AI Prototype: Booking Assistant for a Travel Agency

TravelPlus is an owner-operated travel agency in St. Gallen with three experienced travel advisors and a strong focus on tailor-made trips. The company primarily serves repeat customers and their referrals. The USP lies in personal consultation and the ability to assemble individual travel programs. The problem: every custom trip requires 3-5 hours of research, during which the advisor searches through hundreds of offers from various tour operators to find the optimal combination of flight, hotel, and activities.

Managing director Marco Bianchi faced a dilemma: personal consultation is the heart of his business model, but the research time per customer was economically barely sustainable. With an average revenue of CHF 4,500 per booked trip and 3-5 hours of research time plus 1-2 hours of consultation, margins were thin. He wanted to reduce research time without lowering the quality of suggestions. His discerning customers expect tailored recommendations, not standard packages.

The core problem was the enormous research time: each advisor manually searched through 5-8 different tour operator portals, compared prices, read hotel reviews, checked availabilities, and assembled 3-5 travel options. Information had to be manually transferred into a Word document. With 15-20 customer inquiries per week and 3-5 hours of research per inquiry, the three advisors spent most of their working time in front of screens instead of in customer conversations.

We used our AI Prototype Lab to build a booking assistant in 4 weeks. The prototype works in three steps: First, it captures customer preferences through a structured interview (budget, travel period, interests, comfort level, previous trips). Second, it searches an aggregated database of travel offers and applies AI-based matching to identify the most relevant combinations. Third, it creates a professional travel proposal document with images, descriptions, prices, and a personalized explanation of why this offer suits the customer.

The biggest challenge was not the AI itself but aggregating travel data. Different tour operators describe their offers in different formats. The AI had to learn to compare a 'Meerblick-Suite' from provider A with a 'Sea View Junior Suite' from provider B. We trained a special matching model that understands the semantics of hotel descriptions. Particularly exciting: the advisors tested the prototype with 20 real customer inquiries from recent months and compared AI suggestions with actually booked trips. In 75% of cases, the AI shortlist included the actually booked hotel or an equivalent one.

The prototype proved feasibility and was developed into a production solution. Results after 3 months of productive use: research time per customer dropped from 3-5 hours to 1-2 hours, a 60% reduction. Advisors use the freed time for more and better customer conversations. Customer satisfaction increased because suggestions came faster and were often even better. The AI considers factors a human might overlook during manual search (e.g., rating trends or seasonal price differences). Revenue increased by 15% because advisors can serve more customers per week.

The most important insight: the prototype approach saved the travel agency at least CHF 30,000 compared to direct production development. Had we immediately built the full solution, data aggregation would have been the biggest risk. The prototype showed early that 85% of relevant travel data could be covered through three main sources instead of integrating all eight portals. The advisors emphasized that the AI does not replace their work but enables them to do what they do best: personal, emotional travel consulting.