AI Agents in Controlling: Turning Hype into Practical Value

Why many companies are still hesitant and how to get started the right way.

Artificial intelligence (AI) is everywhere, also in controlling. At least as a buzzword. Yet despite all the attention, many organizations are still struggling to put it into practice. Some are only just getting started, while others simply do not know where to begin.

Part of the problem is the terminology. Terms are often used inconsistently and are not always clearly distinguished from one another. At the same time, expectations can quickly become unrealistic, especially when companies start talking about fully automating entire processes with AI agents.

But is relying on AI-driven automation enough? Or does the use of AI agents require a clearer framework within the organization? This was exactly the topic of our 45th iTalk session on March 20. Florian Bliefert, expert in AI and RPA in controlling, joined us to help clarify these questions and discuss both the opportunities and limitations of AI agents.

One key takeaway: chatbots, assistants, and AI agents are often treated as if they were the same thing, even though they work very differently. A practical distinction helps:

• Chatbots are conversation-based and respond to user input
• Assistants support users with specific tasks
• AI agents go a step further: they independently plan and execute tasks on their own

What matters most is not finding the perfect definition, but creating a shared understanding across the organization. Only when everyone is talking about the same thing can meaningful applications be identified.

Getting Started Is the Biggest Challenge

One thing became clear very quickly during the iTalk: in controlling, the biggest obstacle is not the technology itself — it’s getting started. Many teams simply do not know where to begin. At the same time, there is concern about making mistakes or launching overly complex projects that may ultimately fail to deliver results.

That is why a pragmatic approach is intentionally simple: start small. Instead of launching large-scale transformation initiatives, the first step should be identifying a manageable use case.

Reporting is often an ideal starting point. AI can help identify deviations, generate initial commentary, and structure results for presentation. Because these tasks occur regularly and the results are easy to check, reporting is particularly well suited for early AI adoption.

From Chatbot to AI Agent

As organizations gain experience, the question becomes how far automation can go. Traditional applications typically follow predefined, step-by-step processes. AI agents operate differently. They are given a goal and independently determine how to achieve it.

In controlling, this could include:

• Accessing multiple data sources
• Analyzing and interpreting data
• Creating reports
• Automatically distributing results to recipients

So the key difference is not what these tools can do, but how independently they can do it.

Where Automation Reaches Its Limits

As significant as the potential may be, the limitations are just as clear. One major factor is the level of control required. If outputs need to be reviewed so extensively that the time savings disappear, the overall benefit quickly diminishes.

This means the value of AI always depends on the context. For simple, standardized tasks, automation can create immediate value. For use cases involving high risks or significant business impact, human oversight remains essential. AI can support decision-making — but it does not replace responsibility.

Why Processes Matter More Than Technology

One frequently underestimated aspect is the quality of the underlying processes. Automation does not automatically produce better outcomes. If an inefficient or poorly designed process is automated, the result remains inefficient.

This becomes especially clear in reporting. Even technically flawless reports create little value if they fail to meet the actual needs of their recipients.

The conclusion is straightforward: first review the process, clarify requirements, and reduce unnecessary complexity. Only then does automation become truly effective.

Conclusion

The value of AI in controlling does not come from automating as much as possible. It comes from using AI purposefully in the right applications. AI agents in particular demonstrate the potential of automating more complex tasks. At the same time, their successful use requires far more than technology alone.
Without clear processes, meaningful use cases, and a shared understanding across the organization, the impact will remain limited. The goal should therefore not be to automate everything as quickly as possible. The real priority is to start — in a structured, pragmatic way and with a clear focus on tangible business value.