18 Aug Process to Prediction: Exploring Agentic AI in Financial Services
Financial services institutions are navigating an environment of growing operational complexity. Evolving regulatory pressures, dynamic risk landscapes, and the demand for operational resilience encourage a continuous exploration of how core business processes are designed and executed. While existing capabilities aim to provide a stable foundation, new technologies offer pathways to enhance operational effectiveness.
One such concept is agentic AI. This approach involves deploying autonomous/semi-autonomous actor, or “agents” to perform complex tasks. It’s helpful to think of an agent as a digital specialist assigned to operate within a specific domain. With a defined level of autonomy allowing for strict, pre-configured policy boundaries. They can perform tasks analyse data, make decisions, and initiate actions in response to changing conditions.
Crucially, these agents can be designed with observability in mind. They can operate individually or as part of a larger, coordinated system known as an agentic mesh. In this mesh, multiple agents can work together on complex tasks, with their interactions orchestrated and managed to achieve a broader objective, all while remaining aligned with institutional governance & security requirements.
Exploring Practical Applications: From Risk to Resilience
The potential of agentic AI is best understood by exploring how it could be applied to critical workflows, introducing a level of precision and responsiveness that could augment current processes.
- Early Warning Signals: Given LLM and agentic framework’s capabilities to scan and disseminate information quickly, the ability to adapt to changing circumstances in real time, is a powerful capability agentic AI can enable. It has broad swathe applications across the banking context, in areas such as markets & investment risk, climate risk and operational decision making.
- E.g. Operational & Climate Risk: An agent could be configured to continuously monitor a vast array of data sources: from internal systems to global regulatory news. Upon detecting an emerging risk pattern or a change in climate risk reporting standards, it could automatically draft summaries, trigger alerts for human review, and update preliminary risk models.
- Decision & Task Automation: A significant amount of banking process by design, involves a high degree of manual activities and processing. These are designed with risk in mind, however they can be significantly resource intensive. The ability to automate low risk or complexity, high volume activities is unlocked by the versatility of AI agents to complete various tasks. This becomes table stakes if it’s a competitive advantage being realised by a competitor.
- e.g. Stress Testing & Operational Resilience: In the context of APRA standards like CPS 190 (Recovery and Exit Planning) and CPS 900 (Resolution Planning), agents offer a new approach. An agent could run dynamic simulations of disruption scenarios using live operational data, testing the viability of recovery strategies. This has the potential to move planning from a static, periodic exercise toward a state of constant readiness.
- e.g. Customer Contact Automation: Customer contact and maintaining direct and efficient communication with customers has long been a bug bearer within banking organisations. Agents offer the ability to do this in a seamless way, by automating less complex interactions. While allowing local client service teams to deal with high value interactions. The ability to provide personalised and targeted communications, through agentic AI allows for seamless Customer Experiences to develop.
Building the Foundation: A Framework for Implementation
For any organisation exploring agentic AI, a strong foundation is the key to success. If you can imagine the desired target state is to develop an efficient digital service plane, across the entire organisation. This will allow tasks to be automated, and efficiencies blended in seamlessly.
- Take stock of your existing processes and identify opportunities for enhancement: Intelligent systems require clear instructions. Critical workflows should be thoroughly mapped, with decision points, data dependencies, and success metrics clearly defined. Taking stock of how these processes work, can help identify points of integration or efficiency.
- Aspire for perfect data, know what is good enough to get going: Agents are only as effective as the data they consume. The ability to consume organisational data sources of sufficient quality, allowing for effective service integration, will be a key driver of quality agentic outcomes. A modern data platform with robust APIs is essential to ensure agents have timely and accurate information to act upon.
- Support autonomy and governance: Trust is paramount. Organisations must have clear policy boundaries, defining the scope of an agent’s actions, its decision-making thresholds, and the protocols for escalation to human experts. This can be enabled through deterministic graphs, or agentic mesh, as previously alluded to. Ensuring that the right tasks can be automated to the right levels of autonomy and risk appetite.
- Maintain monitoring and observability: To ensure control and build confidence, all agent activities must be transparent. Maintaining comprehensive logging, audit trails, and feedback loops is integral for oversight. Integration into SIEM and alerting of key decisions is a prerequisite here, we can only manage what we can measure, and this level of visibility will be crucial to ensure safe and transparent integration of AI.
- Strive for pragmatic, incremental adoption: An exploratory journey should begin with a focused, high-value use case to demonstrate potential and build internal expertise. As the organisation matures, these capabilities can be scaled and orchestrated. As patterns prove their efficacy and viability has been showcased, these approaches can be cascaded across other domains and business areas. We also strongly recommend that these are underpinned by building scalable building blocks for AI, no regret patterns designed to scale as your capabilities in the space grows.
The Path Forward
Agentic AI represents a compelling evolution in operational capability. Success in this domain is built on the established disciplines of good governance and data integrity, adding a powerful ability to act with greater speed and insight.
When approached with deliberate planning, agentic AI can move from a concept to a valuable component of the operational toolkit. The aim is to advance automation through intelligence: enhancing precision, control, and readiness to meet the demands of an increasingly complex world.