Thought Leaders
The Architecture of Future Travel: How AI Infrastructure is Replacing Manual Processes
The tourism industry is currently at a turning point globally. Flight schedules are changing by the minute, as are ticket prices. Unforeseen events are spreading across continents faster than airlines can respond to them. Travelers expect instant rebooking, transparent pricing, and seamless journeys, not overnight queues or manual workflows hidden behind a modern user interface.
Despite constant changes in its structure, the entire system still runs on a framework developed half a century ago. Global Distribution Systems (GDS) remain extraordinarily reliable at scale but their architecture was designed in the 1970s – an era before dynamic retailing, continuous pricing, API distribution, or anything resembling AI-driven operations existed.
This discrepancy is no longer a technical debt – it is a structural cost center and the primary barrier to profitability and scalability. The industry is struggling under the weight of manual servicing costs and lost revenue opportunities caused by its own infrastructure.
Over the years of working with airlines, agencies, consolidators, and travel platforms in different regions, I have noticed one clear pattern: the biggest obstacle in the travel industry is not customer support, but the infrastructure itself.
Unless the industry rebuilds the foundation on which travel transactions are based, artificial intelligence will remain a cosmetic addition rather than the operational revolution the sector so desperately needs.
This article explores why outdated systems continue to dominate, why automation is constantly delayed, and what the level of AI-based execution should look like to take the travel industry beyond human-dependent operations.
Why the Industry Still Runs on GDS Infrastructure
GDS platforms still dominate due to deep commercial entrenchment, network effects, and contractual incentives. They provide a globally consistent and contractually reliable source of bookable inventory.They are the only globally synchronized and secure source of bookable travel services. Every day, they coordinate millions of bookings with airlines, OTAs, TMCs, consolidators, and corporate systems.
However, the fundamental problem is that the core logic and architecture of GDS are still based on legacy data exchange standards. Historically, this was EDIFACT, and while modern GDS have long since adopted XML/JSON wrappers and support newer schemes like NDC, the underlying transaction logic and many business processes remain embedded in these aging structures. These standards were sufficient for a slower, less dynamic industry of the past but now impose severe limitations on flexibility, data richness, merchandising, and service capabilities.
This is not to say GDS have lost their value – they have served the industry with exceptional reliability for decades. However, they were never designed for modern requirements: dynamic offers, continuous pricing, complex bundled packages, or AI-driven service. Today’s travel ecosystem is not monolithic. It includes low-cost carriers that often bypass GDS entirely and airlines striving for direct distribution via NDC, though achieving a complete decoupling from GDS remains challenging for most.
The travel ecosystem is a dense, interdependent chain of online travel agencies (OTAs), travel management companies (TMCs), aggregators, consolidators, and mid-office systems, each of which relies on assumptions embedded in outdated standards. Therefore, even minor changes require immense levels of coordination.
Why Automation Stops at Customer Support
Discussions about artificial intelligence in travel are centered only on chatbots, self-service flows, and automated FAQs. This is useful, but mostly on the surface. The real complexity is hidden behind the scenes..
Even a simple customer request – “change my flight,” “refund my ticket,” “apply this waiver” – triggers a maze of operational steps: recalculating fares across multiple booking classes, restructuring Passenger Name Record (PNRs), validating rules, handling involuntary changes, reconciling ticketing deadlines, and navigating refund logic shaped by dozens of conditions.
Agents perform these tasks manually because systems do not provide complete and consistent data. It is not a matter of AI’s lack of capabilities, but rather the lack of infrastructure on which it could operate.
New Distribution Capability (NDC) was meant to modernize distribution and in retailing, it did. But NDC implementation is wildly inconsistent. Each airline and each GDS exposes different schemas, servicing flows, and business logic. The promised ‘standard’ of NDC has, in practice, spawned hundreds of non-standard implementations.Today, a simple exchange works differently depending on whether the booking originated through GDS, NDC, or a direct API.
As a result, automation constantly fails. It’s not because companies don’t want it, but because AI can’t automate what it can’t interpret or execute safely.
The Core Problem: Fragmented Data and Fragile Workflows
Travel transactions rely on a series of steps: availability, pricing, booking, ticketing, payment, recheck, reissue, refund, synchronization. Each step runs on a separate system, built at different times with different data models.
This fragmentation creates weakness:
- GDS, NDC, and direct API content all differ.
- PNR, ticket, order, and fare data are stored separately.
- Servicing logic varies by channel.
- Legacy schemas can’t handle modern offer-and-order complexity.
They disrupt the workflow, leading to lost revenue, compliance issues, or customer dissatisfaction. The industry’s final failsafe has become human agents. Human agents act as the “glue layer,” stitching together systems that were never designed to work together.
Why the Industry Needs a New Architectural Foundation
In 2025, the tourism sector is undergoing the most rapid changes in its history, due to airlines switching to continuous pricing and dynamic service packages, and supply and ordering models changing the principles of retail trade. Artificial intelligence has paved the way for fully autonomous operations. But established infrastructure must keep pace with change.
What the industry really needs:
- Normalized, machine-readable data across all sources
- Transaction-safe orchestration for servicing and changes
- Real-time availability and pricing updates
- Fault-tolerant execution of complex workflows
- Rules encoded as logic, not PDFs or tribal knowledge
This can’t be solved at the UI layer. It requires a foundational shift: an AI-native execution layer beneath the surface of every travel transaction.
Research in the field of cloud computing shows that distributed modular architecture significantly improves scalability and reliability – exactly what is needed for real-time automation in the tourism industry.
Outdated systems cannot meet these requirements. They need an additional layer designed specifically for artificial intelligence.
The Role of AI: From Conversations to Execution
Today, most artificial intelligence initiatives in tourism focus solely on customer communication. This is useful, but it is not revolutionary or truly a priority issue.
I believe that the future of AI in tourism should be behind operational activities. AI should be able to reissue tickets, process refunds, manage disruptions, synchronize bookings across all channels, automatically apply pricing rules, and execute multi-step workflows from start to finish. And it should do so with the same reliability and accuracy expected of experienced agents.
For AI to operate safely, a basic system is needed that guarantees data consistency, workflow traceability, transaction integrity, predictable results, and compliance with airline and regulatory requirements.
Meanwhile, the market for “servicing automation” and refund processing tech is growing – several vendors and airlines report a surge in demand for automation, but also highlight how legacy fragmentation continues to block full-scale deployment.
In short: AI must connect to infrastructure that understands travel logic, not just language. This is the missing layer.
What the Next-Generation AI-Native Travel Stack Looks Like
The established GDS infrastructure will remain a critical system of record in the near term, but its role must evolve. The industry requires a new execution layer that abstracts its complexity and transforms fragmented operations into automated workflows.
This architecture should includes:
- A unified data layer
We normalize data from GDS, NDC, and direct APIs into a format that machines can read. - A deterministic transactional orchestration engine
Executing and recovering from complex servicing flows autonomously, with built-in safety and auditability for every PNR touch. - AI agents with domain-specific expertise
AI agents are powered by codified domain expertise – where fare rules, ticketing logic, and operational procedures are translated into deterministic, auditable execution paths, not just statistical language models. - Real-time monitoring and automated recovery
Ensuring resilience across multi-step, high-risk transactions. - A security and compliance framework
It has to be trustworthy, transparent, and verifiable.
With the introduction of modern AI technologies, the entire stage that can only be performed by GDS is changing.
A Turning Point for Travel Infrastructure
The industry is currently at a pivotal juncture. Emerging technologies and modern society are changing demand, and the retail air ticket business is undergoing structural changes.
The next decade of travel will be defined not by front-end innovation, but by infrastructure innovation. The companies that embrace AI-native architecture will scale more efficiently, operate with greater reliability, and deliver the seamless experiences travelers have expected for years but rarely received. Those that delay will remain chained to systems never built for the complexity of modern travel.
The next decade of travel tech investment must shift from polishing the storefront to rebuilding the warehouse and the supply chain. The winners will be those who invest not in the most conversational AI, but in the most capable transactional AI – the intelligence that operates reliably behind the scenes.
