 different business lines and functions, narrow down to three areas where you want to go deep,” where you’re reimagining the entire domain or workflow has led to a faster path to value, Lerner said.

Choose areas where ROI can actually be tracked

It’s becoming increasingly clear that generative AI’s main strength in saving workers time can’t always be traced back to bottom-line impact, which is leading to some frustration in the boardroom.

“The value of what you’re doing depends on how you’re going to repurpose your time, and that’s really hard to do,” Lerner said. “Because it’s an indirect sort of lever, it’s very difficult to actually measure and get people to agree that there’s value.”

On the other hand, AI tools like call-center copilots and AI-powered marketing campaigns that improve the customer experience can generate incremental value that is measurable, Lerner said. One large bank referenced in the McKinsey report is projecting a 10% revenue increase thanks to a new analytics platform to target new customers and cross-sell products to existing ones.

For buy now, pay later fintech Klarna, leveraging an OpenAI-powered call center agent is estimated to bring some $40 million in profit this year, the company said in a blog post earlier this year. At the time, the AI was doing the work of 700 full-time agents, according to Klarna.

Lerner said he’s starting to see some banks modify forward-looking hiring plans, especially in the contact center, thanks to the increase in self-service and faster resolution times. “That cost avoidance is absolutely measurable,” he said.

Reusability is key

Build something once and redeploy it a hundred times, Lerner said. Doing so can accelerate development times and let companies scale faster because the tool has already gone through the required risk, security, and compliance approvals, he said.

Execution will come down to adoption

Getting workers and customers to adopt a new way of doing something or a new technology is one of the most important parts of the value equation. It’s an old challenge that banks have had with previous technology cycles. When it comes to AI, “most companies have done a pretty bad job of getting adoption to the level that’s going to yield the results that they want to yield,” Lerner said.