What AI can and cannot automate in a travel quote

Artificial intelligence already handles the mechanical part of a quote well —interpreting the request, finding the services that fit, applying the commercial rules and generating the document— and it does not handle the part where the business is defined. The usual confusion runs opposite to what it seems: the hard part is not writing the itinerary, it is guaranteeing that every price in it corresponds to a rate that exists and is still valid.

Why it matters

Because the promise of "AI for quoting" is being made across the entire industry, and not every version means the same thing. There is a huge difference between a system that writes fast and one that quotes accurately, and from the outside they look almost identical in a ten-minute demo.

The cost of not telling them apart is concrete: a rough quote that goes out fast, that the client accepts, and whose error surfaces once a price commitment is already in place.

What gets automated well today?

Why is building an itinerary easy and pricing it hard?

An itinerary is text. If it says "airport transfer and free afternoon in the historic center" instead of "free afternoon in the old town," nobody is harmed. Text tolerates approximation.

A quote is a price commitment. If the number is approximate, the error is not noticeable when you read it: it shows up when the client accepts and the trip has to be operated. And unlike text, there is no way to fix it at no cost, because by then the agency has already given a price.

That is why itinerary generation has become a relatively common capability —anyone with access to a language model has it— and accurate quoting has not. The first is a writing problem. The second is a data problem.

Can an AI make up a rate?

Yes, and it is the central risk in all of this.

A language model, if it is not required to take the price from somewhere, produces a plausible number. It will have the right format, the right currency and a reasonable order of magnitude for that destination and that hotel category. It will look like a rate. And it will not correspond to any real rate.

What prevents that is not a better or bigger model. It is that the model does not decide the price: the quote can only be built with rates that exist in a current data source, and if a rate is missing or does not apply, the system says so instead of filling the gap. It is a design constraint, not an incremental improvement.

Translated into a question to ask any vendor offering AI for quoting: where does every number in this proposal come from, and what happens if that rate does not exist? The right answer is boring and verifiable.

Where is a person still needed?

In three places, and not as a temporary stage:

In the commercial decision. Which supplier makes sense for this particular client, what margin to apply on an account you want to retain, when to give up profitability for volume. That is not a calculation, it is agency policy — and it changes depending on the client, the time of year and the relationship.

In the supplier relationship. An allotment secured with a phone call, a spot rate negotiated for a specific group, a favor asked in high season. No system replaces that, and it is a large part of what gets an inbound tour operator chosen.

In reviewing what falls outside the pattern. The unusual request, the group with special requirements, the exception that had never come up before. Well-placed human oversight does not review everything: it reviews what the system flags as unusual.

What automating quotes does not solve

It does not improve a poorly negotiated rate. It does not organize a rate sheet that arrives in a mess. It does not replace destination knowledge, which is exactly the asset an inbound tour operator sells and which often goes unused today because the team is busy looking up prices.

And it does not remove the person from the process. What it does is move them: from shuffling information to making decisions. Which is, incidentally, the part of the work the client actually pays for.