A straightforward test for whether a task is worth automating, and the three patterns that reliably pay for themselves.
The honest test is arithmetic. Multiply how often a task happens by how long it takes and by what an hour of that person's time costs. If the annual figure is not several times the build cost, do not automate it. Most AI proposals fail this test quietly, which is why so many pilots never reach production.
Three patterns that reliably pay
First, moving data between systems that do not integrate. If someone retypes information from one screen into another, that is pure loss and it is cheap to fix.
Second, first-line enquiry handling outside business hours. Most enquiries arrive when nobody is at the desk, and response speed is the strongest predictor of whether an enquiry converts.
Third, generating documents from structured inputs, quotes, proposals, reports. Predictable shape, variable content, high frequency.
Where to be careful
Anything where a wrong answer carries real cost, pricing commitments, clinical or legal information, contractual terms, needs a human in the loop. The correct design is an assistant that drafts and a person who approves, not an assistant that decides.