Most first automation projects fail by being too big. "Automate customer communication" has no edges. Nobody can test it, and nobody can say when it is done.
A better first target fits in one sentence. One inbox. One kind of request. One tracker where the result lands.
What that looks like
Here is a hypothetical example, not client work. A small facilities team has a shared inbox. Among everything else, it receives equipment repair requests from staff. Someone reads each one, retypes the details into a spreadsheet, and replies to confirm.
The bounded version: the automation reads only that inbox, handles only repair requests, and writes only a draft row in that one spreadsheet. A person looks at the row and approves it. That is the whole scope.
Why the small size helps
- You can list every input the workflow will see.
- You can list every place it can write.
- A person can check a result in seconds.
- When something goes wrong, there is one place to look.
A wide scope hides these answers. A narrow one gives them to you for free.
Checks worth building in
Even a small internal workflow needs a few checks. Requests still contain names, phone numbers and building details, so privacy and access matter here too.
First, decide what the automation may read. If it connects through MCP, review which tools it is granted. A read-only lookup should not carry write access to anything else. If it does not use MCP, the same question applies to whatever credential it holds.
Second, decide what it may not do. In this example, it never sends a reply. It only drafts one.
Third, handle the messy cases. If a message is not clearly a repair request, it stays in the inbox untouched. If a field is missing, the row says so instead of guessing.
The handoff
The handoff should be understandable by the person who inherits it. One page is enough: what the workflow reads, what it writes, which checks run, and who to tell when it misbehaves.
If the owner cannot explain it in a few minutes, it is too big or too vague.
Growing from here
Once the first workflow runs cleanly and people trust the checks, a second request type or a second tracker is a small step. You earn the wider scope with evidence from the narrow one.
This is the kind of workflow we review: small, bounded, and easy to hand off.
I'm Jonathan Melton. I run FusionAL, focused on reviewing and integrating lower-risk internal AI workflows, one bounded workflow at a time.
Top comments (0)