Automation That Is Worth Setting Up, and Automation That Isn't

Automation projects fail for the same reason software projects fail: they are built for the process someone imagines rather than the one that happens. The way to avoid it is to write the process down first, in the order it actually occurs, including the awkward exceptions.

The payback test

Estimate minutes saved per week, multiply by 50, and compare with hours to build plus an ongoing maintenance allowance of roughly two hours a year per automation. If payback is longer than six months, leave it manual and revisit when volume grows.

Build order that survives contact with reality

1. Write the current process as numbered steps, including who does each one.
2. Delete steps that exist only out of habit. Removing work beats automating it.
3. Standardise the inputs — a single enquiry form beats three inboxes.
4. Automate one step, run it for a fortnight, and check the exceptions it created.
5. Document what it does and how to switch it off. An undocumented automation becomes a mystery within a year.

The maintenance cost people forget

Every automation is a small piece of software with dependencies that change. When a form field is renamed or a plan changes, something silently stops. Keep a one-page register: what it does, what it connects, who owns it, and how you would notice it had failed. That last column is the one that matters.