Before you hire an AI vendor, make them answer these questions

Ask who owns the data model, how prompts and retrieval are versioned, what happens when the model changes behavior, how quality is measured, where sensitive data goes, and how the system fails safely. If the answers are hand-wavy, keep your wallet closed.

Also ask what they refuse to do. Good vendors have boundaries. Bad vendors say yes to everything, then discover production constraints with your budget. That is not innovation. That is tuition.

A paid consultation with me is useful before vendor selection because I am not trying to win the build. I no longer take on implementation or building projects. I only offer paid consultations and strategic advisory, so the advice is not a disguised sales pitch for delivery work.