How to Decide What to Automate First
A scoring method that takes about an hour and consistently beats intuition, because intuition picks the most annoying process rather than the most automatable one.
Score every candidate on four axes
| Axis | Score 1 | Score 5 |
|---|---|---|
| Frequency | Monthly or less | Many times daily |
| Rule clarity | Two good people would handle it differently | Written down, no disagreement |
| Error tolerance | A mistake is expensive or irreversible | Mistakes are cheap and caught easily |
| Data availability | Lives in someone's head or on paper | Already in a system with an API |
Multiply the four. Anything scoring above 200 is a strong first candidate. Below 60, leave it alone for now regardless of how much it annoys you.
Why the most painful process is the wrong first project
The thing everyone complains about is usually complex, political and badly documented — which is precisely why it is painful. Starting there means a long build, a contested definition of success, and a high chance of a visible failure that poisons the next three attempts.
Start somewhere boring and high-volume. Win. Then spend the credibility on the hard one.
The question to ask before any of this
What happens if we just stop doing this?
A surprising number of recurring processes exist because of a decision nobody remembers making, for a person who has left, to satisfy a requirement that has changed. Automating one of these makes a pointless process permanent. Deleting it is free and instant.
Run this scoring across every repetitive process you can name before picking. The exercise takes an hour and routinely changes the answer — the highest-scoring candidate is very rarely the one people nominate first.