The vast majority of banks have adopted artificial intelligence, but new research from nCino reveals the industry may be measuring the wrong thing entirely.
nCino's AI in Banking Benchmark, fielded in spring 2026 with 150 senior banking executives, found that 84% of US institutions are now running AI at the enterprise level. The pilot era is over. But nCino argues the more important story sits beneath that headline figure: banks are not running one AI, they are running three distinct deployments simultaneously, each with different capabilities, governance requirements, and risk profiles.
Those three deployments are generative AI, adopted by 91% of institutions; predictive AI, at 87%; and agentic AI, at 81%. Each operates differently.
Generative AI handles document drafting, summarisation, and knowledge retrieval, grounded in banking context at the point of use rather than trained on customer data.
Predictive AI, which many banks have run for years without labelling it as such, powers risk scoring, fraud detection, and credit decisioning, trained on each institution's own portfolio and loss history.
Agentic AI, the newest and fastest-growing category, executes multi-step workflows autonomously, from gathering borrower documents to routing exceptions to the appropriate reviewer.
The productivity case is well-documented. McKinsey estimates generative AI can improve productivity in core commercial and institutional banking activities by between 30% and 90%, with potential operating-profit gains of 9% to 15%. Accenture projects early adopters will see 22% to 30% productivity gains over three years.
nCino's research also highlights a significant governance gap. Each deployment demands a different oversight framework. Generative AI requires output review and source citation. Predictive AI falls under model risk management, governed in the US by SR 11-7, requiring statistical validation, documentation, and ongoing monitoring. Agentic AI, which acts rather than simply answers, demands human sign-off at every material decision point. A bank applying one governance standard across all three, nCino warns, is almost certainly overseeing at least one of them incorrectly.
The report frames orchestration as the new competitive frontier. Running one AI is a project; running all three well is an operating model.
People remain essential. nCino found 55% of banking executives are already investing in workforce reskilling, and 89% expect staff to be working alongside AI agents within five years.