AI Agent for Airport Transfer Quotes and Dispatch: Human Final Call

AI Agent for Airport Transfer Quotes and Dispatch: Human Final Call

How to structure per-route transfer quotes, flight timing and driver confirmation, with final dispatch by the control desk and clear limits.

•

Want to apply this in your company?

Book a call with Juan →

Superintelligence is not what organizes an airport transfer, and treating it as if it were creates the wrong expectations. What works in practice is a workflow: an AI agent receives the request, builds a per-route quote from an approved rate table, checks the flight time against driver availability, and passes the final confirmation to the control desk, which dispatches. People who own pricing, drivers and exceptions stay in charge of the operation, with a record of every step.

Key takeaways

  • A quote depends on the route (origin and destination), and the agent only answers with prices that appear in the current approved table.
  • Flight timing changes the operation. The booking stores the flight number, the expected landing time and the agreed waiting tolerance, not just the time the customer typed.
  • The driver confirms before dispatch, the acceptance is recorded, and there is a plan for refusal or no response.
  • Final dispatch belongs to the control desk. The agent prepares, a person approves, and any last-minute change goes through them.
  • Passenger data enters the workflow only in the minimum amount needed, with attention to LGPD.

What the customer asks and what the agent can answer

The most common question is simple: how much does it cost to take one person from the airport to a specific address? The answer, however, depends on the route, vehicle type, time of day, a possible waiting surcharge and any access fees. A well-designed agent answers only when the route is registered and the price comes from a known rule. When the request falls outside the table, it does not improvise a number. It records the details and sends the case to the control desk.

This sounds obvious, but it is the first source of error in operations like this. A language model is good at conversation and unreliable at remembering tables that change. For that reason, the pricing rule should live outside the agent's free text, in a source of truth that the operation controls.

How to structure per-route quotes

Route-based quoting works best when the company separates calculation from judgment. Calculation (a fixed table, a rate band by vehicle type, an agreed waiting surcharge) should be deterministic. Judgment (a new route, a group booking, non-standard luggage) should go to a person.

Step by step, here is how to build this stage:

  • List the routes the company actually serves, with airports, destination areas and available vehicle types.
  • Define who approves pricing rules, and record the version of the current table.
  • Separate the deterministic rules from the ones that require human analysis.
  • Write the explicit limit: the agent quotes only registered routes and forwards the rest.
  • Record each quote with the route, table version, request time and, where approval applies, who approved the reply.

With this design, the control desk stops writing quotes from scratch and starts reviewing exceptions, which is where human time makes a difference.

Flight timing: what to track and what not to promise

The flight is the axis of the transfer. The booking should store the flight number and the expected landing time, along with the waiting rule agreed with the customer. If the company checks flight status through an authorized source, the agent can flag delays or cancellations, but the decision to reschedule the driver belongs to the control desk.

The most common mistake is promising an exact pickup time. The safer rule is to agree on a window: the driver waits after landing within the agreed tolerance, and any change beyond it is recalculated by the control desk.

Illustrative scenario: a passenger books a transfer for a flight that arrives late. The agent records the change the passenger reported, alerts the control desk with the new expected landing time, and does not confirm a new pickup time on its own. The control desk adjusts the schedule, confirms the driver and notifies the customer through the agreed channel.

Driver confirmation and the control desk's decision point

Driver confirmation is a two-way step. The agent checks the schedule, sends the offer through the channel the company uses, records the acceptance or refusal, and starts a timeout. If there is no response, the case returns to the control desk with a list of alternatives. It does not sit idle in the system.

A simple decision guide helps separate what can be automatic from what needs approval:

  • Confirm automatically: registered route, vehicle available, time within the window, and no recent change to the booking.
  • Send to the control desk: route outside the table, flight delay beyond tolerance, group larger than standard, or special luggage.
  • Decline with an explanation: address outside the served area, or no driver available for the requested period.
  • Never decide alone: swapping an already confirmed driver, refunds, extra charges or price exceptions.

Final dispatch is an action of the control desk. Even when the agent suggests a schedule, a person with access to the full operational picture releases the driver.

Passenger data, LGPD and records

A transfer involves a name, a phone number, a flight number, a hotel or home address and, sometimes, travel time details. This is personal data and should be handled with the same care as any customer record. The practical recommendation is to collect only what the operation needs to deliver the service, define how long that data is kept, and keep documents and payment data out of the chat.

The National Data Protection Authority (ANPD) publishes guidance on LGPD obligations, and it is worth reviewing the official documentation before defining the workflow. For assessing the risks of AI systems in a structured way, the NIST AI Risk Management Framework is a useful reference. It is also important that every automated decision has a readable record, because the company must be able to explain why a request was quoted in a particular way.

When service runs through WhatsApp, the rules for sending messages also matter. Before proactively telling passengers about a schedule change, it is worth understanding the 24-hour window rules of the official API, covered in the article on the 24-hour WhatsApp window.

How to pilot one route before expanding

The safest way to start is with one route or a small set of real routes. Our method follows four steps: map the current process and its baseline, limit data and actions, pilot a real workflow with records and an agreed metric, and then hand the operation to the team, deciding afterward whether the scope should grow, be adjusted or stop. In practice, the metric can be simple, such as the share of quotes the control desk approves without editing, agreed before the start.

The pilot also helps separate what AI does well from what should stay with people. A procurement agent, for example, can compare quotes and request approval, a logic similar to the transfer workflow. The AI agent for procurement is worth reviewing as a design comparison.

Superintelligence, general AI and what is realistic today

The term superintelligence generally describes systems with capabilities far beyond human ability across almost all intellectual tasks. It is a subject of technical debate, not a product category. For a transfer company, what exists today and can be run responsibly is narrow AI inside a controlled workflow. The difference between the two changes investment and risk expectations, and it is worth reading about superintelligence and ChatGPT before writing the internal proposal.

The most useful reading for leadership is this: the gain comes from workflow discipline, with an approved table, clear limits, a control desk with real authority and complete records. None of that depends on superintelligence.

Frequently asked questions

Can superintelligence run an airport transfer operation on its own? No. An AI agent can organize requests, quote registered routes and collect driver confirmations, but the operation needs pricing rules, scheduling and a control desk that decides exceptions. Treating the system as the single source of decisions increases the risk of errors and of promises the company cannot keep.

Can the agent give the transfer price directly to the customer? Yes, as long as the amount comes from a table approved by the operation and the route is registered. When the request falls outside those conditions, the right step is to record the details and forward them to the control desk, without risking a number. This rule should be written into the workflow design and tested before real use.

How does the system handle a flight delay? The booking should store the flight number and a waiting tolerance agreed with the customer. If the change goes beyond that tolerance, the agent alerts the control desk, which recalculates the schedule and informs the passenger through the agreed channel. The agent should not confirm a new pickup time on its own.

Which passenger data is needed, and how long should it be kept? Collection should be limited to what is necessary to deliver the service, such as name, contact details, route, time and flight number. The retention period must be set by the company based on LGPD and its own business obligations, and it is worth checking the documentation of the National Data Protection Authority before finalizing the policy.

Sources

airport transferAI agentsdispatch operationsLGPDquotes