Agentic AI: The next level of artificial intelligence in banking
In recent years, artificial intelligence (AI) has established itself in banking as a powerful assistance technology that extracts, condenses, formulates, analyses or optimises specified information (Financial Stability Board, 2017; Bank for International Settlements, 2024). To date, its use has focused in particular on areas where AI can perform clearly defined, operationalisable tasks with increased efficiency. Examples include applications in customer interaction (e.g. chatbots), fraud detection, anti-money laundering and counter-terrorist financing, and risk assessment in the insurance and credit business (Bank for International Settlements, 2024; European Banking Authority, 2025; Bank of England, 2024; European Central Bank, 2025). Essentially, traditional AI models in these fields of application operate within narrowly defined, a priori defined spheres of action and decision-making and continue to rely on human control and intervention in key process steps (cf. IBM, 2026).
As a logical further development of this paradigm, the focus is shifting from pure assistance to systems that not only support tasks but also execute them in a targeted and independent manner. In the course of this further development, “agentic AI” is increasingly coming into focus. IBM (2026) defines agentic AI as AI systems that are designed to achieve defined goals largely independently and with only a minimal amount of supervision. These systems are based on AI agents, which are learning-based models that replicate human decision-making behaviour and process tasks in real time according to the situation (IBM, 2026). This represents a shift in emphasis from primarily reactive assistance to AI systems that proactively, i.e. goal-oriented and largely independently, drive tasks forward and specifically involve the user (McKinsey & Company, 2025). Examples of tasks include interpreting goals, breaking them down into individual tasks, interacting with people and systems, executing activities and dynamically adapting them, and communicating with systems (McKinsey & Company, 2025; IBM 2026). Figure 1 below describes the general steps that agent-based systems take to execute their operations.
The paradigm shift is technical and has a strong operational impact
Figure 1 outlines the general process steps of agentic systems. They range from data collection and target interpretation to continuous adaptation and orchestration, and establish a qualitatively new, independent operationalisation principle compared to classical AI (cf. Gonzalez et al., 2026; Elshan et al., 2022). For banks, the difference between classic and AI-based automation and agentic execution is therefore not only semantic, but above all operational in nature: agent-based systems learn independently to develop successful strategies that lead to the best possible results (cf. Silver, 2016; Silver et al., 2017). In particular, goals are broken down into subtasks, interim results are validated, tools are used selectively, and process paths are adapted to the situation (cf. Deloitte, 2025; IBM, 2026). This represents a profound change for day-to-day banking operations, as it is not just a matter of efficiency gains, but also of how autonomous, AI-supported decisions in particular can be made in accordance with the applicable regulatory requirements, documented in a comprehensible manner and monitored effectively. In order to capture these banking-specific effects not only selectively, but also in their breadth and depth, a regulatory framework is needed that systematically highlights the changes along the internal value chain of banks.
Agentic AI in the banking model: from assistance to controlled execution
A banking model is a suitable regulatory framework for this purpose, as it enables systematic classification and makes the subsequent effects along the banking processes and value chain structured and traceable. The Gabler Banking Dictionary describes the banking model as a structured view of all the functions of a banking operation, which systematically relates selected model objects depending on their purpose and target audience (Alt & Zerndt, 2020). The implications of agentic AI along banking processes are systematically presented and discussed below.
Management processes
In management processes (planning, control and monitoring), the leverage lies less in the pure automation of customer processes than in the operationalisation of control signals: agentic approaches could continuously merge key figures, risk indicators and deviations and convert them into verifiable proposals for action with documented assumptions. The NIST AI Risk Management Framework can serve as a reference framework for managing AI risks. On the banking side, it is particularly applicable where governance, monitoring and traceability are understood as integral requirements for model-based decisions (cf. National Institute of Standards and Technology, 2023; Board of Governors of the Federal Reserve System, 2011). From a stability perspective, the analysis by the Financial Stability Board (FSB) in this context emphasises in particular that the increasing use of AI in finance raises the requirements for monitoring and the adequacy of supervisory and regulatory frameworks (Financial Stability Board, 2024).
Sales processes
In sales processes (channel management), added value arises in particular from the structured preparation and execution of the entire customer interaction. Agentic systems could, for example, not only summarise a consultation occasion, but also convert it into a sequence of information gathering, clarification questions, document compilation, conversation structure and follow-up. Among other things, systems are mentioned that prioritise target customers, automatically track contacts, prepare customer meetings with structured dossiers, support condition and price decisions with current signals, and derive information for the further development of customer advisors from meeting minutes (McKinsey & Company, 2025).
Transaction and settlement processes
In transaction and settlement processes (initialisation, recording, verification, approval, processing), Agentic AI is particularly convincing where multi-stage processes currently get stuck due to media breaks, exceptional cases and incomplete information. Agentic approaches could complete transaction data in a targeted manner, check plausibility sequentially, control internal systems using tools, consolidate the necessary evidence and transfer exceptions to structured processing on a case-by-case basis. The analyses documented by ECB Banking Supervision on creditworthiness, scoring and fraud detection provide a practical reference for banks (European Central Bank Banking Supervision, 2025).
Transaction-related processes
In transaction-related processes (monitoring, transaction management, exception handling), the strength of agent-based systems lies in the structured execution of clarifications and the traceable processing of exceptional cases. Agentic approaches could, for example, generate a consistent case dossier from alerts and transaction traces, structure chains of evidence, trigger clarifications with internal departments or systems, compare interim results and control processing in such a way that the case history remains traceable and auditable. In particular, according to an analysis by the Bank for International Settlements (BIS), the use of AI for fraud detection, including the fight against money laundering and terrorist financing (AML/CFT), is one of the most prominent areas of application (Bank for International Settlements, 2024).
Cross-transaction processes
In cross-transaction processes (including customer and account management, product development and maintenance, risk and internal monitoring), the contribution of agentic AI lies primarily in combining cross-departmental work steps into a continuous process. For example, an agentic system can consolidate events relevant to a customer portfolio, verify anomalies, obtain the necessary data from different systems, generate a consistent view for reporting and internal control, and prepare and transfer the next work steps to the relevant departments. In product maintenance, it can trigger changes to parameters or documentation, check the necessary dependencies, and coordinate consistent updates across channels, systems, and reports. In the context of risk and reporting, it can identify data gaps, initiate subsequent deliveries, perform consistency checks, and use this information to create a traceable report draft that transparently identifies sources and assumptions.
Support processes
In support processes (human resources and accounting, document management, legal reporting, procurement, IT/security), Agentic AI is particularly useful where many sub-steps currently have to be coordinated across systems before a consistent result can be achieved. Agentic systems could, for example, consolidate documents relating to a process from different sources, request missing information in a targeted manner, initiate checks and approvals, monitor status and deadlines, and use this information to generate standardised output documents, such as internal reports, templates or notifications. In procurement, they can structure recurring processes such as requirements clarification, bid comparison and documentation. In document management, they can complete files neatly and prepare them for audit. In accounting, they can coordinate clarifications in the event of discrepancies and compile the necessary evidence.
Strategic implications: What needs to be decided now
The strategic question is not whether agentic AI will come, but in what form banks will make it “bankable”. In particular, the greatest leverage is located where banking processes are not automated selectively, but redesigned end-to-end (McKinsey & Company, 2025). In a similar vein, it is emphasised that agentic scaling across extensive banking functions requires a corresponding level of integration and governance maturity (Deloitte, 2025). Anyone who wants to substantially anchor agentic AI in banks will therefore not treat it as an extension of individual assistant functions, but as a design task for a bank-wide target operating model. The banking model serves as a regulatory framework for precisely locating effects and responsibilities along the value creation processes and functions. It is crucial that processes and functions, role and responsibility logics, and governance and control mechanisms are coordinated in such a way that agentic systems can take on comprehensible tasks in clearly defined process sections without undermining the bank’s management and control architecture. This includes, in particular, defining system access, documentation and verification paths, approval and escalation mechanisms, and monitoring as integral components of process design.
Bibliography
Alt, R., & Zerndt, M. (2020). Bank model. In Gabler Banklexikon. Springer Gabler.
https://www.gabler-banklexikon.de/definition/bankmodell-70673/version-377473 (Banking Act).
Bank for International Settlements. (2024). Regulating AI in the financial sector: Recent developments and main challenges (FSI Insights on policy implementation No. 63). Financial Stability Institute. https://www.bis.org/fsi/publ/insights63.pdf.
Bank of England. (2024). Artificial intelligence in UK financial services.
https://www.bankofengland.co.uk/report/2024/artificial-intelligence-in-uk-financial-services-2024.
Board of Governors of the Federal Reserve System. (2011). Supervisory guidance on model risk management (SR 11-7).https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm.
Deloitte. (2025). Agentic AI in banking. Deloitte Insights.
https://www.deloitte.com/us/en/insights/industry/financial-services/agentic-ai-banking.html
Elshan, E., Zierau, N., Engel, C., Janson, A., & Leimeister, J. M. (2022). Understanding the design elements affecting user acceptance of intelligent agents: Past, present and future. Information Systems Frontiers, 24(3), 699–730. https://doi.org/10.1007/s10796-021-10230-9.
European Banking Authority. (2025). AI Act: Implications for the EU banking and payments sector. https://www.eba.europa.eu/sites/default/files/2025-11/d8b999ce-a1d9-4964-9606-971bbc2aaf89/AI%20Act%20implications%20for%20the%20EU%20banking%20sector.pdf.
European Central Bank Banking Supervision. (2025). AI’s impact on banking: use cases for credit scoring and fraud detection.
https://www.bankingsupervision.europa.eu/press/supervisory-newsletters/newsletter/2025/html/ssm.nl251120_1.en.html.
Financial Stability Board. (2017). Artificial intelligence and machine learning in financial services: Market developments and financial stability implications.
https://www.fsb.org/uploads/P011117.pdf.
Financial Stability Board. (2024). The financial stability implications of artificial intelligence.https://www.fsb.org/2024/11/the-financial-stability-implications-of-artificial-intelligence/
Gonzalez, G. R., Habel, J., & Hunter, G. K. (2026). AI agents, agentic AI, and the future of sales. Journal of Business Research, 202, 115799. https://doi.org/10.1016/j.jbusres.2025.115799
IBM. (2026). What is agentic AI? IBM Research.
https://www.ibm.com/think/topics/agentic-ai.
McKinsey & Company. (2025). Agentic AI is here. Is your bank’s frontline team ready?
https://www.mckinsey.com/industries/financial-services/our-insights/agentic-ai-is-here-is-your-banks-frontline-team-ready.
National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0) (NIST AI 100-1).
https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf.
Silver, D. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529, 484–489.https://www.nature.com/articles/nature16961
Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I., Huang, A., Guez, A., … Hassabis, D. (2017). Mastering the game of Go without human knowledge. Nature, 550, 354–359.
https://www.nature.com/articles/nature24270
