Agentic Automation: Mission Automation in regulated industries

Agent systems offer a wealth of possibilities and promise for modern automation. But is this the right tool for every job, and how can this innovative and powerful technology be used to create economic value?

The new trend: agent systems

Independent, autonomous, and context-sensitive work with the help of technology. What initially sounds like a pipe dream has become a tangible reality since last year. Agent systems are the keyword and have not only impressed experts, but also presented companies that want to use this technology with new challenges. While companies in less regulated sectors are tapping into new business areas and potential, companies in highly regulated sectors with a multitude of requirements are wondering how agent systems can be used in a transparent, structured, and profitable way. Yet it is precisely these companies, such as insurance companies, public authorities, and banks, that can benefit from the autonomous capacity to automate complex and often analog processes in an intelligent and uncomplicated manner. To understand how agentic automation is revolutionizing the world of automation, it is necessary to understand the journey this technology has already taken and where it originated.

Robots that learn to think

When you think of automation, you think of robots. It was no different in the digital automation landscape. Robotic process automation describes the deterministic processing of digital processes using software robots. The key feature of RPA is that it simulates human input. RPA is quick and easy to implement, enabling companies to rapidly leverage efficiency potential in software-based processes. The penetration of RPA increased despite its limitations: The deterministic approach that RPA follows requires data and processes to be structured and transparent in order to be automated. Automating documents (e.g., policies) whose content first had to be interpreted was difficult to impossible and reduced the potential for automation.

In order to increase the potential for automation, an approach has emerged that uses artificial intelligence to try to close this gap: AI-powered automation (AiPA for short). Here, deterministic RPA processes continue to be enhanced with artificial intelligence at specific points in order to make processing more “intelligent.” Processing does indeed become more intelligent, and potential that was previously unattainable is now within reach. So what is the problem? AI models must be specifically trained for the purposes for which they are to be used and secured with quality gates.

When we talk about agent systems or agentic automation, we are referring to autonomous, independent, and adaptive systems. This is achieved through a combination of an AI model and so-called “tools” that enable an agent to perform actions independently. So instead of being trained for “only” one purpose, an agent can interpret data, documents, and requests, break them down into individual tasks, and independently extract, classify, and process them using assigned tools. Such agents are controlled via prompts that provide work instructions and explain how an agent should understand an input and what to do with it. From there, the system takes over the interpretation. If more data is needed to make a decision, available tools are searched, executed, and the results are processed independently. The advantages are obvious: flexible content and even complex structures can be interpreted and processed without the need to describe them in detail. How should insurance companies, banks, or public authorities use this technology if they cannot assess whether the agents are working correctly?

The intelligent revolution in process automation

At adesso, we have addressed this problem and use AI agents as part of a larger automation structure. Our Agentic Automation uses agents in an isolated, targeted, and transparent manner by equipping them with tools and instructions and integrating them into a process flow that is controlled by a process workflow. Non-deterministic approaches meet deterministic process design, because in highly regulated areas in particular, processes are subject to regulations, procedures, and laws that are rarely flexible. The data that comes in is completely different. It is often complex, unstructured, and varied. We follow the philosophy of “flexible input, structured output.” With the help of a workflow-based process structure, we define exactly what results we expect in the process and align the agents accordingly. This means that we take advantage of flexible processing, but pour it into predefined structures, thereby creating transparency and preventing erroneous results. Our agents are allowed to work flexibly, but they must arrive at one of the results we expect in our process. The path becomes flexible, but the goal remains the same.

RPA, agentic automation, and workflow control provide the components needed to establish a future-proof, sustainable, and intelligent process platform for the long term. The graphic above depicts such a business automation platform, which enables complete, transparent processing even for highly regulated companies.