Ten Things You Should Not Automate Yet

We make money building automation, which is exactly why this list exists. A vendor who cannot tell you when *not* to buy is a vendor optimizing for the invoice. Ten holds, with reasons:

1. A process you haven't done manually at least twenty times. Automation freezes a process. Freeze one you barely understand and you have preserved your own ignorance at machine speed.

2. A process that changes monthly. If the workflow is still evolving, the maintenance on the automation will cost more than the labor it replaced. Stabilize first.

3. Apologies. When your business drops something important, a human writes the note. Customers can smell templated contrition, and they price it in.

4. The exception path. Automate the eighty percent happy path and route the rest loudly to a person. Systems that try to handle every edge case handle none of them well.

5. Anything where you cannot define "correct." If two smart employees would disagree on the right output, the machine's version of the disagreement will just be less visible. Resolve the policy first; automate the policy second.

6. Firing, discipline, and bad news. Structure the information with AI if you like. Deliver it with a person, every time, forever.

7. Your one irreplaceable relationship. Every business has an account where the owner personally answers. Keep it that way. The margin on that account is the relationship itself.

8. A process whose data source you don't control. Scraping a portal that changes without notice, parsing an upstream report a partner formats by hand — you are building on land you do not own. Get a stable interface or wait.

9. Quality control on the automation itself. The system that checks the system should not share the system's blind spots. Sampling by humans stays, permanently, at whatever rate the risk justifies.

10. The thing you're avoiding deciding. The most popular automation request we decline: "route these somewhere" where the real problem is nobody has decided who owns them. Automation executes decisions. It cannot make the one you're dodging.

The best automation portfolio has a visible list of things it deliberately excludes. If everything qualified, nobody was thinking.