Booking.com said generative artificial intelligence is providing it with a much better view of traveler intent.
Matthias Schmid, senior vice president of accommodations for the online travel platform, said the company now has a "much better understanding" of who the customer is and can surface more relevant hotel content as a result.
Subscribe to our newsletter below
Schmid, who was speaking during a session at ITB Berlin this week, said the technology "is not an evolution, it is a revolution" and that Booking.com is "at the forefront and well positioned" to apply generative AI.
He said the company has been using machine learning for more than a decade and has hundreds of models in place with use cases including making offers more relevant, language translation, pricing and fraud detection.
Booking.com divides generative AI into three different buckets, he said. The first is a focus on internal productivity for software engineers who use coding assistants. AI tools are also used to summarize customer center calls, and the company encourages employees to access and use generative AI assistants more widely to aid productivity.
The second bucket is using the technology to improve its machine learning models and make them more agile. The final use is around traveler intent with the aim of making it easier to transact on its marketplace.
"We now have a much better understanding if someone is a female traveler, for example, or a family with two underage kids and based on that we surface different property descriptions. We make the description of the hotel much more relevant for that traveler's intent," he said.
"OTAs are really good at search, book and pay. That's the core thing an online travel agency does. Until now we were not very good in the planning stage. You came to Booking when you knew the destination and the period when you wanted to travel. Now with gen AI, it's different. You can start much earlier in the planning phase."
The approach is not dissimilar to that shared by Expedia Group in its recent fourth quarter and full-year 2024 earnings call. The OTA's CEO Ariane Gorin outlined three strategies for using AI including operational efficiencies and driving more value for travelers.
Schmid said there is also a big benefit for partners. Currently Booking.com is very dependent on content that partners provide to it about properties but, with generative AI, it can gather information from many more sources such as from images or customer reviews.
Booking.com launched its generative AI trip planner in June 2023, a conversational user interface with a booking widget. Since then it has developed smart filters to focus in on travelers' requirements and property Q&As, which enabler users to ask more specific questions of properties for a more tailored approach.
Schmid said the functionality is live in the United States, United Kingdom, Australia, New Zealand and Singapore and will be launched across European countries in the coming months.
The company is also close to launching a trip assistant for the post-booking stage as well as a car rental pick-up assistant to help travelers understand the required documentation to pick up a car.
He also spoke about experimentation with agentic AI and its partnership with OpenAI.
"We did some experiments where we asked the agent to do a reservation and as a customer you can see what the robot or agent does," Schmid said.
"It's still a little bit clunky but we're convinced that six months from now these problems will be solved. It's also quite interesting as an experiment because if the agent has a problem in a certain step in the booking process it's quite likely that a human also has a problem during that step. So, it's a great indication of where there is improvement across the user experience."
The company is also thinking about how search is going to develop given its significant investment in paid search and Google but "the jury is still out" on how that will evolve with generative AI startups, Schmid said.
He also said that Booking.com is integrated with four or five different large language models and has built a generative AI integration layer over the past 18 months that decides which LLM is best suited to specific tasks.
"In this integration layer we also built in all the privacy checks, we also make sure there is no hallucination so there are a lot of checks happening and this architecture allows us to democratize gen AI across the organization," Schmid said.
He said it meant product and engineering teams can access LLMs in this secure, protected environment.
"Very often when you talk about gen AI it sounds like it is plug and play. Yes it is on a small scale but if you apply it on an enterprise level it's complex."
Schmid went on to discuss the importance of data and whether there would be increased collaboration around data sharing or increased competition. He said companies needed to think about whether their data is really proprietary or already being scraped by the generative AI platforms.
"it's really for companies to identify which of the core data is proprietary and whether it should be protected," he said.
"For us it's all the customer behavior, historic transactions, search behavior so there I don't think we'll see industry collaboration because this is the new gold."
On whether generative AI would level the playing field for smaller organizations, Schmid there were certain use cases such as translations where it would.
"If it's related to more complex use cases, it's a lot of effort, it's a lot of investment and it takes time."
Schmid ended by saying that using AI is not a choice and urged everyone in the travel industry to begin experimenting with it.