Every operations stack I have inherited had the same shape: six workflows in the CRM, two Zaps someone built in 2023, and a contractor’s scenario in a third tool that nobody dares to open. All of them fire on the same record. Something breaks weekly, and the fix is another workflow on top.

None of those automations were built by careless people. Each one solved a real problem on the day it was built. The trouble is that “does this solve a problem?” is the wrong test. Almost anything passes it.

The test

The test I use now is simpler and much harsher: does this automation remove a decision, or hide one?

An automation that removes a decision takes something the business has already agreed on and makes it happen without a human remembering to. Deal closes, invoice draft is created. Lead comes from a defined territory, it goes to that territory’s owner. No judgment involved, because the judgment was made once, up front, and written down.

An automation that hides a decision takes something the business has not agreed on and quietly picks an answer. Lead scoring thresholds nobody can explain. Stage changes triggered by activity counts. Enrichment that overwrites a field a human filled in deliberately. These feel like efficiency. What they actually do is bury a disagreement inside a workflow, where it can go wrong for months before anyone notices.

How to tell them apart

Ask three questions about any automation before you build it, or keep it:

  • Can I state the rule in one sentence a non-technical colleague would agree with? If the rule needs a diagram, it is a decision, not a rule.
  • Who would notice if it silently stopped working? If the answer is “nobody, for a while,” you have built something that hides its own failure.
  • What did a human do before this existed? If they did nothing, be suspicious. You may have automated a step the business did not actually want.

Fewer, louder automations

The stacks that hold up over time have fewer automations, not more, and every one of them is boring. They log. They route. They create the next record. They fail loudly, in a channel someone reads. Nobody has to explain them because there is nothing to explain.

The ones that move the mess around faster are clever. That is usually how you can tell.