How gen AI helps share expert travel advice at Hotelplan Suisse

In a travel market dominated by mass tourism and standardized package deals, our customers choose Hotelplan Suisse for exceptional, memorable travel experiences. Our team of more than 400 experts provides a personal touch, supporting customers with insider knowledge and expertise, local tips, and up-to-date information.

Our customers trust us to deliver an unforgettable experience, tailored to their taste. With specialists on each of the 82 destinations we serve, our team boasts a wealth of expertise. The challenge we faced, however, was how to share that expertise across the organization and still get it to the right customer at the right time.

Travel booking periods are seasonal, and during high demand, our sales and operations team faced a bottleneck when directing customer queries to the relevant travel expert, delaying their ability to provide relevant information. Complex inquiries, possibly involving partner companies, could result in a one-day wait for an answer or a week-long wait for a complete itinerary.

Not only was this creating unnecessary friction, but it also meant we risked losing new customers. We were determined to find a more efficient way to give customers the expertise they needed as quickly as possible.

Generating fast responses with the help of large language models

Our first idea was to build a peer-to-peer support network. However, we found it took time for the right expert to see the question and respond, meaning customers continued to experience delays. With the emergence of generative AI, we saw the potential to allow our team to query a large language model (LLM) to find answers to customer questions.

However, the information provided by the first third-party LLM we used was too generic for our purposes. It wasn’t until we tried Google AI PaLM 2, and discovered we could layer our proprietary data over the model, that we knew we had found the solution.

Working on a proof-of-concept with Google Cloud and its partner Datatonic, we added a few basic data sets to the model to build a simple chatbot in less than two months. While the output was initially limited by the small amount of data we had fed it, as we began to add more, we noticed a rapid improvement in the chatbot’s ability to deliver accurate, detailed responses to the kinds of questions our customers ask every day.

Because we could add unstructured data to the model, we constantly added new information from our experts to train the model, making it increasingly accurate.

Removing the bottleneck from the sales process

With the solution currently being tested by our sales team, members can now enter queries into the chatbot to provide our travel experts with the necessary information to consult our customers with accurate, detailed travel expertise in real time, removing the bottleneck in the booking process.

Any questions that stump the chatbot are escalated to our travel experts, who now have more time to respond in detail. Those answers are then fed back into the model, creating a feedback loop to continually improve responses. Because many of the data sources to the model have been configured to update automatically, those responses remain current.

The biggest challenge now lies not with the chatbot and our ability to quickly provide our customers with expert information, but in making it readily accessible for our travel experts within their working environment.

To this end, we are planning to develop a Google Chrome extension to ensure the chatbot is always visible and available for every team member when handling customer queries. We are also considering how the chatbot might help handle customer queries outside office hours.

Creating expert content at scale with generative AI

Building on our success with PaLM 2, we are now exploring other ways to use generative AI to provide expert travel advice to our customers. One such project involves using Gemini to generate a first draft of website content such as hotel and destination descriptions, fine-tuned to the tone of voice of the three brands under the Hotelplan Suisse umbrella.

We are also using Gemini to generate blog posts bylined by our team members. With hundreds of experienced travelers at Hotelplan Suisse around the world, we have a vast amount of knowledge to share. However, not all our employees enjoy writing, so much of their unique expertise remains untapped.

By storing all this travel experience in Cloud Storage, we are using Gemini to automatically generate blog posts based on these personal experiences, creating a large amount of new content packed full of tips and advice for our customers.

With this project set to go live in the coming months, the plan is to provide even more of a personal touch by enabling customers to use these blogs to find the travel experts that best relate to their needs and connect with them directly. For example, a parent could search for a blog post with tips on traveling with small children, and then connect with that travel expert for more information.

The more we work with Google Cloud’s generative AI solutions, the more potential we see and are excited about all the possibilities. With the support of the Google Cloud team, we are confident that innovative solutions are not only achievable but will also enable us to continue providing exceptional travel experiences for our customers long into the future.