From our concierge AI work
Worked examples
Use cases drawn from IIG’s work building destination concierge AI and operational AI in the Caribbean.
Institutional memory
“Have we seen this before?” — making SOPs, audits, and renovation deliverables queryable
The first valuable use case at a smaller airport or resort is rarely what vendor brochures suggest. It is the question nobody can currently answer quickly: have we seen this before? Safety incidents, audit findings, corrective actions, ICAO annex updates, and revised SOPs accumulate every year. The portion of that knowledge searchable next year, by anyone except the original author, is small.
IIG deploys a local LLM grounded against the operator’s own document corpus. A safety officer can ask whether a recent ramp incident matches a five-year pattern. A compliance officer can identify which procedures need revision after an annex update. A new operations manager can see how a recurring weather scenario has historically been handled, and why.
The technical and engineering departments often have the most acute version of the problem: a major renovation generates thousands of pages of drawings, specifications, manuals, warranties, and certifications — delivered in the contractor’s structure rather than the operator’s. The corpus becomes inaccessible the moment the project closes. A local LLM reads the entire delivery, extracts metadata, proposes a structure that fits the operator’s operations, and surfaces what is missing or contradictory.
LessonNone of this requires the model to invent anything. It is institutional memory made queryable, on data sensitive enough that on-premise is the only acceptable deployment.
Proactive passenger communication
St. Maarten road incident → missed flights — closing the loop with AI
On Sint Maarten, where the road network cannot easily route around an incident, a single accident on a key arterial can turn a 35-minute drive to the airport into a two-hour ordeal. The airport typically learns about it when passengers start missing flights. By then, every downstream consequence is already in motion.
An AI connected to local traffic data, the flight schedule, and the check-in feed can identify which passengers are most likely affected and propose the message that should go out, when, and through which channel — SMS, WiFi-portal push, AI concierge alert, or hotel front desk notification. The duty manager keeps decision authority; the AI removes the manual reconstruction work that currently happens under pressure.
LessonThe same connectivity layer IIG deploys for Destinito WiFi and AI concierge becomes the proactive channel during disruption — not a separate system.
Single-runway cascade
Modeling operational cascades in real time at single-runway airports
Most Caribbean regional airports operate with one runway, which means any incident creates a cascade: a medical diversion arriving unscheduled, a runway inspection extending, a thunderstorm sitting over the approach. Working the consequences out in real time is currently the job of an experienced duty manager, supported by phone calls and institutional memory.
An AI with access to surface movements, weather, schedules, and capacity can model the cascade in seconds, identify the choke points, and propose the sequence of decisions a duty manager would otherwise have to construct under pressure. It does not replace the duty manager. It gives them a faster, clearer picture, with the reasoning shown.
LessonOperational AI is hardest where the consequences are real. Retrieval-augmented generation, source citation, and confidence indicators are not optional — they are how the duty manager learns to trust the system.
Destination concierge
Multilingual AI concierge grounded in Destinito’s structured data
For traveler-facing use cases, IIG’s concierge AI runs against the structured Destinito destination database: verified businesses, hours, menus, prices, location, events, transport, and accessibility data. The model is not free-styling the destination from training data; it is reading a curated, verified source that the operator controls.
That grounding is the difference between an AI that can be deployed in regulated travel environments and one that cannot. Every answer can be traced back to its source record. When the underlying data is updated, the AI’s answer updates — no model retraining required.
LessonThe model is an implementation detail. The data, the governance, and the use case are what matter.