Most automation projects that disappoint were aimed at the wrong task. The tool worked. The choice did not.
The three-part filter
Does it happen often?
Daily, or many times a day, is the zone. Something you do twice a month takes longer to build and maintain than to keep doing by hand, and it will quietly break between uses.
Can you write the rule down?
If you can describe the decision as a sequence of ifs, it can be automated. If your answer includes "it depends, you get a feel for it," that is judgment, and judgment is exactly what automation is bad at.
What does missing it cost?
A lead that sits unanswered for two days is usually a lost lead. A social comment nobody sees is a customer who asked publicly and got nothing. Those are worth automating even at modest volume, because the failure is expensive.
- 01
Happens often?
Daily, or many times a day. Twice a month is not worth the build.
- 02
Can you write the rule?
If the answer needs "you get a feel for it," it is judgment.
- 03
What does missing it cost?
An unanswered lead is usually a lost one. That justifies modest volume.
The tasks that clear the filter almost every time
- Lead routing. A form or DM arrives, gets categorized, lands in the CRM, notifies the right person. Every step is a rule.
- Comment and DM triage. Instagram and Facebook comments sorted into real enquiries against noise, with the enquiries pushed somewhere a human will see them.
- Follow-up gaps. Not marketing blasts. The "you asked for a quote four days ago and nobody replied" gap that costs real money.
- Data entry between systems. Anything you copy from one tab into another on a regular basis.
- Reporting. Pulling the same numbers into the same sheet every Monday.
The tasks that do not
- Anything requiring a judgment call, unless you are routing it to a human rather than deciding it.
- Rarely run processes. Annual, quarterly, occasional. Write a checklist instead.
- Work whose rules change monthly. You will spend more time editing the workflow than running the task.
- Anything touching regulated personal data without a lawyer in the conversation first. Health records especially. See the n8n cost comparison for why compliance comes before tooling.
A useful test: could you hand this to a new assistant with a one-page instruction sheet and expect it done correctly? If yes, automate it. If the sheet would have to say "use your judgment," do not.
Start with one
The common mistake is mapping twelve processes and building all of them. Build one, run it for a month, and watch where it fails. It will fail somewhere, and the failure teaches you more about your own process than the planning did.
Then build the second, knowing what you now know. That compounding is real: the comment and DM automation I run for doctor practices started as one workflow sweeping Facebook comments, and my own outbound system is now ten modules that each began the same way.
Budget honestly
A workflow is not finished when it runs. It needs monitoring, because a silent failure is worse than no automation at all. You stop checking the inbox you automated away, and nobody notices for a fortnight.
See what an n8n build and handoff includes, or start with the automation consultation if the map itself is the problem.
