Automating Poor Processes Doesn’t Make Them Better

Companies are currently investing a lot of time and money in Agentic AI. New agents are expected to answer inquiries, prepare decisions, review documents, or handle multiple process steps. The discussion usually revolves around models, platforms, and use cases.

Yet the real problem often lies elsewhere entirely.

Many companies know surprisingly little about how their processes actually work.

On PowerPoint slides, processes often look clearly structured. In practice, however, there are special cases, manual corrections, Excel files, email loops, and workarounds that have evolved over the years. No one planned them. Nevertheless, they have become part of the process.

And that’s exactly where the risk begins.

A common misconception is that if we automate a process, we solve its problems.

Often, the opposite happens.

Suppose an approval process takes five days. The reason for this could be poor coordination between different departments, missing information, or unclear responsibilities or delegation rules. If this process is automated, these problems do not disappear. At best, they simply move through the system faster.

If you automate inefficiency, you get automated inefficiency.

The technology may work perfectly well. Nevertheless, there is no business benefit.

Before discussing agents or automation platforms, an organization should be able to answer a much simpler question:

How does the process actually work today?

Not in the documentation or in the target vision.

But in reality:

• Who makes decisions?

• What data is used?

• Where do delays occur?

• What exceptions occur regularly?

If you can’t answer these questions, you shouldn’t start automating just yet.

Many companies rely on workshops and assumptions. This quickly leads to discussions about how a process is actually supposed to work.

However, the more interesting question is how it actually works.

That’s why Operational Transformation is gaining importance. The goal isn’t to create more documentation. The goal is transparency. Companies need to recognize how work flows through the organization, where friction occurs, and which decisions have the greatest impact on costs, quality, or processing times.

This doesn’t have to be a tedious, manual process of gathering detailed information; today, there are suitable tools that automatically support process mining and discovery and provide deep insights into the reality of how processes actually play out.

Only once this picture is in place can meaningful automation decisions be made.

An AI agent does not operate in a vacuum.

For an agent to perform tasks reliably, it needs organizational knowledge, business rules, and process context. Without this context, errors occur. Decisions become inconsistent, and employees must constantly intervene.

That’s why successful agentic AI initiatives often depend less on the quality of the model than on the quality of the underlying processes.

Clear process documentation, well-defined responsibilities, and specified data sources are not trivial matters. They form the foundation for every agent’s work.

Many companies want to automate entire value chains right away. That sounds ambitious, but it often leads to high complexity while experience is still limited.

The better approach is usually much less spectacular.

One process.

A clearly defined use case.

A measurable result.

This allows you to gain experience, limit risks, and gain real insights. Automation can then be expanded step by step.