The workflows that actually pay back
Not every repetitive task is worth automating. Here's the short heuristic we use to decide what to ship first.
A common mistake on automation projects is to start with the workflow that feels most painful. The painful one is usually painful because it's complicated, edge-case heavy, or politically sensitive — exactly the wrong place to start.
Here's the rough heuristic we apply in the first week of any engagement.
The four signals
Score each candidate workflow on:
- Volume. How many times does this run, per week, across the team? Under 20: probably not worth it. Over 200: very likely worth it.
- Time per run. Below 2 minutes is rarely worth automating — the coordination cost wins. 5–30 minutes is the sweet spot. Over an hour usually means the workflow is actually several workflows in a trench coat; split it before you score it.
- Determinism. What share of the steps are mechanical (click, copy, paste) versus genuinely judgment-driven (read this email and decide what to do)? You want at least 70% mechanical for the first project.
- Blast radius if it goes wrong. Sending a wrong invoice is recoverable. Pushing a wrong stock update to a marketplace is not. Start where the worst case is "we run it again."
A simple scoring rule
We multiply weekly volume × minutes per run × determinism share, divide by blast-radius weight (1 = harmless, 5 = scary), and rank. The top 1–3 workflows usually pay back the entire engagement in the first month.
What to avoid on project one
- Workflows that touch payments or anything irreversible.
- Workflows where the ops team has strong reasons to keep doing it manually (compliance attestation, customer-facing judgement calls).
- Workflows that depend on a system about to be replaced.
Start boring. Ship something visible in two weeks. The interesting stuff gets easier once the team has seen what "good" looks like.