What workflows should not be automated first?
The wrong first automation can make AI feel unreliable. The right first build is narrow, frequent, measurable, and safe enough to improve with real use.
Do not start with vague work
If nobody can explain where the workflow starts, who owns it, or what done means, automation will only make the confusion move faster.
Start by mapping the workflow. If the map is still unclear, the first project should be cleanup, not automation.
Do not start with high-stakes autonomy
AI should not make consequential decisions alone on day one. Healthcare, legal, financial, customer-facing, and billing workflows can still be good targets, but the first build should be review-first.
A safer first version drafts, flags, routes, summarizes, validates, or prepares the next action for a person to approve.
Do not automate work that barely happens
A workflow can be annoying and still not be worth automating first. Frequency matters because repeated use reveals edge cases, creates measurable savings, and teaches the team to trust the system.
Save rare, complicated workflows for later unless the cost of failure is unusually high.