The AI conversation at most smaller Caribbean operators has moved from whether to how. It has not yet moved to what.
Eighteen months ago, the executives I worked with were asking whether AI was a real thing or a passing wave. That question has settled. The newer question is which model, which vendor, which cloud, which budget line. Reasonable questions, but the wrong first questions — asking them first is the single most common reason an AI initiative produces a presentation deck rather than a working capability.
The right first question is what to point the AI at. Not which model. Not which vendor. What data, which documents, which operational context the AI is meant to operate on. The value of an AI implementation is almost entirely determined by the foundation underneath it — and most operators have not yet inventoried that foundation, let alone made it AI-ready.
Two specific signs an operator is stuck on the wrong question.
The first: the conversation is happening between IT and a vendor, with operations absent. If the people who use the system day to day are not in the room, the AI being scoped will not survive contact with the actual operation.
The second: nobody can answer "what does the AI need to read?" If the answer is hand-waving — "our documents," "our data," "our procedures" — the project will stall at the first integration. Sharper operators can name specific document types, specific systems, the gaps between them.
While the strategic AI conversation moves slowly, the tactical exposure has often already begun. In a recent engagement, the risk manager at one regional Caribbean airport had spent months pushing leadership for in-house AI. His concern was not technical. Management was already using personal $20-per-month accounts to draft documents and analyse operational issues. On those plans, the user is the product — inputs train the model. Someone asking the right questions from outside the airport could surface a workable profile of operational performance, financial pressure, and internal disputes, assembled from inputs no one realised were leaving the building. The mitigation he proposed was modest: move onto team or enterprise plans where data is contractually excluded from training, with the secondary benefit that staff conversations become organisationally visible at appropriate access levels. Finance opted for the cheapest plans. The risk manager was right.
The series that follows works through everything underneath that shift: the economics of on-premise AI in 2026, what to give the model to read first, how operational data becomes AI-readable, and the governance question that does not get asked until something has gone wrong.
AI is no longer a question of whether or which. It is a question of what underneath — and the operators who get this right will spend the next two years building capabilities their competitors will need three years to match.
hashtag#AirportOperations hashtag#AviationAI