Not every manual process deserves automation. Some are rare enough, or judgment-heavy enough, that the cost of building and maintaining automation exceeds the time it saves. Before scoping a workflow automation project, it's worth scoring candidates instead of automating whatever is most annoying this week.
A two-axis scoring model
Score each candidate process on two axes, 1–5:
- Impact — frequency × time per occurrence × how many people are involved. A process that happens 200 times a week at 10 minutes each outranks one that happens twice a month, even if the second one feels more painful.
- Automatability — how rule-based the decision is. A process with clear, consistent criteria scores high. A process where two experienced people would make different calls on the same case scores low — that's a sign it needs an AI-assisted step with human review, not full automation.
Prioritize high-impact, high-automatability processes first. High-impact, low-automatability processes are usually better candidates for AI-assisted triage — narrowing the decision, not making it — than for full automation.
The maintenance cost people forget to score
Automation isn't a one-time cost. Every automated workflow needs monitoring, and every upstream system change (a new field in the CRM, a changed API response) is a chance for silent failure. Processes tied to systems that change frequently carry a higher ongoing cost than the initial build suggests — factor that into the impact score, not just the build estimate.
A simple prioritization example
| Process | Impact | Automatability | Priority |
|---|---|---|---|
| Support ticket routing | High | High | Automate first |
| Invoice reconciliation | Medium | High | Automate |
| Contract exception review | Medium | Low | AI-assisted triage, human decides |
| Quarterly board reporting | Low | Medium | Defer |
The takeaway
Automation ROI isn't a single number you calculate once. It's a relative ranking that should change as your systems, headcount, and volume change. Revisit the scoring quarterly instead of treating the initial automation roadmap as fixed.