How AI Is Transforming Travel Personalization

Personalisation has been a goal in travel for a long time. The idea is easy to state: show each traveller the flights, hotels and packages that suit them, at a price that works for both sides. In practice, most travel sites still show the same results to everyone, sorted by price. Advances in machine learning and large language models have made better personalisation more practical, but they have also made it easy to build features that look impressive and do little.

This article looks at where AI is useful in travel personalisation today, where it is not, and what travel businesses should get in place first.

Recommendations that use the right signals

Recommendation engines work by finding patterns in behaviour: what people searched, viewed and booked. In travel, those signals are sparse. A typical leisure customer books a few trips a year, so there is far less data per person than in retail or streaming.

Useful recommendations in travel tend to rely on context more than long personal history:

• The current search, including dates, party size and destination.

• Patterns from similar searches, for example families searching beach destinations in school holidays.

• Agency or market preferences in B2B, such as which hotel chains an agency tends to sell.

For a B2B platform, the agent is often the best signal. Ranking hotels by what an agency has successfully sold before, in that destination, is a simple model that can make results more relevant without needing any personal data about the traveller.

Dynamic pricing with clear limits

Airlines have used revenue management for decades. Smaller travel businesses, such as charter operators and hotel wholesalers, can now apply similar ideas to their own inventory. Models can suggest price changes on a departure based on how quickly seats are selling compared to similar departures in the past.

We think these models should advise rather than decide. In Vee Money, pricing is driven by rules that people can read and change: markups by agency, product, season and channel. A model can suggest adjusting a rule, but the rule stays visible and a person approves it. This keeps prices explainable to agencies and avoids surprises such as different prices for the same seat with no clear reason.

Pricing based on individual characteristics also raises fairness and regulatory questions in many markets. Pricing by demand, season and channel is generally better understood than pricing by who the customer is.

Language models for search and service

Large language models make it possible to accept requests in plain language, such as "a week in Greece in June for two adults and a child, near a beach, under a set budget", and turn them into structured searches. This can help customers who do not know exactly where they want to go.

The important word is structured. The model should translate the request into parameters that the booking engine understands, and the results should come from real inventory with real prices. Generated text that describes hotels or fares without checking live availability creates expectations the business cannot meet.

Language models are also useful for operations: summarising long supplier emails, drafting responses to common customer questions, or classifying incoming requests so they reach the right team. These uses are lower risk because a person reviews the output.

Data comes first

Most personalisation projects fail on data, not on models. Before investing in AI features, a travel business should check whether it has:

• Clean booking records with consistent product, supplier and destination data.

• Search and click events linked to bookings, so it is possible to learn what led to a sale.

• A clear consent basis for using customer data, and a way to respect opt-outs.

A unified platform helps here. When search, booking, operations and finance share one data model, it is much easier to connect what a customer looked at with what they booked and what it earned.

Where to start

For most travel businesses, the most practical first steps are modest. Improve ranking of search results using booking history. Use language models to help staff with routine text work. Add pricing suggestions on own inventory, with people approving changes. Measure each change against the previous approach before expanding it.

AI can make travel platforms more relevant and more efficient, but only on a foundation of good data, real inventory and clear rules. Treat it as a set of tools for specific problems rather than a feature in itself, and keep the traveller's trust at the centre of every decision.