The findings: Now that most financial institutions (FIs) have deployed or piloted genAI, some recurring lessons have emerged. Here are the biggest takeaways from a recent EY study on banks’ AI use cases:

There’s widespread use and understanding: 

• 77% of banks have launched or soft-launched genAI apps (up from 61% in 2023).
• Just 31% have started to implement agentic AI, while 99% are aware of the opportunities it offers.

The failure rate is high: Just 16% of use cases reach deployment, and 40% of implemented use cases didn’t meet expectations.

Governance is the most important step: Governance helps set expectations, parameters, and metrics before the bulk of the money is spent—helping prevent project failure and disappointment. Seventy-five percent of banks have formal AI governance committees. 

Among EY’s banking respondents, governance emerged as the most important lesson:

• 79% would prioritize governance if starting AI implementation over. 
• 79% would focus specifically on enhancing data governance to improve the quality of available data, address privacy concerns, and assist with compliance.

Engaging stakeholders is paramount to the success of pilots: Seventy-one percent of respondents would have engaged stakeholders earlier, involving them in planning meetings. This would help FIs ensure alignment and reduce delays that could arise from misunderstandings and disagreements partway through.

Our take: According to Biz4Group’s AI Banking App Development Cost Guide, AI features tend to cost FIs $40,000 to $400,000 per use case. Smaller FIs in particular can’t spend the money if just 16% of those experiments reach deployment. 

Given these risks, FIs with more limited budgets should consider starting with a highly targeted, well-governed pilot project that addresses a specific, high-value problem rather than broad experimentation.