Every automation project starts with a list. Someone opens a document, writes down everything the team does by hand, and the list runs to forty items. Then nothing happens for three weeks, because forty is not a starting point.
The trick isn’t picking the biggest item. It’s picking the one that will still be running in six months.
Volume beats complexity
A decision made two hundred times a week is worth automating even if each instance saves ninety seconds. A decision made twice a month isn’t, however painful each one feels.
This is counterintuitive because the painful cases are the ones people remember. Ask a team what wastes their time and they’ll describe the worst hour of their month, not the fifteen minutes they lose every single morning without noticing.
Go by the logs, not the anecdotes.
Judgement is the disqualifier
The second filter matters more. For each candidate, ask how much judgement it takes.
- If two people would reliably reach the same answer, it’s a rule
- If they’d usually agree but sometimes not, it needs a confidence threshold
- If they’d routinely disagree, it isn’t automation, it’s a decision that needs an owner
Only the first category works cleanly on day one. The second works with a human fallback. The third will burn your credibility if you try.
Reversibility is the safety net
Prefer workflows where a mistake is cheap to undo. Routing a request to the wrong team costs a few minutes. Sending the wrong invoice costs considerably more.
Start where being wrong is survivable, and you’ll learn far faster than a team that started somewhere it couldn’t afford to be wrong.
What this looks like in practice
The first workflow most teams should automate is dull: classify an incoming thing, attach the context someone will need, put it in the right place. High volume, low judgement, trivially reversible.
It won’t demo well. It’ll save more hours in its first month than the impressive workflow saves in its first year — mostly because the impressive workflow won’t be finished yet.
When to move on
Once the first workflow has run for a month without anyone babysitting it, take the next one. Not before.
The teams that stall are the ones that started three workflows in parallel to show momentum, then discovered they had three half-instrumented systems and no idea which one was misbehaving.
