When do I automate a process and when do I deliberately not?

Automation is not the same as improvement. Technology usually makes an unclear process unclear at greater speed. I therefore start not with the tool, but with which step is predictable, where exceptions arise and which risk makes human judgement essential.

Process first, technology second

I reduce the current way of working to concrete steps: trigger, required information, decision, action, handover and control. Only then do I assess whether AI or automation genuinely adds value.

My decision framework

1. Volume and repetition. Does the task recur often enough to make standardisation worthwhile?
2. Regularity. Are the input and desired outcome recognisable, or is the process dominated by exceptions?
3. Risk. What happens if the outcome is incorrect, incomplete or reaches the wrong person?
4. Confidentiality. May the chosen environment process the information, and who has access?
5. Human judgement. Which step requires knowledge of relationships, context, tone or executive sensitivity?

Three possible design choices

Fully manual suits rare, sensitive or highly variable situations. AI-supported suits work where collecting or structuring takes time but review remains essential. Partly automated suits predictable steps with clear exception routes and ownership.

Human in the loop is a design choice

A review point should not be bolted on afterwards. I define in advance who checks, what they check, which source they use and what happens when there is doubt. External communication, decisions, personal data and confidential documents always retain an approval step.

Measuring value beyond speed

Time saved matters, but it is not enough. I also look for less repair work, clearer ownership, faster follow-through, consistent quality and a method colleagues can actually use.

Professional decision framework based on my working method, updated 29 October 2026.