What It Takes to Make AI Booking Work at Scale

This "pilot purgatory" is where most travel tech companies currently reside, struggling to bridge the gap between a generative AI response and a confirmed, paid booking. However, the landscape is shifting as early movers begin to reveal the results of long-term development. HotelPlanner, a major player in the group travel and hotel booking space, has recently emerged as a primary example of this transition. By launching Reservations.ai, the company is moving beyond the theoretical, offering a voice-driven, fully integrated booking system that has been battle-tested within its own high-volume marketplace before being offered to the wider industry.

The genesis of Reservations.ai was not a sudden pivot to catch the generative AI wave, but rather a strategic response to the existential crisis of the COVID-19 pandemic. As global travel ground to a halt, HotelPlanner’s core business—group bookings—evaporated. This left a massive workforce of group sales managers and hotel directors with decades of institutional knowledge but no customers to serve. Tim Hentschel, co-CEO of HotelPlanner, recognized that this human expertise was the company’s most valuable latent asset. The company transitioned its traditional call center into a gig-powered model, tasking these experts with handling diverse travel inquiries while meticulously recording the interactions.

By the time the AI revolution reached a fever pitch in 2023, HotelPlanner was sitting on a goldmine: 10 million highly rated, recorded calls detailing the nuances of travel sales, customer objections, and localized hotel knowledge. This data served as the bedrock for training "Skyler," the AI-powered travel agent at the heart of Reservations.ai. Unlike generic large language models (LLMs) that pull information from the open web, Skyler was trained on specialized, high-conversion sales data specific to the travel industry.

The technical hurdles to making an AI booking agent work at scale are significant. In the early stages of development, Hentschel admits the results were "stunningly bad." Initial conversion rates hovered around a meager 1% to 2%. The AI struggled with the complexities of human speech—accents, local slang, and the non-linear way people often plan trips. To reach human-level performance, the development team had to refine the back-end technology to process natural language in real-time without the latency that often kills a phone sale. Over a year of iterative testing against live traffic, the conversion rate climbed to between 14% and 16%, effectively matching the performance of a professional human reservationist.

Today, Reservations.ai is a comprehensive ecosystem rather than a simple tool. It possesses the same access to inventory as any human agent at HotelPlanner, spanning direct hotel connections, millions of global flight routes, rental car fleets, and even event tickets for concerts and sporting events. This multi-supplier integration allows the AI to function as a holistic travel consultant. A customer can call, speak naturally about their needs for a weekend trip to London—including a hotel near Wembley, a flight from New York, and tickets to a specific show—and the AI can navigate those disparate databases simultaneously to provide a curated itinerary.

Crucially, the system handles the "last mile" of the transaction: payments and confirmations. Many AI tools in the market can suggest a hotel, but they often hand the customer off to a web form to finish the booking. Reservations.ai is designed to take credit card information via voice, process the payment securely, and issue the confirmation within the call. This end-to-end capability is what Hentschel believes differentiates the product in a crowded market. The system currently handles roughly 500,000 calls per day, a volume that would require a massive, cost-prohibitive human workforce to manage manually.

The rise of agentic AI—AI that can take actions rather than just provide information—is forcing a total rebranding of the company. HotelPlanner is increasingly positioning itself as a "reservations company" rather than just a "hotel company." This is reflected in the shift toward the Reservations.ai brand at the top of their corporate hierarchy. The logic is clear: if the AI can sell a flight or a car as effectively as a hotel room, the brand should reflect the entire travel journey.

For the broader travel industry, the emergence of Reservations.ai presents a "build vs. buy" dilemma. Online Travel Agencies (OTAs), Destination Marketing Organizations (DMOs), and Travel Management Companies (TMCs) are all racing to integrate AI, but the cost of entry is rising. Building a proprietary voice-activated booking system requires not only sophisticated engineering but also an enormous repository of proprietary data to train the models. Hentschel argues that for companies that haven’t spent years managing their own call centers and data streams, building in-house is likely a mistake that could take until 2030 to bear fruit. Licensing existing, proven technology allows these companies to go live in weeks rather than years.

However, the integration of AI does not signify the end of human labor in travel. HotelPlanner’s data shows that even with a highly competent AI, approximately 30% of customers still explicitly ask to speak to a human. This highlights a critical insight: AI is best used as a tool for "overflow" and efficiency rather than total replacement. During peak booking seasons or major global events, AI can handle the surge in volume, ensuring zero hold times for customers, while human agents focus on high-value, high-complexity cases or customers who require a personal touch. This hybrid model optimizes the bottom line by reducing the need for overstaffing while simultaneously improving the customer experience.

Looking toward 2027 and beyond, the roadmap for Reservations.ai involves moving into multimodal interactions. While voice is the fastest way for a human to communicate a complex request, it isn’t always the best way to consume information. The next iteration of the technology aims to combine voice and touch. For example, a traveler in a car might use voice to initiate a search for a hotel ("Skyler, find me a four-star hotel with a pool within five miles"), but then use a touch screen to quickly swipe through photos of the suggested rooms. This synergy between natural language and visual interaction represents the next frontier of the user interface in travel.

The travel industry has a history of falling for "hype cycles," from the early days of the dot-com boom to more recent trends like blockchain and the metaverse. Hentschel observes that the "AI hysteria" likely peaked in early 2024, and the industry is now entering a more sober period of realistic application. The companies that will thrive in this next era are those that view AI not as a magic wand, but as a sophisticated utility that requires human direction and high-quality data.

As travel companies look to the future, the success of Reservations.ai suggests that the most effective AI tools will be those that were born "in the trenches" of actual operations. By training technology on millions of real-world interactions and testing it against live, high-stakes bookings, HotelPlanner has bypassed the theoretical limitations that currently stall many AI initiatives. For the traveler, this means a future where the friction of booking—the endless searching, the long hold times, and the disconnected systems—finally begins to dissolve into a seamless, conversational experience. For the industry, it marks the transition from AI as a novelty to AI as the essential infrastructure of global commerce.