What is AI-based Process Automation?

The Difference from Classical Automation

Classical automation follows fixed rules: If subject X, then folder Y. This works as long as cases are predictable.

AI comes into play where input is unstructured – a freely formulated email, a scanned delivery note, a protocol. The model provides an assessment, not certainty.

What This Means for Implementation

A reliable process typically looks like this:

1. Rule-based software initiates the process.
2. The model suggests a categorization – specifying how certain it is.
3. If certainty is high, the process proceeds; if low, it is sent for review to a person.
4. Every decision is logged.

The third point is the most important. A system that makes decisions without hesitation even in uncertain situations produces errors that go unnoticed.

Typical Meaningful Applications

• Pre-sort incoming emails and assign them to the correct process
• Extract invoices and delivery notes
• Summarize long documents with references to sources
• Prepare draft responses for human approval

Where Caution is Required

Whenever a wrong assessment could cost money, legal compliance, or trust: Creditworthiness, terminations, medical questions, legal inquiries. Here, a model may assist, but must not make final decisions.