The fastest way to build an expensive automation is to encode an inefficient manual process exactly as it is performed today, including the workarounds nobody remembers inventing.
Map before you build
- Observe
- Map steps
- Mark handoffs
- Find waits
- Cut, then automate
Handoffs and waits are where the time goes. A step that takes two minutes but waits a day for approval is not a two-minute step.
Three questions per step
- Does this step exist because of a real constraint, or because of a past tool?
- What happens if it is skipped? If the answer is nothing, delete it.
- Does it need judgement? If yes, keep a human on it and automate everything around it.
Deleting a step is a bigger win than automating it, and it costs nothing to run.