AI launch approval checklist for company leaders

Use evidence to decide whether an artificial intelligence (AI) pilot is ready for routine use.

The National Institute of Standards and Technology (NIST) includes an AI definition covering predictions, recommendations, and decisions. See the NIST AI glossary. Before routine use, check the business outcome, permissions, quality, human review, cost, and operating owner.

Use the six checks below for your launch review. NIST offers voluntary guidance for managing AI risks. Its AI Risk Management Framework (AI RMF) covers design, development, use, and evaluation. Check the NIST AI Risk Management Framework page for its latest revision status before using this guidance.

Check	Decision	Required evidence	Proposed accountable role
Business outcome	Does the pilot address a defined business problem?	Current results, target outcome, measurement method, and review date	Business sponsor
Data permissions	May the system use each data source for this purpose?	Data inventory, access approvals, applicable terms, and retention rules	Data owner
Quality evidence	Does the system meet agreed criteria on representative tasks?	Test cases, results, known failure patterns, and agreed minimum results	Product or technical lead
Human review	Can a person review, correct, or stop a consequential action?	Named reviewer, review instructions, and tested stop or handover steps	Process owner
Spending limits	Can the organization detect and contain unexpected costs?	Approved budget, monitoring method, alerts, and a stop threshold	Finance or service owner
Operating owner	Who maintains the system and responds to problems?	Named role, support process, monitoring schedule, and change controls	Operations owner

Record the current result before the pilot. Use it as the comparison point for later tests. Define the target outcome and measurement method before reviewing results. Agree on minimum results before testing. Use those agreed results as the minimum for launch.

NIST’s AI RMF Core calls for clear, documented roles for managing AI risks. Give a named reviewer authority to check, correct, or stop the system. Test the review process before launch. Specify who receives an issue and what that person can do. Confirm that the process works under realistic conditions.

Record missing evidence as an open decision. Record each evidence item, its owner, and any unresolved limitation. Set a review date for open items. For provider engagements, record deliverables, ownership, and handoff terms. Chiri’s solutions page says customers own what Chiri builds.

Stop and fix

Pause launch when data permissions remain unclear, test results fall below the agreed minimum, or no accountable owner exists. Also pause when people cannot complete human review or the team cannot monitor spending against an approved limit.

Assign each gap to a proposed accountable role. Record the corrective action and the evidence needed to close it. Review the decision after that evidence is available. Do not substitute a schedule or demonstration for evidence. If a gap cannot be resolved, narrow the use or keep the pilot limited.

Questions leaders often ask

What evidence supports the launch decision?

The evidence should show agreed results on realistic tasks. Record data permissions, spending limits, operating ownership, and the tested human review process. Keep unresolved items visible to the decision owner.

What if data permissions are unclear?

Pause that use until the data owner confirms permission and applicable limits. Record the decision and supporting evidence. Do not assume access means permission for every purpose.

Who should make the final launch decision?

The accountable executive should decide using evidence from the business sponsor and operating owner. Relevant data, finance, product, and process owners should address their assigned checks. Record the decision, unresolved limitations, and any conditions for launch.