Not every manual process is worth automating
It's tempting to treat automation as a universal good — if a task is manual, automate it. In practice, some manual processes are manual for good reasons: they're low volume, they require judgment that's genuinely hard to systematize, or the cost of getting them wrong is high enough that a human check is worth the time it takes.
The pattern worth automating
The workflows that pay off are usually high-frequency, well-defined, and low-judgment — the kind of task where the tenth time looks almost exactly like the first. Approvals that follow a consistent rule set, data entry between systems that don't talk to each other, and status updates that get manually copied from one place to another are common examples.
Where automation alone isn't enough
The workflows that look automatable but aren't usually involve judgment calls disguised as rules — an approval that's "usually straightforward except when it isn't," for instance. Automating only the straightforward path and routing everything else to a person, rather than trying to automate the exception handling too, tends to work better than an all-or-nothing approach.
How to prioritize
A useful starting question is how often a task happens and how much judgment it actually requires — not how much judgment it appears to require on paper. High-frequency, low-judgment work is the clearest starting point. Everything else is worth a closer look before committing engineering time to it.