Seven checks before launching an AI project
An ACORN GLOBAL checklist for evaluating models, vendors and project proposals before spending money or sharing sensitive data.
Use these seven questions to make an AI project testable, reversible and understandable. This is a free planning aid, not a certified audit or performance guarantee.
A practical readiness checklist
An observable goal
Describe one task to improve and measure the current baseline. Model activity alone is not a business outcome.
A human decision owner
Name the person accountable for results, exceptions and shutdown. Specify which decisions always require approval.
Lawful, minimal inputs
Check each source’s provenance, usage permissions and sensitivity. Exclude unnecessary personal and confidential data.
Representative cases and failure tests
Collect ordinary, difficult and deliberately misleading examples. Test correct answers and appropriate refusals.
Clear cost and capacity limits
Set volume and spending caps, measure per-task cost and latency, and define behavior when a provider quota is hit.
Evidence and rollback
Keep version identifiers, inspected sources, error logs and observed results. Retain a manual fallback and a safe shutdown path.
Deliverables and acceptance
Specify what will be handed over, how it will be tested and who can accept it. Payment alone never verifies delivery.