A fixed automation follows the same steps every time and is right for most business processes: predictable, testable, cheap. It is what tools like n8n, Make and Zapier do, and what custom code does when those tools run out.
AI enters when a step needs judgement, such as reading a document or classifying a message. Even then the automation around the model, with its validation and approval points, does most of the work. Our automation ROI calculator puts a number on what a process costs to leave manual.
A good first automation is one that is reversible, high volume and boring, because it proves the approach without putting anything at risk.