Essay
Automating chaos only scales chaos
A workflow without an owner, a metric and a stop rule becomes a faster way to spread a bad decision.
Automation is attractive because it looks like relief. A team that copies data between an ERP, a CRM and a spreadsheet will ask for a bot before it asks why three systems hold the same fact. The bot can be built. The disagreement about which fact is true remains, and now it runs on a schedule.
The same pattern shows up with generative AI. A model can draft the SOP, classify the ticket or summarize the meeting. If the underlying process has no owner, the draft becomes another artifact nobody maintains. The company feels modern and stays slow.
A test before the build
Before an automation or an AI assist is funded, three answers should already exist. Who changes the rule when the business changes. What number moves if this works, and in what window. What would make the company turn it off.
If those answers are missing, the work is not “an integration.” It is a decision the company has postponed. Building first locks the postponement into software. Unwinding it later costs more than the original manual pain.
Useful automation removes a repeated, already-understood step. It does not invent the operating logic the step was hiding. Start with the step the team can explain without opening a tool. Then decide whether a machine should do it.