Editor's Note:David Hickey is the leader of digital solutions at Baker Tilly, a consulting, tax, and assurance firm headquartered in Chicago, Illinois. This article reflects the author's personal views only.

Financial institutions are at the forefront of leveraging AI agents to achieve more streamlined, efficient, and automated operations. Whether in banking, credit unions, capital markets, insurance, or asset management, the application of agentic AI across different financial sectors shares similarities.

Agentic AI represents a sophisticated form of autonomous AI that simulates human cognitive processes. These AI "agents" are designed to make independent decisions, collaborate, and adapt to their environments. By ensuring agents focus on specific tasks and receive targeted training, they can operate reliably and safely in autonomous mode. As agents continuously learn and refine their capabilities, they can significantly enhance efficiency, save time, and streamline complex workflows that traditionally required human intervention.

Building Agent Personas

In financial services, building agent personas is crucial to unlocking the full potential of agentic AI. An agent persona is a digital representation of a role within an organization, designed to perform specific tasks and interact with systems and humans in ways that mimic real-world interactions. Think of it as a combination of a job description and standard operating procedures for a task executor.

These personas begin with well-defined system prompts, including identifying the specific tasks and responsibilities associated with the role, as well as the skills and knowledge required to execute them effectively. Like human employees, agent personas need personality traits that align with the organization's brand and values. This is especially important for agents that interact with people outside the organization, including determining the tone and style of communication, as well as the level of formality and empathy the agent should exhibit.

Once the persona is defined, it must be supplemented with additional context for executing its role. Context grounding is essential for various applications in intelligent automation. For example, in the financial industry, it sets the parameters for AI agents' access to databases, standard operating procedures, regulatory information, and other trusted sources. This ensures agents always use the most current information and understand how to leverage it in task execution.

By grounding AI models with specific business context, organizations can ensure their AI systems are not only accurate but also aligned with their unique operational needs and regulatory requirements.

After defining the persona, personality, and providing context on data sources, the next step is to define the specific boundaries within which the agent is authorized to operate. This involves providing the agent with relevant use cases or scenarios to observe its responses. Unexpected responses are used to refine the system prompts to guide correct behavior.

The latest prompt-building tools evaluate the robustness of use cases against the defined persona and recommend additional use cases to strengthen the prompts. External black-hat agents can also be used to test the boundaries provided to the agent and help refine system prompts.

As agents execute tasks, continuous training and development are essential to ensure they stay current and handle emerging tasks. This includes tracking key performance indicators and gathering user feedback to identify areas for improvement. Regular evaluations help ensure agents are achieving their goals within the overall context of their role.

Orchestrating Complex Workflows

In the dynamic world of financial services, the ability to model and manage complex, long-running workflows is critical. Recent technological advances have made this possible, allowing organizations to seamlessly integrate agents, robots, and humans into end-to-end processes.

However, managing long-running workflows involving multiple subprocesses and systems presents several challenges. One of the primary challenges is ensuring each step of the process is executed accurately and efficiently. This requires meticulous coordination and monitoring at the transaction level to prevent errors and delays. Additionally, the complexity of integrating various systems and tools can lead to compatibility issues and data silos, hindering the smooth flow of information.

Providing an orchestration layer enables the seamless integration of various agents, robots, and human tasks into a unified workflow. This orchestration ensures the components of the process work together, reducing bottlenecks and enhancing overall efficiency. By coordinating and monitoring each step, the orchestration layer helps prevent errors and delays, ensuring the entire workflow runs smoothly and effectively.

For example, in a loan application, an agent can initially review documents, automatically request missing information or clarifications from the applicant, and then initiate subsequent steps such as credit checks, income and employment verification, and predicting the likelihood of loan default. The completed application package is then sent to an underwriter for review, and once approved, the agent immediately generates a loan offer and sends it to the applicant.

This transaction-level concierge service advances the transaction as soon as process steps are completed, ensuring no unnecessary delays and timely execution of each step. Agents only request human support when needed, allowing human employees to focus on more complex and strategic tasks or dedicate more time to direct client interaction. This approach not only enhances workflow efficiency but also ensures the overall process is smooth and responsive to customer needs.

One of the greatest advantages of central orchestration is its ability to preserve overall context throughout the process. As transactions seamlessly transfer between agents, robots, and people, central orchestration ensures context is maintained throughout. This means each component of the workflow is aware of previous steps and the overall goal, supporting more informed decision-making. This continuity is crucial for maintaining the integrity and efficiency of the process.

This visually orchestrated end-to-end process can also provide a level of visibility and accountability for audit and compliance departments. By having a comprehensive view of the entire workflow, organizations can ensure every step of the process is transparent and traceable. This transparency not only improves accountability but also allows for better decision-making and process optimization.

Furthermore, the ability to visualize the entire process allows for continuous improvement and optimization. Organizations can analyze data collected from these workflows to identify bottlenecks, inefficiencies, and areas for improvement. By making data-driven decisions, organizations can enhance processes, reduce costs, and improve overall performance.

The Road Ahead

As AI technology continues to advance at an exponential rate, the impact of agentic AI will become even more profound. These agents will become increasingly sophisticated, capable of handling complex tasks with greater precision and making nuanced decisions.

Despite the current fragmentation of systems and applications in the AI landscape, the future holds immense promise. The financial services industry stands at the forefront of a transformative era driven by agentic AI.