Banks’ “fear of missing out” around artificial intelligence is fueling related spending in the sector.

That’s one of the implications from recent findings from global consulting firm Accenture, which discovered bank leaders continue spending more on AI because they see it as essential to remaining competitive. 

Banks’ AI spending last year exceeded $40 billion, by one estimate, and companies across industries are grappling with sticker shock tied to the use of AI tools. Yet returns on those investments aren’t quite there. 

“There’s a high level of confidence they’re going to get something out of it, but it hasn't exactly been measurable just yet,” Mike Abbott, Accenture’s global banking lead, said in a recent interview.

Banks are still more focused on expense benefits than revenue generation opportunities surrounding generative AI uses, and are finding value weaving it into scaled, repeatable processes, including call center operations; underwriting processes; marketing and content generation; and regulatory reporting automation, Abbott said.

But only 20% of bank leaders said they’re seeing widespread, sustained value from AI initiatives, Accenture said, pointing to scale challenges. Accenture received responses from 212 retail banking and 110 capital markets C-suite executives across 20 countries between April and June.

“The reason why they've all struggled is: task-wise productivity – getting 10%, 15% for each person – does not add up to system-wide productivity,” Abbott said. “They're giving AI tools to individuals and saying, ‘Hey, do your job a little bit better.’” 

Most banks “have not figured out how to reconfigure the work, fully, around AI yet,” he said.

Abbott expects that to change in the coming years.

Banking “is hyper-competitive, so once one figures it out … the rest will figure it out,” Abbott said. “Banks are pretty good at copying each other.” 

Editor’s note: This interview has been edited for clarity and brevity.

BANKING DIVE: How are banks thinking about AI spending and use right now? 

MIKE ABBOTT: Especially as you look across the broad spectrum of banks, I think there is definitely a FOMO feeling driving initial investment. There's a fear of missing out in terms of not using it. 

At the same time, there's not as much of a worry right now around getting a return on investment. But many banks are starting to see that ROI.

The real question all of them are asking, when you dig underneath, is when is it going to scale, and where am I going to see the big results? 

When I’m in the board room for these conversations, it feels like everyone's looking for that magic elixir, to claim victory. The challenge with AI is there's no one person to claim victory because it's impacting every job, pretty much, in the banking world.

Do you think that FOMO has led to wasted AI spending?

I think it has, in some scenarios. You’re seeing the smarter banks realize that: How do I put AI and the right type of AI into the process? We’re starting to see an optimization occur: I’m going to use the large language models where I need them, increasingly use small language, open-source models for document ingestion and basic functions, because they’re faster, deeper and hallucinate a lot less. And there’s this emergence of transactional foundational models, where banks are using their own data to create their own models. People who have been at this for a couple of years are starting to optimize the models. 

What do you make of the AI cost backlash?

Traditionally, what you do when you budget for something is you budget for headcount. You get so many full-time employees and that's your budget. 

I saw someone recently say, “No, I'm going to change that. Let’s say your budget is $25 million for your area. Your budget is $25 million for people and AI. You figure out how you want to optimize it.” 

What you're going to start seeing is a change in the way that people think about the cost of AI, relative to the function, and they'll be returning authority to leaders to say, how do you want to optimize it? 

Right now we're in euphoria, and that's reflected in the spending. 

When do you expect to see more ROI pressure related to AI spending?

I expect ROI to take center stage next year, in a couple of different ways. Do I really need to use that advanced, large language model? There’s going to be model optimization on the pure expense side. 

And then there’s going to be optimization in terms of what should I prioritize, where I can get an ROI? 

The biggest shift you’ll see next year is this shift from serial to parallel thinking. Every bank in the world has been designed off of some form of Six Sigma process engineering over the last 35 years. Process engineering taught us to optimize against a limited number of people. But with generative AI, you can take those processes that were gated, stretch them apart, and you can parallelize. You can make your organization think in parallel. When that happens, that generates a lot of ROI.  

Can you provide examples where you expect to see that in banking? 

Think of a mortgage today: It's a serial process that you go through. Now imagine if you have AI agents that can do that simultaneously. Everything can get processed instantaneously.

To handle lost or stolen cards inquiries, today, a bank builds that capability for the call center, for their mobile app, for their website, for their branch. Imagine if you just build one lost/stolen agent, and you can project it into any of those channels.

Probably the biggest example of all is software development. Because software development is design, build, test, deploy – it's a very serial process.

We're already seeing it in a few elite banks right now, the ones that are pushing the envelope. I expect to hear more about that being mainstream next year. 

What’s essential for banks to realize AI’s value at scale?

Banks should not confuse collaboration with consensus. You want to be highly collaborative. But banks that are consensus-oriented do not execute in this space, because there's too many questions.

And you have to invest in your people. The funniest thing I get: requests for proposal saying, “Can you give me somebody with five years of generative AI experience?” And I think to myself, when you find them, you let me know where they are. Generative AI’s only been out for three years! 

The most important thing you have is your people, and investing in the talent you have.