A common conversation starts with “we should be using AI for this”. Sometimes that is right. More often the task turns out to be a repetitive, rule-based job that a simple automation handles faster, more predictably and at a fraction of the cost.
Three questions to ask first
Is the task the same every time? If the steps can be written down as a short list, a plain automation will follow them exactly, every time. A model adds variability you do not need.
What happens when it goes wrong? If a mistake is easy for a person to spot and fix, you have more freedom to experiment. If a mistake goes straight to a customer, you need a human check regardless of the technology.
Is the input genuinely messy? Free text, scanned documents and conversations are where AI earns its place. Clean data in a spreadsheet does not need it.
A worked scenario
Imagine a small studio that receives class enquiries by email. Each one needs to be added to a sheet, answered with the timetable, and followed up if there is no reply. That is three fixed steps. An automation that reads the form, writes the row, sends the timetable and schedules a follow-up solves the problem completely, with nothing to “train” and nothing that can answer unpredictably.
If the same studio wanted to answer open-ended questions about which class suits a beginner with a knee injury, that is a different problem. There, a well-scoped assistant with a person reviewing replies might help. The point is to decide based on the task, not the technology.
Our rule of thumb
Start with the simplest thing that removes the repetitive work. Add AI only where the input is messy and a person stays in the loop. This is general guidance; the right choice depends on your specific process.