Practical Q&A
Five questions before an AI project
A practical FAQ about scope, source data, evaluation and ownership.
What should the first version do?
Pick one task with an observable finish: find evidence, prepare a draft, reconcile a record or move a reviewed item through a workflow. A narrow task makes the value and failure cases easier to see.
What data does it need?
Start with the sources a person already uses to do that task. Check whether they are accurate, current and available to the intended users. More documents can add contradictions as easily as useful context.
How will we know it helped?
Define an accepted result and a baseline. Include review time, corrections and costs. A polished demonstration is a useful checkpoint; measured use over representative tasks is stronger evidence.
Where does a person remain involved?
Identify who reviews important outputs and who can authorize changes. Give the system a clear way to stop when information or permission is missing.
What should I own after delivery?
Agree on access to your data, documentation, operating responsibilities and a way to export or move the system. Bring those questions to the first conversation.
This entry is an introductory FAQ, not a transcript of questions submitted by readers.