What it takes to make AI work at scale in travel

Across the travel industry, artificial intelligence (AI) is reaching an inflection point, with pressure mounting to demonstrate the value of implementation at scale. Gone are the days of aspiration, experimentation and discussion; now is the time to deliver on the promise of this transformative technology. 

Here I want to explore the potential—and challenges—ahead this year, looking at how a system of record, neutral execution and the industry-wide orchestration of specialized agents will be vital. 

Traveler expectations are driving change 

AI has been part of travel technology for almost two decades. What is different now is that travelers themselves are seeing the power of new tools, including ChatGPT and Claude. They realize what they can do and want to see that translated into the travel experience. 

This is particularly true during the inspiration phase of a trip, so far the part of the journey most impacted by AI. Here travelers are exploring destinations and sketching out itineraries with AI assistants before they reach a booking platform. I’ve done it and most likely so have you. 

This is increasingly where our travel ideas will be sourced, and our industry cannot afford to ignore this changing behavior. 

Neutrality at the execution layer 

But there are challenges that are often overlooked or even dismissed by those promising to unlock the power of AI. Travel is a highly regulated industry with near-zero tolerance for error. Availability, pricing, ticketing, passenger identity and airport operations all demand accuracy, security and resilience. 

There is no space for error or hallucination. 

If companies simply place an AI layer on top of existing infrastructure, the system may work initially, but struggle at scale. That’s why many travel companies are rethinking their underlying applications so AI can function effectively across the entire workflow.

In an industry like travel, a core platform must act as the system of record, holding the authoritative booking, ticketing and servicing state. AI must be connected to this foundation to be reliable at scale. Amadeus is the trusted system of record in travel, and our focus is on embedding AI directly into core platforms that manage booking, pricing and servicing, where reliability and scale matter most. 

Much of the infrastructure travel relies upon was built over decades and was not designed to interact with AI. Many existing application programming interfaces (APIs) and data structures are thus difficult for large language models to interpret. As a result, companies will need to adapt their applications to make them more AI-friendly, while also meeting strict security, compliance and operational requirements. 

Orchestration is vital to AI in travel 

At an industry level, AI becomes effective not through a single model but by orchestrating multiple capabilities across highly interconnected systems, workflows and partners, while maintaining control over decisions and outcomes. This means coordinating the many players in a journey, including airlines, hotels, airports, transport providers and activities.

Technically, the industry is moving toward specialized AI agents that handle defined tasks such as content search, booking management or service recovery. These smaller agents tend to perform more reliably with fewer hallucinations. 

They must then be coordinated to complete end-to-end tasks, such as booking a trip across multiple systems, with orchestration enabling communication, sequencing and accountability. This is a role Amadeus can undertake as AI will reinforce and augment our platform.

Commercial imperative 

Travel companies feel urgency around AI, with a broad recognition it could reshape how travelers discover, plan and book trips. Many are seeking guidance on how to participate. Across the industry, we have talked about seamless travel journeys for years. AI is now the thing that could actually make it happen.

Amadeus helps bridge the gap, working with major technology providers such as Microsoft and Google that are building the AI assistants likely to influence how travelers access services. We are also investing directly, including the acquisition of SkyLink, an AI-first company with a proprietary architecture and multilayer orchestration engine enabling conversational booking and servicing in seconds.

Building AI prototypes is now relatively straightforward. Operating them at scale is harder. Successful companies invest in monitoring, testing and explainability and measure impact in business terms such as conversion, service resolution time and cost-to-serve. Because AI systems are non-deterministic, testing must evaluate ranges of acceptable outputs, supported by strong observability in production.

Using AI appropriately is equally important. AI excels at understanding intent, while traditional algorithms handle precise calculations and optimization. Strong solutions combine both, with AI capturing intent and orchestration ensuring accurate pricing, availability, ticketing and servicing.

On the horizon 

We expect AI will follow a path similar to cloud computing, evolving from transformation to standard infrastructure. It will become deeply integrated and embedded within platforms that combine system-oriented and event-driven architectures with AI components and agents. 

Rather than replacing existing systems, AI will complement them, creating a hybrid architecture where traditional platforms and AI work together to power travel experiences.