AI Dispatch for Limo Companies: Flight Changes, Drivers and Pickups
An AI agent can watch flight changes, suggest drivers and send pickup confirmations for an airport transfer company. Dispatch keeps the final calls.

An AI dispatch agent for a limo or airport transfer company can watch flight status changes, rank driver options for each trip and send pickup confirmations to passengers, all through tools and permissions the company sets. The agent handles the repetitive checking and drafting, while a dispatcher keeps authority over driver decisions, exceptions and anything that changes price or service promises. The safest first step is a narrow role with written approval rules, not a general booking assistant.
Key takeaways
- Flight-change handling is the strongest first job because the trigger is clear and the agent can propose a change before it acts on one.
- Driver assignment works best as a ranked suggestion with visible reasons, not as an automatic dispatch.
- Every booking needs a pickup status that a person can see before the shift starts, so unconfirmed trips never hide in a message thread.
- Dispatch keeps exceptions, repricing, disputes, vehicle problems and any decision that involves a refund or a waiting fee.
- Start with one airport, one vehicle class and one passenger channel, then widen the scope only after the logs show the agent following its written rules.
Why airport transfer dispatch breaks at the edges
Most airport transfers run without trouble. The failures cluster around change: a flight lands early, a delay pushes arrival past the driver's shift, a passenger replies by text to a number nobody monitors, or an airline moves a flight to another terminal. Dispatchers rebuild the day by phone and memory, and the record of what was promised often lives in three places at once. That is the work an agent can take over, because it is mostly reading, comparing and writing things down consistently.
The value is not that the agent is faster at one call. The value is that it watches every booking continuously, applies the same buffer rules each time and leaves a readable trail for the next person who picks up the shift.
What an agent can run for flight changes
Flight-change handling is the cleanest starting point because the input is a structured event and the output is a bounded set of actions. A reasonable plan looks like this:
- Receive flight status events for every booking that has a flight number and date, and accept only events from an authenticated source.
- Compare the new arrival time with the scheduled pickup, the airport's buffer rule and the assigned driver's current hours.
- Draft a revised pickup time using the company's own buffer for that airport and terminal.
- If the change falls inside the buffer, send the approved change message through the passenger's original channel and log it.
- If the change falls outside the buffer, or the terminal or drive changes, place the trip in the dispatcher's queue with the two or three options already drafted.
- Write the source event, the decision and the message to the booking record so the history is never reconstructed from memory.
Webhook-style delivery needs the same care as any inbound integration. Stripe's documentation on webhooks describes verifying that an event really came from the provider before acting on it, and the same discipline applies to a flight feed or a booking platform. An agent that trusts unverified events will eventually move a pickup because of a bad message.
Driver assignment: suggest, rank and explain
Driver assignment is where most dispatchers feel the pressure, and where an agent is most useful as a reviewer rather than a decider. For each trip, the agent can rank the available drivers by vehicle class, licence and permit status, hours already scheduled, the last drop-off location and any passenger needs stored in memory, such as a child seat, a wheelchair accessible vehicle or a preferred language. The output is a short recommendation with the reasons written next to it, so the dispatcher can accept it in seconds or override it with one note.
The agent should not reassign a driver on its own. A reassignment changes who a passenger is waiting for, and it can affect fatigue rules, insurance records and the driver's own schedule. Keeping that choice with a person is what makes the suggestion trustworthy.
The agent proposes the driver. The dispatcher commits the driver.
Illustrative scenario: A passenger's inbound flight is moved to an earlier arrival that is still inside the airport buffer. The agent updates the pickup time, sends the approved change message and notes the source event on the booking. The same flight, if it moves to a different terminal that changes the drive, goes to the dispatcher with two drafted options and the driver ranking already attached.
Pickup confirmations and the unconfirmed list
Pickup confirmation is routine work that often goes undone under pressure. The agent can send the confirmation at a fixed point before the pickup, asking the passenger to confirm the meeting point and time, and it can give each booking one of a small set of states: scheduled, confirmed by passenger, unconfirmed, changed or disputed. The unconfirmed list, sorted by pickup time, is the single most useful screen for a dispatcher starting a shift.
Passenger replies deserve particular care. Messages are untrusted input, and a passenger, or someone pretending to be one, can write text that looks like an instruction to the agent. OWASP's guidance on prompt injection prevention is clear that an agent should act only on a fixed set of intents, with outputs checked against those intents before they reach a system. In practice, a reply that says a new address is a change request for a human. A reply that asks the agent to change the price or cancel the booking is simply routed to dispatch.
What stays with dispatch
Some decisions carry judgment, money or liability, and they should stay with a person even when the agent is working well. A practical boundary list for most transfer companies includes:
- Accepting, declining or repricing a trip, including last-minute upgrades.
- Waiting time, meet-and-greet exceptions, tolls, parking and any fee that a passenger or corporate account will see.
- No-shows, missed passengers and disputes about what was promised.
- Vehicle breakdowns, accidents, driver fatigue or any safety concern.
- Corporate account changes, contract terms and any statement that promises a refund or a credit.
- Final driver reassignment whenever the suggestion is close, unclear or conflicts with a driver's own request.
These lines are not a sign that the agent is limited. They define where a useful agent stops, and they are easier to defend to a client or an insurer when the record shows who made each decision.
How to set the permissions for a first build
Access control is the part most often skipped in early demos. OWASP's authorization guidance points toward least privilege, meaning each tool the agent uses gets only the rights it needs. For a transfer company, that usually means read access to the flight feed and the booking calendar, write access to booking notes and status fields, permission to send templated messages on the approved intents, and no access to pricing, payment or refund functions at all. Each permission should have an owner in the company who can review it, and the review should happen on a schedule, not only after something goes wrong.
How Be Human runs this as a managed service
Be Human starts with one useful role for the dispatch desk and defines what the agent remembers, including airport buffers, vehicle classes, driver credentials and passenger needs, as well as the tools it can call and the approval rules for each action. The success criterion is written before the build, for example that every trip has a named status and a timestamp by a set point before pickup. After launch, the managed service covers monitoring of the agent's actions, recurring evaluations against real booking cases, maintenance when a flight feed or booking platform changes, permission reviews and controlled improvements that are tested before they reach live dispatch. The work is remote and available in English and Portuguese. If you want to find the first dispatch task worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Can an AI agent replace a dispatcher at an airport transfer company? In a well-run first build, no. The agent watches flights, drafts changes, ranks drivers and tracks confirmations, which removes a large share of manual checking from the dispatcher's day. The decisions about drivers, exceptions, money and safety stay with the dispatcher, and that boundary is part of the design.
What happens when a flight is delayed and the passenger does not reply? The booking moves to the unconfirmed list with the new estimated arrival attached, so the dispatcher sees it before the shift. The agent can send one follow-up at the approved point, but it should not invent a confirmation or change the pickup without the buffer rule allowing it. A human then decides whether to wait, call or reassign.
Does the agent need access to our pricing or payment system? It should not. A dispatch agent needs the booking calendar, the flight feed, the driver roster and the messaging channel, and nothing that sets prices, issues refunds or stores payment details. Keeping those functions out of reach limits the damage that any mistake or manipulated message could cause.
How do we know the agent is behaving correctly? You check it against cases you already understand. Build a small set of past bookings with known outcomes, including a few tricky flight changes and a few messy passenger replies, then run the agent against them after every change to its rules, tools or memory. The logs of live actions, reviewed on a schedule, show whether it stays inside the approved intents.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


