17. What should AI-assisted charter procurement automate—and what must remain a human decision?

Put this to: Jeff Agur and Andrew Day, paired with the buyer’s information-security lead.

Why it matters now: NBAA’s April 2026 coverage describes a tool designed to cross-check FAA and DOT information for charter validation. That is evidence of a proposed use case, not independent proof of operational effectiveness. Separately, NIST’s generative-AI risk profile identifies confabulation as a risk to manage. (nbaa.org)

Question: “Where can automation accelerate this procurement process without turning incomplete documents into false certainty?”

A good answer should define a controlled pilot:

1. Extract quote and contract fields with page-level provenance.
2. Preserve ‘unspecified’, conflicting and unreadable fields rather than guessing.
3. Calculate totals in tested software, with visible tax and inclusion assumptions.
4. Require human confirmation of critical dates, entities, obligations and exceptions.
5. Keep operational acceptance with the operator and specialist legal or tax conclusions with the relevant advisers.
6. Minimise itinerary data, restrict access and define retention and deletion.
7. Treat uploaded documents as evidence, never as instructions authorising payments or changing approval rules.

Follow-up: “Show a case where the system must stop and request clarification. Who can override that stop, and what is recorded?”

Evidence to request: an evaluation set, error categories, correction workflow and audit trail. Test it against permissioned real documents before using its output for live approvals.

Expert quote: [Pending interview, attribution and approval.]