A survey of 250 banking and financial services organisations across the US, UK, France, Germany, and the Netherlands finds the AI market at an inflection: the majority are no longer experimenting. Forty-two percent say they are running AI in production at scale across multiple teams and systems, and 83% increased AI investment in the past year, with 38% reporting significant increases.
Scaling is where the difficulty concentrates. Forty percent of organisations say AI initiatives stall when they try to extend them organisation-wide. The research, conducted by Hanover Research in June 2026 and published in a white paper titled The Trust Gap: How Regulated Enterprises are Setting the Bar for Agentic AI, points to regulation and governance as the dominant friction points. Seventy-eight percent of respondents say regulatory considerations significantly or extremely limit their ability to deploy AI. Auditability ranks as very or extremely important for 88%. The top requirement before trusting AI in production: strong governance controls and approval workflows.
The shift toward agentic AI shows up clearly in the data. Forty-three percent of firms say conversational AI is too dependent on prompt quality for practical operational use, and the same proportion are already blending conversational and agentic approaches. Among organisations with more than one billion dollars in annual revenue, that figure rises to 47%. Half the organisations surveyed identify anomaly detection and signal correlation across systems as the leading AI use case they are ready to deploy agentically.
The mainframe skills picture is acute: 81% of leaders rate their organisations' mainframe skills gap as very or extremely significant. Yet the study finds that most organisations view mainframe infrastructure as an AI asset rather than a liability — 87% believe AI will close that gap within two years. Compliance is now outpacing traditional cost-reduction as the investment rationale: 44% cite reduced compliance risk and easier audits as the primary trigger for further AI funding, compared with generic efficiency arguments.
On the return side, 91% of respondents say AI-driven diagnostics are credible for reducing mean-time-to-resolution, with most expecting a reduction of more than 20% when implemented well.
"Enterprise AI has clearly moved beyond the pilot phase, but scaling is where many organisations hit a wall, with regulatory, auditing and governance considerations common barriers," said Neil Fowler, SVP of hybrid cloud engineering at Rocket Software. "As we move beyond the chatbot, advanced agentic AI solutions are proving they can support mission-critical systems. What's more, while the mainframe skills gap is a known challenge, this Study shows that the mainframe itself is an AI advantage. With the right AI partner and agentic platform in place, organisations can confidently scale their AI initiatives, reduce operational complexity, and modernise without disrupting existing systems."
The full study is available as a white paper from Rocket Software.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
