Why AI cannot automate bad processes

Many businesses begin an AI project with the wrong question: “Which tool can we use?” They should ask: “How does this process actually work today?”

Most processes look more structured on paper than in everyday practice.

The documented process rarely matches reality

In reality, there are:

• manual intermediate steps,
• exceptions known only to certain employees,
• data gaps between ERP, CRM and document systems,
• approvals by email, Excel or word of mouth, and
• different approaches to the same case.

Adding AI to this creates faster access to an unclear process, not genuine automation.

AI then takes on both the desired work steps and:

• unnecessary loops,
• contradictory rules,
• missing responsibilities,
• poor data handovers, and
• long-established workarounds.

The core problem is rarely the model

Businesses know their intended process but not how it actually runs.

Before integrating AI, make the following visible:

• Where does the process start and end?
• Which systems and data sources are involved?
• Where do media breaks occur?
• Which decisions follow clear rules?
• Which exceptions still need human judgement?

Only then can you decide what to automate, integrate or change organisationally.

AI can accelerate a process, but cannot decide whether it is well designed.

Processes before tools

• Processes before tools.
• Integration before automation.
• Implementation before theory.

AI creates value where it is technically well integrated into a understood business process, rather than simply introduced quickly.