The risk is not the model, it is the boundary
Travel is a trust business. A confident, wrong answer about a visa requirement, a supplier policy or a cancellation window does real damage — to the traveler first, and to the advisor's reputation immediately after.
The question that decides whether AI helps or harms a travel business is not which model is used. It is what the agent is allowed to read, what it is allowed to say, what it is allowed to do, and where it must stop.
Four boundaries every travel agent role needs
Before an intelligence role goes anywhere near a client, four things are written down:
- Scope: the specific work the role performs, and the work it explicitly does not
- Sources: the approved material it may answer from, with a precedence order when sources disagree
- Actions: what it may do in the system, and what requires a person
- Approval: the point where an advisor reviews or releases the output, and how escalation works
Supervision before the client sees anything
A supervision layer checks output before it reaches a traveler: is this grounded in an approved source, does it follow brand and tone, does it carry any disclosure the situation requires, and does it need a human release.
When the answer is not in an approved source, the correct behaviour is to say so and route to an advisor. A system that would rather invent than defer is not ready for client contact.
Disclosure and the advisor relationship
An agent supports the advisor; it never impersonates one. Semi-assisted work such as partner onboarding is prepared for the advisor to review and send — it does not accept terms or make commitments on the advisor's behalf.
Affiliate and referral arrangements are disclosed where they appear, and they never interfere with the booking path the traveler expects.
Attribution as a standard
Every AI-produced output is attributed and recorded. When something goes wrong — and eventually something will — the business needs to be able to see which role produced what, from which source, and who released it.