Regnology finds agentic AI gap in reporting adoption

Regnology has published research on the use of agentic AI in regulatory reporting, based on a study of 276 practitioners across 22 countries.

The report identifies what it calls an "agentic gap" between AI pilots and production use in financial institutions. Most organisations have begun testing or exploring AI in operations, but only a small minority have embedded it into live regulatory reporting processes.

According to the findings, 87% of respondents are exploring, piloting or embedding AI in operations, while 13% have no current plans. Production use remains limited: embedded use stands at between 8% and 15% across every institution tier, and 16% of respondents overall say they have reached that stage.

The main obstacle, the research argues, is not the underlying technology but the controls around data, governance and operational trust. It sets out a framework for institutions to assess where AI can take on parts of the reporting workload and where human supervision remains necessary.

Separate analysis by Oliver Wyman, commissioned for the study, suggests the financial incentive is significant. Regulatory reporting typically accounts for 1% to 3% of total expenditure at the banks profiled, with 30% to 50% of reporting spend tied to internal process execution.

The study describes this internal running cost as the largest single cost pool in reporting, reflecting how much work is still completed manually. In illustrative Tier 1 bank case studies, Oliver Wyman estimated that about 15% to 25% of reporting spend could be addressed by agentic workflows, although the study said this figure was directional.

Skills shortage

Moving AI from trial use into live reporting requires more than software investment, according to the report. It points to the need for changes in operating models, stronger governance and people who understand both regulatory requirements and how those rules can be translated into controlled system behaviour.

"Banks are conservative by nature, and in regulatory reporting they are right to be. The tolerance for error is close to zero, and a manual process may be inefficient, but it is familiar and readily defensible. The harder part is rarely the technology. It is finding people who understand the regulatory logic in depth and can turn it into something a system can apply safely. That combination is scarce, and it is what Regnology brings alongside institutions rather than asking them to assemble it alone," said Rob Mackay, Chief Executive Officer of Regnology.

The study presents adoption as uneven in execution rather than awareness. Many firms, it argues, have already accepted that AI will play a role in regulatory operations, but have not yet put in place the controls needed to rely on it in recurring reporting cycles.

Its proposed approach is to map AI authority to the risk and repeatability of each workflow. Institutions should baseline existing processes before piloting changes, build governance into systems before deployment and ensure outputs can be traced and reconstructed.

Process by process

The study also highlights regulatory scrutiny of AI systems. Institutions need to align deployment with requirements such as the EU AI Act, particularly in functions where reporting accuracy and accountability are central.

"Agentic is not one thing. It runs from explaining what the numbers mean, to recommending what should happen next, to carrying out defined work under human oversight. How much authority each process can carry is a decision institutions make process by process, not once for the whole organization," Linda Middleditch, Chief Product & Engineering Officer of Regnology, said.

Regnology linked the findings to its own product strategy through RGI, which it describes as an intelligence layer combining explainability, AI-assisted decision support and agentic workflows under human oversight. The report presents that model as a response to institutions seeking to automate parts of reporting without removing accountability from human operators.

Respondents came from financial institutions, professional services, technology and supervisory bodies across Europe, North America and Asia-Pacific. The research was conducted between February and July and focused on how far organisations have progressed from experimentation to production in AI-led reporting operations.

One of the clearest messages in the data is that the scale of an institution does not appear to guarantee faster production adoption. Resources may support experimentation, but embedded use in reporting has remained low across all tiers, suggesting common barriers in data quality, governance design and internal confidence.

This leaves regulatory reporting as a potentially costly but structured target for AI deployment. With reporting absorbing up to 3% of total bank expenditure in the banks studied, and a large share of that spend concentrated in internal operational work, the report argues that the gap between pilot projects and trusted production has become the central issue for financial institutions.