Every growing business reaches a point where something isn't getting done, and the decision is: hire someone or automate it. Most people make this call poorly, in both directions.
Some automate things that genuinely need judgment and end up with a system that produces work nobody trusts. Some hire for things that could be automated, then hire again when that person drowns in the same repetitive work.
The questions that actually help
Does this work require judgment? If the task involves deciding between options based on context that varies significantly — a complaint that could go several directions, a situation outside the norm — that's judgment work. Automation can support judgment but can't replace it.
Is it the same or essentially the same every time? Data entry, format conversion, moving information between systems, sending the same type of message on a defined trigger — these are automation candidates. Low variation, clear rules.
How often does it happen? Something that happens twice a month probably isn't worth the build time. Something that happens 50 times a day is a different calculation entirely. The return on an automation scales with frequency.
How stable are the rules? Some processes don't change for years. Others shift quarterly as the business evolves. Automations built on stable processes have long useful lives. Those built on frequently changing rules require ongoing maintenance that can exceed the initial build cost.
The answer that's usually right
Most job descriptions contain both types of work. The customer service rep who handles complex inquiries also processes routine refund requests. The admin who does meaningful project coordination also copies data between spreadsheets.
The practical move is usually to automate the mechanical part of an existing role rather than eliminating the role. This makes the person more effective at the work that actually requires a person.
When a role is mostly mechanical and scales with volume — you hire another person every time the business grows — that's when you're looking at replacing it with automation. That deserves careful evaluation before committing.
