Despite proven use cases, many lenders are fumbling opportunities to unlock the potential of AI.  

Of more than 200 surveyed lenders, 38% struggle to see a return on investment from AI, according to a Jan. 14 report by global technology and data provider Experian.  

One reason is that many lenders underestimate or overestimate the need for humans to augment AI tools while trying to find the right balance of automation and labor, Vijay Mehta, executive vice president of global solutions and analytics at Experian, told FinAi News.  

In addition, the financial services industry is still trying to figure out how to measure the success of AI against traditional financial metrics, Mehta said. Some banks, for example, are choosing to maintain headcount on their compliance teams and prioritizing AI to increase human surplus, he said. 

“But then you get into this very nuanced measurement of what is human surplus, and how do you actually get a baseline for it,” he said. “The other part of this is there are so many tools right now. It’s overwhelming for a typical business to kind of identify which tools are right.” 

Over piloting  

Prolonging the pilot phase is another mistake that lenders make when deploying AI, Mehta said. 

“It’s almost like a death by proof of concept. The goal for any initiative should always be solving a problem No.1, and then getting it out of a test environment and into a production environment.” 

— Vijay Mehta, Experian

For example, 80% of AI pilots in financial services never make it to production, according to a 2025 report by technology provider Fintellect AI.  

While some testing is necessary, lenders must be disciplined in reinventing workflows and accept the possibility of failure, Mehta said.  

“If you work backward and you say, ‘We’re going to set our metrics up front and define what success looks like, and we’re comfortable saying this may not be the answer for every problem,’ most organizations are going to have better success,” he said.  

FIs also must overcome organizational resistance, which can lead to AI project failures, Sonata Bank Chief Innovation Officer Will Rhoads told FinAi News. 

“Once it goes live, you start having exceptions,” he said. “People don’t trust the system. They go back to the old ways, to the tribal knowledge, and it falls apart.”  

Turning the corner?  

Excessive testing has hindered some AI deployments, but it’s possible that financial institutions are beginning to learn from their mistakes, with 67% of lenders planning to complete or implement generative AI strategies in 2026, according to a Nov. 13 report by financial services research and advisory firm Celent.  

With generative AI it’s striking “how quickly financial institutions are moving from pilots to production,” Craig Focardi, principal analyst at Celent, stated in the report. 

“The urgency to compete on both efficiency and customer experience is turning gen AI from a curiosity into a core capability,” he said.  

Further signaling this shift, banking giant Lloyds Bank plans to bring agentic AI pilots into production this year.  

Register here for the inaugural FinAi Banking Summit, taking place March 2-3 in Denver. View the full event agenda here.