- Ticket classificationA fixed set of queues, priorities or categories, and a history of tickets that people have already sorted. The label set is closed and the right answer is on record, which is what a supervised fine-tune needs.
- Document extractionInvoices, forms, contracts or reports go in; a fixed set of fields comes out as structured data. The test is exact: each field is right or it is not, and the output parses or it does not.
- Intent routingThe first step of an assistant or an agent: work out what the person wants and send the request to the right tool, flow or model. It runs before everything else, so its latency is added to every reply.
What they have in common.
- A closed output. A label from a list, a record with fixed fields, a route from a known set. The answer can be marked right or wrong.
- Volume. The task runs often enough that the price and the delay of each call add up.
- History. There are past examples whose correct output a person or a system of record stands behind.
- Stability. The categories, fields or routes are not being redefined every month.
A task with all four is worth testing. A task missing one of them usually wants a different tool, and the decision guide says which.