Start with a question worth paying to answer

It is tempting to collect everything available and look for a buyer later. A better first move is to define a question a buyer needs answered. That does not prove someone will pay, but it gives the work a proposition to test before broad collection begins.

For automotive AI, the question can be specific: can a system tell when an appointment request is ready to act on, distinguish current availability from an outdated listing, or recognize when a service question needs a person? Each question implies a different deliverable.

Then ask whether a source can support the task and whether the proposed use can be permitted. If either answer is no, collecting more material is unlikely to resolve the problem.

A well-scoped pilot names the buyer task, source boundary, intended use, and acceptance criteria. The result may be a larger project, a narrower one, or a decision to stop. The useful outcome is evidence for the next decision.