AI Internal Tool Production Checklist

Start with the operational decision

Document the user, repeated task, source record, exception, and next action. A prompt that names charts but not decisions will produce a dashboard-shaped demo rather than a useful tool.

• • Named user and cadence
• • Decision or exception
• • Required context
• • Allowed action
• • Measurable 30-day outcome

Treat data and queries as a contract

Record the system of truth, table or view grain, joins, filters, time zone, exclusions, freshness, and owner for every important metric. Validate against a trusted benchmark before optimizing layout.

Review every write path

Prefer narrow business actions to unrestricted editing. Test required fields, invalid transitions, duplicate submissions, confirmation steps, permissions, and recovery for each mutation.

• • Role permission
• • Input validation
• • Confirmation
• • Audit context
• • Recovery or reversal

Test production-shaped records

Use long text, missing relationships, empty queues, large tables, unusual dates, partial failures, and users with different roles. Confirm that an operator can understand and fix every failure state.

Assign ownership after launch

Name owners for source schema changes, metric definitions, template and query updates, access requests, and operational exceptions. Monitor adoption, resolution time, failed actions, and repeated manual work.