Data and deployment questions for an AI project

Data and deployment questions for an AI project

Confirm each data flow, external provider and access permission in the actual deployment. Separate storage from request processing: a local database can still send selected content externally. Verify the architecture and configuration behind any “self-hosted” label.

• Storage: Where are source documents, customer records and outputs stored?
• External requests: Which fields are sent to models or services?
• Providers: Which providers and account settings are used?
• Access: Who can read, export, change or delete records?
• Other copies: How are logs, backups and temporary files handled?

Agree action and approval boundaries

Distinguish read-only analysis, drafting and sending messages. Agree which actions are enabled, which require approval, and how uncertain or failed results are handled. A demo does not authorize operation of a real account.

Verify boundaries during acceptance

• Use representative data and inspect a real request path.
• Record every unverified boundary as an open question.
• Include security, retention and compliance requirements in the project scope; do not infer them from marketing terms.